personalized treatment assignment

• Study Protocol
• Open access
• Published: 09 December 2022

The PerPAIN trial: a pilot randomized controlled trial of personalized treatment allocation for chronic musculoskeletal pain—a protocol

• E. Beiner 1 ,
• D. Baumeister 1 ,
• D. Buhai 1 ,
• M. Löffler 2 , 3 , 4 ,
• A. Löffler 2 ,
• A. Schick 5 ,
• L. Ader 5 ,
• W. Eich 1 ,
• A. Sirazitdinov 6 ,
• C. Malone 2 ,
• M. Hopp 7 ,
• C. Ruckes 7 ,
• J. Hesser 6 ,
• U. Reininghaus 5 ,
• H. Flor 2 ,
• J. Tesarz 1 &

PerPAIN consortium

Pilot and Feasibility Studies volume  8 , Article number:  251 ( 2022 ) Cite this article

1876 Accesses

3 Citations

1 Altmetric

Metrics details

The therapy of chronic musculoskeletal pain (CMSP) is complex and the treatment results are often insufficient despite numerous therapeutic options. While individual patients respond very well to specific interventions, other patients show no improvement. Personalized treatment assignment offers a promising approach to improve response rates; however, there are no validated cross-disease allocation algorithms available for the treatment of chronic pain in validated personalized pain interventions. This trial aims to test the feasibility and safety of a personalized pain psychotherapy allocation with three different treatment modules and estimate initial signals of efficacy and utility of such an approach compared to non-personalized allocation.

This is a randomized, controlled assessor-blinded pilot trial with a multifactorial parallel arm design. CMSP patients ( n = 105) will be randomly assigned 1:1 to personalized or non-personalized treatment based on a cluster assignment of the West Haven-Yale Multidimensional Pain Inventory (MPI). In the personalized assignment condition, patients with high levels of distress receive an emotional distress-tailored intervention, patients with pain-related interference receive an exposure/extinction-tailored treatment intervention and patients who adapt relatively well to the pain receive a low-level smartphone-based activity diary intervention. In the control arm, patients receive one of the two non-matching interventions. Effect sizes will be calculated for change in core pain outcome domains (pain intensity, physical and emotional functioning, stress experience, participant ratings of improvement and satisfaction) after intervention and at follow-up. Feasibility and safety outcomes will assess rates of recruitment, retention, adherence and adverse events. Additional data on neurobiological and psychological characteristics of the patients are collected to improve treatment allocation in future studies.

Although the call for personalized treatment approaches is widely discussed, randomized controlled trials are lacking. As the personalization of treatment approaches is challenging, both allocation and intervention need to be dynamically coordinated. This study will test the feasibility and safety of a novel study design in order to provide a methodological framework for future multicentre RCTs for personalized pain psychotherapy.

Trial registration

German Clinical Trials Register, DRKS00022792 ( https://www.drks.de ). Prospectively registered on 04/06/2021.

Peer Review reports

Introduction

As the leading symptom of a wide variety of musculoskeletal disorders, chronic pain is one of the most serious health problems worldwide [ 1 ] with enormous consequences on psychological as well as physical health [ 2 , 3 ]. It is estimated that more than 20% of the world’s adult population suffer from chronic pain and every 10th person is diagnosed with chronic pain every year [ 4 , 5 ]. While initial causes of pain can often be adequately treated (e.g. surgery, joint replacement, anti-inflammatory drugs/biologicals), in many cases, the pain persists and becomes chronic. This is especially true for chronic musculoskeletal pain (CMSP) conditions where a specific cause for the pain can no longer be detected [ 3 , 5 ]. These complex persistent CMSP conditions are manifested by a kaleidoscope of symptoms that are temporally dynamic. Various therapies are available, and a multimodal treatment principle is generally recommended [ 5 ]. However, the treatment of CMSP is usually difficult and the treatment outcomes are unsatisfactory [ 4 ]. Various randomized controlled trials (RCTs) for the treatment of CMSP have led to negative results despite promising outcomes from preclinical and early clinical studies [ 4 , 6 , 7 ]. Accordingly, the most recent Cochrane review found that multimodal pain treatment achieves only small effects on pain and disability immediately after treatment, which had disappeared by follow-up [ 4 ].

There is variability between patients in their response to different pain therapies (even in effective treatments), which can significantly limit overall effects in clinical trials. Often psychobiological mechanisms of pain aetiology and maintenance are not considered [ 6 , 8 ]. This led to calls for personalized pain therapy with a focus on specific disease-eliciting and disease-maintaining mechanisms and the development of empirically based algorithms that determine the optimal treatment or treatment combinations for individual patients to improve both the clinical care of patients with pain and the success rates of established pain management procedures [ 6 , 9 ]. A major problem may be related to the fact that patients with comorbid mental disorders such as anxiety and depression and a high burden of psychobiological factors may be even more in need of specialized mechanism-based treatments tailored to their specific multiple needs [ 6 , 10 , 11 ]. Across individual diagnostic categories (e.g. fibromyalgia (FM), osteoarthritis (OA) or chronic non-specific back pain (CBP)), chronic pain may be maintained by similar psychobiological, including comorbid mechanisms leading to the need of cross-diagnostic treatment approaches taking into account relevant core mechanisms [ 6 ]. There are several core biological and psychological mechanisms that interact with CMSP in complex ways and can determine the development and maintenance and spread of comorbid psychopathology [ 12 ]. Variability of these biological and psychological mechanisms in different clinical presentations of chronic pain appears greater between individual patients than between different underlying diseases, suggesting that mechanistic etiologies of the pain chronicity process and subsequent successful treatment are likely to be at the individual level rather than at the disease level [ 13 ].

To implement a personalized treatment approach, however, the characteristics of individual patients or subgroups of patients that show common disease mechanism maintaining their pain, which increase or decrease the response to a particular treatment, need to be identified. Based on previous work, including our own, we can distinguish three subgroups of musculoskeletal pain patients: (1) patients who are characterized by extensive dysfunctional behaviours with high levels of fear of pain, proneness to the rewarding and punishing consequences of pain, (2) patients characterized by high levels of (interpersonal) distress and comorbid depression/anxiety and (3) a group that is not characterized by a specific psychobiological characteristics and comorbidity, commonly termed “adaptive copers” [ 14 ]. At this point in time, these mechanistic differentiations are solely based on cluster analyses of questionnaire data; however, we have identified pathomechanisms that may be specific for these subgroups [ 15 , 16 ]. These include functional brain imaging results [ 15 , 16 ], as well as evidence for specific mechanisms related to maladaptive aversive learning, memory processes, and reduced capacity for pain-memory extinction including fear of pain and anxiety [ 12 , 17 ]. Similarly, we previously found that high levels of stress and emotional distress, exposure to psychological trauma and high levels of depression were associated with hyperalgesia to deep and widespread pain and that emotional abuse leads to exacerbated spinal pain summation in chronic pain patients [ 18 ].

Further, earlier evidence suggests that this clustering is associated with divergent treatment responses. In a feasibility study [ 19 ], we treated a group of CMSP patients with high level of emotional distress with an Emotional Distress Desensitization-tailored treatment approach (EDDT) using eye movement desensitization reprocessing (EMDR), focusing on alterations of distressing experiences [ 20 ]. We adapted the trauma EMDR treatment manual to the specific needs of CMSP patients with high level of emotional distress [ 21 ]. This emotional distress-tailored treatment induced a significant and clinically meaningful reduction of pain intensity and disability [ 19 , 22 ]. Significantly, the therapy also led to a normalization of the aforementioned changes in the somatosensory function.

Based on findings on the role of maladaptive aversive learning, memory processes and reduced capacity for pain-memory extinction including fear of pain and anxiety involved in chronic pain, an operant treatment approach based on an increase in healthy behaviours and a decrease in pain behaviours was recently expanded to brain-based extinction retraining. This Pain Extinction and Retraining-tailored treatment approach (PERT) yielded excellent results in CMSP patients in pilot studies and increased effect sizes when specifically provided to patients characterized by high dysfunctionality, fear of pain and anxiety, whereas those with high levels of distress profited less [ 23 ].

In addition, we also developed and tested a new pain diary that focuses on pain-free phases and activity rather than pain and disability [ 17 ]. With this positive pain diary, we aim to investigate momentary mechanisms, prognostic markers and treatment outcomes in daily life using smartphone- and sensor-based ecological momentary assessment (EMA) [ 24 ]. We found significantly lower pain and stress and higher mood ratings at the end of a 4-week trial period when the positive activity diary was implemented [ 17 ]. Quality of life was significantly improved as well. Ecological momentary assessments provide unique opportunity for real-world, real-time, interactive, adaptive and personalized administration of interventions based on the dynamics of individuals’ experience and behaviour and their interaction with contextual factors in daily life [ 25 , 26 ]. Therefore, implementing this new pain diary as an ecological momentary low-level diary intervention for real-time and real-world activity-based attention modulation (EMDI) may be a useful intervention for patients with pain but low levels of comorbidity. This EMDI therefore complements the other two approaches, which are more likely to address patients with high levels of emotional stress or dysfunctional behavioural patterns.

Although all three therapeutic approaches address different core mechanisms and patient subgroups, to our knowledge, there are no approaches to date that have investigated the improvement of efficacy through such precise therapy allocation. Furthermore, there is a lack of validated and cross-disease allocation algorithms that would allow a personalized allocation to the different pain interventions. Against this background, this trial will test the feasibility and safety of a personalized pain psychotherapy allocation with three different treatment modules and explore the initial signals of efficacy and utility of such an approach compared to non-personalized allocation.

The primary objective of this study is to investigate the feasibility and safety of a personalized pain psychotherapy approach with three different treatment modules—EDDT, PERT and EMDI. A secondary objective is the initial estimation of efficacy and utility of such a personalized approach compared to a non-personalized approach on different core pain outcome domains.

The overarching aim of our personalized pain psychotherapy allocation is to allocate an individual to a therapeutic intervention that fits the best based on the individual characteristics of the patient. In our study, we will focus on the allocation of patients to three well-established and well-validated pain patient clusters whose categorization is based on the widely used West Haven-Yale Multidimensional Pain Inventory (MPI) [ 27 ]. Based on this questionnaire, patients will be assigned to one of the three clusters [ 14 ]. According to their cluster’s assignment, patients will be allocated “personalized” to the best fitting psychological treatment. In the control arm of this study, patients will be allocated to a treatment that does not match their cluster assignment.

This study is part of the consortium ‘PerPAIN - Improving outcomes in chronic musculoskeletal pain through a personalized medicine approach’ funded by the German Federal Ministry of Research and Education (01EC1904A). This report focuses on the implementation of the pilot study subproject ‘Personalized treatment for chronic musculoskeletal pain: a randomized double-blind controlled trial with multifactorial parallel arm design’. The paper presents the study protocol for the trial, adhering to the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) Statement [ 28 ], while the results of the trial will be reported in line with the CONSORT 2010 Statement: extension for randomized pilot and feasibility trials [ 29 ]. An additional file shows this in more detail (see Additional file 1 ). All participants must provide written informed consent before inclusion in the study. The study has been approved by the Ethics Research Committee II of the Faculty of Medicine, University of Heidelberg (2020-579N) and will be carried out in compliance with the Helsinki Declaration.

Musculoskeletal pain patients will be offered participation in this study until at least 342 patients have been enrolled in baseline assessment and 105 patients have been included into our randomized controlled proof-of-concept study. All 105 patients enrolled in the study will undergo detailed phenotyping before and after intervention, which, together with the response/non-response data from the RCT, will serve to improve the allocation algorithm for future studies. All patients eligible will be randomly assigned 1:1 to a personalized or non-personalized treatment arm (Fig. 1 ). Personalization is achieved by a targeted allocation either to a therapy focused on emotional distress, a therapy focused on dysfunctional behaviours and maladaptive learning or allocation to a minimal intervention group where adaptive responses to pain will be registered. After baseline assessments (Fig. 1 , T 0 ) for relevant pathomechanisms, each patient included in the trial will receive either 12 weekly sessions of cluster-matching therapy (intervention arms)—where patients are allocated personalized (i.e. according to their individual pain mechanism)—or non-cluster-matching therapy (control arms) —where they are allocated non-personalized (i.e. randomly assigned between the other two interventions). After the last session, all procedures included in the baseline assessment will be performed as a posttreatment evaluation (Fig. 1 , T 1 ). Furthermore, a 3-month follow-up by means of telephone interview and questionnaire package will be implemented for patients taking part in the intervention study (Fig. 1 , T 2 ). At T 2 , the maintenance of the effects on pain and pain-related impairments will be assessed. In order to guarantee a correct evaluation of our primary outcomes in pain change, we will include an extended psychological assessment (details on outcomes are given below).

Intervention scheme/trial flow. Note: the same treatment interventions are applied in personalized as well as in non-personalized interventional study arms. However, the arms differ in the assignment to the respective treatment modules: While in the personalized assignment arms patients with high level of emotional distress receive an emotional distress-tailored intervention (EDDT), patients with high avoidance/maladaptive-learning behaviour receive an extinction/retraining-tailored treatment intervention (PERT) and patients with neutral risk profile receive a low-level diary intervention (EMDI), patients in the non-personalized arm receive a non-personalized intervention (i.e. patients with a high level of emotional distress receive either extinction/retraining-tailored or a low-level-tailored intervention, but not the emotional distress-tailored treatment). CMSP, chronic musculoskeletal pain; PERT, Pain Extinction and Retraining-Tailored Intervention; EDDT, Emotional distress desensitization and reprocessing-tailored intervention; EMDI, Ecological Momentary-tailored Low-Level Diary Intervention

The inclusion and exclusion criteria are chosen to ensure good representativeness and to cover the heterogeneity of different musculoskeletal pain conditions and diagnoses.

Key inclusion criteria : CMSP >3 months as symptom of (1) non-specific chronic back pain, (2) osteoarthritis, (3) fibromyalgia syndrome or (4) rheumatoid arthritis. Symptoms must be present for at least 3 months to ensure clinically relevant chronicity; the ability to see and use a mobile telephone (incl. with visual aids), age ≥18 years (no upper age limit) and ability to provide informed consent.

The key exclusion criteria will contain insufficient or unclear treatment of the underlying disease (according to the practitioner’s judgement), application for retirement/pension pending, on-going psychotherapy, severe, pharmaceutically treated acute life-threatening physical comorbidity or physical comorbidity which is incompatible with participation in the study according to the practitioner’s judgement, severe mental disorder (inability to consent, suicidality, psychosis spectrum disorders), neurological comorbidity (epilepsy, traumatic brain injury, seizures, multiple sclerosis, neurodegenerative diseases) and pregnancy. Participants will be excluded from the MRI assessment, if they have implants or metal parts in their body, which are not compatible with 3T MRI, or diseases which prevent lying still for 60 min. Clinical evaluation will be carried out by study physicians with experience in the diagnosis and management of chronic pain conditions. All patients will undergo a clinical evaluation by the Heidelberg Study Center for Clinical Pain Research within the Department of General Internal Medicine and Psychosomatics of Heidelberg University Hospital [ 30 ].

Participants of the CMSP group will be recruited through our tertiary pain clinic at the University Hospital Heidelberg, the affiliated academic teaching hospital of the University of Heidelberg in Baden-Baden as well as the Central Institute of Mental Health in Mannheim. Should the recruitment rate prove to be insufficient, recruitment will be extended to cooperating general and specialist medical practices in the Rhein-Neckar area. All prospective participants will be screened before inclusion in the study. Screening data will be confirmed by a study physician specifically trained in the assessment of chronic pain conditions. CMSP patients will be entered into the trial until the necessary sample size is reached.

During the planning phase of the study, we involved patient representatives of patients with CMSP. Further, members of relevant patient groups were invited to an initial focus group to ensure patient-friendly study procedures and to reduce the patient burden at all stages. In addition, all participants will be asked systematically about their experiences and impressions during the study and results will be discussed together with patient representatives at regular meetings. For the communication of the results to patients, we will organize a patient forum after the completion of the study.

• Treatment personalization

With our pilot study, we aim to prepare a multi-centre randomized controlled trial (planned RCT) to assess the feasibility and safety of a mechanism-based personalized assessment and treatment approach for CMSP patients across disorders. To further improve the personalized allocation algorithm for the planned RCT, a comprehensive characterization of the study participants will be performed with dynamic quantitative-sensory testing paradigms, immunological and neuroendocrine assessments, psychological questionnaire assessments, longitudinal data collection in everyday life (EMA, ECG and activity sensors) and neuroimaging (fMRI) pre- and post-treatment.

As this pilot study compares a personalized treatment allocation based on patient characteristics with a non-personalized treatment allocation, the same treatment interventions are applied in both study arms. However, arms differ in the assignment to the respective treatment modules. In the personalized assignment arm, patients with high level of emotional distress receive an emotional distress-tailored intervention (EDDT: Emotional Distress Desensitization-tailored treatment), patients with high avoidance/maladaptive-learning behaviour receive an exposure/extinction-tailored treatment intervention (PERT: Pain Extinction and Retraining-Tailored Intervention) and patients with neutral risk profile receive a low-level smartphone-based intervention (EMDI: Ecological Momentary-tailored Low-Level Diary Intervention). For every intervention, a special standard operation procedure (SOP) protocol is written regarding the 12-week design and special psychological/pain characteristics of the patient groups. Patients in the non-personalized arm receive a non-personalized intervention allocation (e.g. patients with a high level of emotional stress receive either exposure/extinction-tailored or a low-level smartphone-based intervention, but not the emotional stress-tailored treatment). For this study, matching to the subtypes will be based on a cluster analysis of the West Haven-Yale Multidimensional Pain Inventory (MPI) as well as the age and gender of participants, which have been identified as significant predictors of MPI clusters in previous research. Cluster status of individual participants will be identified by using soft k-means clustering of previous data collected from CMSP patients by the study teams. Using this data, clusters were identified, and the most likely cluster status for participants in the current study will be determined by an automated script implemented in an electronic data collection tool (REDCap) using a random-forest decision algorithm using the established clustering algorithm. Based on this algorithm, participants in the intervention arm will be assigned to the intervention that best fits according to their cluster, while patients in the control arm will be randomly assigned to one of the interventions that do not fit their cluster.

Interventions

The individual interventions consist of 12 weekly face-to-face sessions (EDDT, PERT), or a minimal intervention condition in which smartphone-based diary queries are monitored daily over a period of 12 weeks (EMDI). The efficacy and safety for all three treatment modules have been demonstrated in previous studies, yet their superiority in outcomes depending on patient cluster has not been formally addressed. During the trial, patients are encouraged not to change their existing therapies over the course of the therapy. All additional treatments will be recorded regularly. In order to ensure that therapy protocols are appropriate and that therapists learn to follow these protocols in early stages of the study, a subsample of pilot participants will be invited to take part in the interventions without being part of evaluation of the personalized treatment allocation algorithm to aid the quality assurance of interventions.

Emotional distress desensitization and reprocessing (EDDT) intervention

EDDT is a stress-reducing intervention that combines the use of well-established trauma intervention elements (including imaginal exposure and cognitive and self-control techniques) and the use of specific EMDR elements such as bilateral sensory stimulation (e.g. left–right eye movements or bilateral hand-tapping induced by the therapist’s fingers) and the dual focus of attention principle. With the dual focus of attention principle, patients simultaneously focus on distressing memories and an external bilateral sensory stimulus. This procedure is suggested to facilitate information processing of emotionally distressing memories (e.g. traumatic events or pain sensations) and thereby cause a decreasing or even an elimination of the emotional distress related to these memories. Recent studies have shown that the dual application of bilateral sensory stimulation is highly efficient in triggering the inhibitory effects of the thalamus on the amygdala [ 31 ]. This strategy is increasingly being used to treat patients with chronic pain and its efficacy in pain treatments has been shown in several randomized controlled trials [ 22 , 32 ]. Promising results have been reported especially for chronic musculoskeletal disorders, including back pain [ 33 ] and inflammatory joint pain [ 34 ]. The treatment protocol for this study is based on a standardized manual [ 21 ] and the possible targets for processing will encompass disturbing memories, current pain perceptions and pain-related fears and cognitions, and anticipated future painful situations together with the associated cognitions, emotions and bodily sensations. Participants who are allocated to the intervention group will receive a manualized and 12-session outpatient psychotherapeutic EDDT intervention (weekly session for 100min).

Pain Extinction and Retraining-tailored (PERT) intervention

PERT is a structured cognitive-behavioural derived intervention that focuses on retraining and alterations of maladaptive brain responses and includes elements such as activity increase, reduction of compensatory behaviours, medication management, altering social interactions, training of pleasant events and improving sensory discrimination. The treatment is based on an extension of a standardized manual [ 35 ] and consists of 12 weekly 100-min sessions led by two psychologists and is conducted in groups of 5–6 patients. To ensure a good patient-therapist relationship, all patients will have two individual sessions with the therapist prior to the group therapy. The group therapy includes video feedback of expressions of pain as well as training of pain-incompatible behaviours, increase of work-related and social activities and medication management. Patients are encouraged to participate in role plays to reduce dysfunctional pain behaviour and promote healthy behavioural responses. To increase the transfer of the treatment effects from the clinic to everyday life of the patients, the training also includes several sessions together with their spouses. Spouses will be fully informed and consented just as study participants. In previous studies, we could show that PERT interventions are superior to multimodal pain interventions and that maladaptive brain activation patterns can be altered by such interventions [ 15 , 36 , 37 , 38 , 39 ]. In a study that is currently under review (Thieme, Kleinböhl et al., submitted), we analysed post hoc to what extent a matching of PERT to patients who are highly dysfunctional compared to those who are characterized by traumatic events/high stress/depressive responses or patients who have no specific dysfunctional or stress-related characteristics boost the treatment effect. Patients who happened to receive the fitting personalized treatment showed an almost 2-fold increase in effect size and a 50% reduction in drop-out. In this previous study, we did a post hoc analysis—the aim is now to a priori assign patients based on an optimized treatment algorithm.

Ecological momentary low-level diary intervention for activity-based attention modulation (EMDI)

We will use a positive activity diary as a low-level intervention with the goal of diverting the focus of attention away from pain and negative body and emotional experiences. This smartphone-based ecological momentary intervention app will not assess pain, negative mood or stress but the absence of pain, positive mood and positive events as targets for refocusing on positive events rather than pain. This will include enhancing, consolidating and interactive ecological momentary intervention components. In previous studies, we developed and tested a new pain dairy that focuses on pain-free phases and activity rather than pain and disability. We found significantly lower pain and stress and higher mood ratings at the end of a 4-week trial period when the positive activity diary was implemented (Nees et al., submitted). We therefore suggest that implementing this diary as an EMI may be a useful and safe low-level intervention for patients with pain who do not show any significant comorbidity or other psychobiological pathogenetic mechanisms.

Randomization and allocation

For the allocation of the participants, a computer-generated list of random numbers will be used. Randomization for the pilot study will be carried out by the Interdisciplinary Center for Clinical Trials (IZKS) Mainz using permuted blocks of variable lengths. The randomization lists will be implemented in the electronic case report form system (eCRF) such that randomization can be performed electronically as new participants are added to the study. After providing informed consent and the acquisition of baseline data, participants will be randomly assigned to experimental intervention (personalized allocation) or control intervention (non-personalized allocation) via this eCRF system. Allocation will be managed by an independent study manager involved neither in sequence generation, assessments or treatment. All processes will be monitored and controlled for correctness by the IZKS Mainz. This process will be started only after the enrolled participants will have completed all baseline assessments and it is time to allocate the intervention. Randomization will be stratified by cluster, so that patients have a 50% chance of being allocated to the treatment matching their cluster, and 50% of being allocated to one of the two interventions (25% chance each) not matching their cluster.

Patients, therapists and the researchers carrying out participant assessments will be blinded to the respective condition (personalized vs. non-personalized), and success of blinding will be evaluated to explore possible source of bias. The double-blinding of patients and data collectors (i.e. patients, therapists and psychologists) is kept during the course of the study and statistical analysis will be also done with blinding maintained to avoid bias. Randomization authorities from the Interdisciplinary Center for Clinical Trials (IZKS) Mainz will be instructed to report any suspected breach of the blinding procedures.

Sample size calculation

The study will be performed in an exploratory fashion and is planned as a pilot study to describe the feasibility and safety of personalized treatments for CMSP as a primary outcome, as well as an initial estimation of signals efficacy and utility as secondary outcomes. Even though a formal power calculation is not necessary in a pilot study, we nevertheless performed a power calculation to better estimate the initial signals of efficacy and utility.

For power considerations, we use a mixed-model with repeated measurements (MMRM), a two-sided significance level of 5% and a sample size of 90 patients (→45 per study arm →90 total from a total of 105 patients at an assumed dropout rate of 15%). The MMRM takes into account also the interim measurements for each patient and allows therefore a fuller use of the data than an ANCOVA. The sample size of 105 patients allows to calculate our feasibility measures with a 95% confidence interval width from 11 to 19% (for varying proportion 0.1–0.5) based on a Wald interval for binary measures [ 40 ]. Moreover, patients with data with at least one measurement can be used for the primary analysis; therefore, we expect fewer patients not to be usable for the analysis compared to an ANCOVA. Between repeated measurements, a correlation of 0.3 is assumed. Then, the study has a power of 80% to detect a true effect size of f = 0.22, a power of 85% for f = 0.23 and 90% power for a true effect size of f = 0.25. Dropouts will be analysed according to the underlying reasons for dropout and distinction will be made between those dropouts that are preventable by modification of study design and those that are not preventable by study design adaptations.

The primary objective of this study is to investigate the feasibility and safety of a personalized pain psychotherapy approach with three different treatment modules. Based on the success of recruitment, randomization, assessment of outcomes, treatment adherence, treatment satisfaction, compliance and acceptability, clinical feasibility will be judged as ‘feasibility given’, ‘readjustment’ for the main trial necessary or ‘feasibility not given’.

Feasibility and safety outcomes will be documented and reported comparatively

Recruitment : number of participants recruited over the study period

Successful recruitment of at least 105 participants (feasibility given)

Recruitment of 105 participants after modification of recruitment strategy (readjustment necessary)

Less than 105 participants after modification of recruitment strategy (feasibility not given)

Assessment of inclusion criteria : proportion of potential participants assessed after written consent obtained

95% of potential participants completed after giving written consent (feasibility given)

75% of potential participants completed after giving written consent (readjustment necessary)

Less than 75% of participants completed after giving written consent (feasibility not given)

Randomization : number of participants successfully randomized after completion of eligibility screening and baseline assessments

Successful randomization of at least 105 participants after completion of eligibility screening and baseline assessments (feasibility given)

Randomization of 90 participants (15% dropout rate) (readjustment necessary dependent of retention rate)

Randomization of less than 90 participants (feasibility not given)

Retention and adherence

85% retention rate of 105 randomized participants, for end-of-treatment assessment at least at one of the end-of-treatment assessments (T 1 /T 2 ) (feasibility given)

Retention of less than 85% of randomized participants, but reasons for dropout can be prevented by modification of study design (readjustment necessary)

Retention rate of less than 85% of those dropouts that are preventable by modification of study design (feasibility not given)

Treatment satisfaction, compliance and acceptability

Negative treatment effects: Unwanted and negative effects of the treatment will be assessed via the Negative Effects Questionnaire (NEQ) [ 41 ] supplemented by the subscales for negative effects on working place, partnership and family/peers of the Inventory for Assessing Negative Effects of Psychotherapy (INEP) [ 42 ].

Severe adverse events (SAEs) will be documented and reported descriptively.

Dropouts will be analysed according to the underlying reason for dropout and distinction will be made between those dropouts that are preventable by modification of study design and those that are not preventable by study design adjustment.

A secondary objective is the initial estimation of the efficacy and utility of such a personalized approach compared to a non-personalized approach on the different outcomes of the will outcome domains (pain intensity and physical functioning, emotional functioning and stress experience, participant ratings of improvement and satisfaction). Therefore, the exploratory analysis of the data will select the following candidate endpoints:

Pain severity: Pain severity will be assessed via the pain severity subscale of the Multidimensional Pain Inventory (MPI-D) [ 27 ]. This subscale summarizes the items pain now, pain in the last week and suffering related to pain. The MPI-D is a valid instrument with a high reliability (Cronbach’s α ≥ 0.90).

Disability/quality of life will be assessed via the Short-Form-Health Survey 12 (SF-12) and the Oswestry Disability Index (ODI). The SF-12 measures the impact of physical and mental health status on everyday life during the last 4 weeks. With the SF-12 the mental as well as the physical component can be assessed. It has a high validity and a good reliability for both, the mental and the physical component, with Cronbach’s α ≥ 0.75 and α ≥ 0.82, respectively [ 43 ]. The ODI captures the level of disability at the moment caused by pain in various activities of daily living (e.g. lifting weights, ability to care for oneself, ability to walk, ability to sit, sexual function, ability to stand, social life, quality of sleep and ability to travel). It is a validated questionnaire with good internal consistency [ 44 ].

Anxiety and depression: the level of anxiety and depression of the last week will be assessed via the Hospital Anxiety and Depression Scale (HADS). With 14 items, the HADS is a self-assessment scale that measures anxiety and depression through two subscales. Seven items for each subscale are rated by the patients on a 4-stage response format. It was in particular developed for somatic disorders and physical symptoms were therefore excluded. The HADS has high validity and reliability [ 45 ].

Positive and negative affect: Affectivity will be assessed via the Positive and Negative Affect Schedule (PANAS). This scale captures both positive and negative affectivity by asking for an assessment of positive and negative states of mind during the last week on a 5-point scale ranging from “not at all” to “extremely”, assessed on 20 items. It has a good validity and reliability with Cronbach’s α ≥ 0.86 [ 46 , 47 ].

Global impression of change: Global impression of change will be assessed from both the patient’s (patient’s global impression of change, PGIC) and the therapist’s perspective (therapist’s global impression of change, TGIC). It is a 7-point scale with answers coded from “very much improved” to “very much worse”.

Days out of work (socioeconomic aspect) and change in medication since start of treatment will be assessed via additional items.

Activity levels will be assessed based on EMA sampling, with 10 queries per day, assessing for example pain intensity, attention to pain, pain catastrophizing, fear of pain and positive/negative affect.

For endpoints 1–6, the change between T 0 and T 1 (or 3-month follow-up) will be used as the outcome variable. In addition, the candidate endpoint domains will be investigated after 4 and 8 weeks to explore temporal therapy effects and dose-response relationships.

Statistical analyses

Descriptive statistics (absolute and relative frequencies for variables with nominal and measures of position (mean, median) and variability measures (standard deviation, interquartile range and range) for variables with interval or ratio scaling) will be used to compare participant characteristics between the study arms.

Feasibility objectives will be quantified. If appropriate, 95% confidence intervals will be calculated based on the Wilson Score interval for binary measures. We favoured the Wilson Score interval against Wald interval due to their good properties for small numbers of trials, compared to Wei and Hutson [ 40 ]. For the analysis of the primary outcome, we will use the complete and pseudonymized data set and follow the intention-to-treat approach which includes all patients in the group they were allocated to by randomization. This approach preserves the allocation of treatment by randomization and it will be as close as possible to the ITT ideal of including all randomized patients. As sensitivity analysis, the analysis will be repeated in the per-protocol (PP) set, excluding patients with major protocol violations. Additional sensitivity analyses will explore the effects of the treatments the patients actually received. The null-hypothesis H0 (no difference occurs in the population means with regard to pain severity at T1 between the experimental intervention and the control intervention) will be tested against H1 (there is a difference in the population means of pain severity differences between the two groups at T1 and T0). Provided that the model assumptions are fulfilled, the null hypothesis will be tested using a (robust) mixed linear regression model at a significance level of 5%. The primary outcome (pain severity at T1, difference to T0) will be analysed using mixed-effects model repeated measurements after 4, 8 and 12 weeks (=T1). The analysis will be adjusted for condition, gender and baseline (T0) pain severity. Compound symmetry will be assumed. Because of the exploratory nature of the study, the observed initial signals of efficacy and utility will be considered more important than the p -values. The corresponding p -values of these tests will be interpreted purely descriptively. We will compute effect sizes and interpret them together with the respective 95% confidence intervals. Regarding secondary endpoints, exploratory data analysis will be performed using appropriate analytical methods depending on the respective parameter. Details of the statistical analyses will be fixed in a statistical analysis plan developed by the statistics department of IZKS Mainz prior to database closure. The safety analysis includes calculation and comparison of frequencies and rates of adverse events. Furthermore, statistical methods are used to assess the quality of data and the homogeneity of intervention groups. All analyses will consider gender aspects. Prior to all analyses, we will pre-specify a statistical analysis plan.

To more closely assess the ecological validity of our design in a naturalistic setting, a second analysis will be carried out where 1/3 of matched patients are randomly sampled and statistically reallocated to the non-matching arms. This is to allow a comparison between personalized treatment and truly random assignment (rather than matched vs non-matched), which may more closely mirror the ultimate implementation of personalization methods in clinical services or during a confirmatory multi-centre RCT.

Feasibility

Feasibility will be assessed according to the following criteria: recruitment, assessment of inclusion criteria, randomization, retention and adherence, treatment satisfaction, compliance and acceptability. On the basis of these criteria, we will assess whether (1) feasibility is completely given, (2) feasibility is limited but likely to be achieved with adjustment of the study procedure or (3) feasibility is not given.

We will continuously monitor the trial for any operational issues (i.e. failure in appointment management, no-show of patients). Concerning data collection, we will prioritize short questionnaires to reduce participant burden. To encourage retention at each study timepoint, non-responders will receive up to five reminders in total via phone, mail and e-mail. Outcome assessments may be completed in multiple sittings.

Missing data

Applying the participant retention strategies outlined above, we will try to minimize the missing outcome data. Notwithstanding, we will record reasons participants are lost to follow-up. Prior to multiple imputation of missing values for primary and secondary outcomes at the item level, we will conduct sensitivity analyses to assess the robustness of the missing data assumption.

Dissemination policy

Regardless of the magnitude or direction of effect, the results of this trial will be presented at relevant national and international conferences and as published articles in peer-reviewed journals. Publication of the study results will be based on the CONSORT-SPI 2018 statement for social and psychological interventions and the CONSORT extension for adverse effects. To reach health care policy and practice audiences (e.g. government bodies) concerning the scale-up of the model, we will present the findings at policy-maker- and service-provider-run conferences. Aiming at directly informing the work of policy-makers and practitioners, we will report the findings in plain language formats to them and compile an executive summary drawing together with key findings of all aspects of the intervention with a series of journal articles included as appendices.

Trial status

At the time of submission, patient recruitment to the trial has commenced. The anticipated study completion date is November 2023 . This trial was prospectively registered on the German Clinical Trials Register (DRKS) with study ID DRKS00022792 on April 6, 2021.

Availability of data and materials

Not applicable.

Abbreviations

Chronic non-specific back pain

Chronic musculoskeletal pain

Emotional Distress Desensitization-tailored treatment approach

Ecological momentary assessment

Ecological momentary low-level diary intervention

Eye movement desensitization and reprocessing

Fibromyalgia

Hospital Anxiety and Depression Scale

Inventory for Assessing Negative Effects of Psychotherapy

Mixed-model with repeated measurements

West Haven-Yale Multidimensional Pain Inventory

Negative Effects Questionnaire

Osteoarthritis

Oswestry Disability Index

Positive and Negative Affect Schedule

Pain Extinction and Retraining-tailored treatment approach

Patient’s global impression of change

Randomized controlled trials

Severe adverse events

Short-Form-Health Survey 12

Buchbinder R, van Tulder M, Öberg B, Costa LM, Woolf A, Schoene M, et al. Low back pain: a call for action. Lancet. 2018;391:2384–8.

Tesarz J, Eich W, Baumeister D, Kohlmann T, D'Agostino R, Schuster AK. Widespread pain is a risk factor for cardiovascular mortality: results from the Framingham Heart Study. Eur Heart J. 2019;40(20):1609–17.

Wettstein M, Eich W, Bieber C, Tesarz J. Pain intensity, disability, and quality of life in patients with chronic low back pain: does age matter? Pain Med. 2019;20(3):464–75.

Article   Google Scholar  

Williams AC de C, Fisher E, Hearn L, Eccleston C. Psychological therapies for the management of chronic pain (excluding headache) in adults. Cochrane Database Syst Rev. 2020;(8):Art. No.: CD007407. https://doi.org/10.1002/14651858.CD007407.pub4

Cohen SP, Vase L, Hooten WM. Chronic pain: an update on burden, best practices, and new advances. Lancet. 2021;397(10289):2082–97.

Edwards RR, Dworkin RH, Turk DC, Angst MS, Dionne R, Freeman R, et al. Patient phenotyping in clinical trials of chronic pain treatments: IMMPACT recommendations. Pain. 2016;157:1851–71.

Article   CAS   Google Scholar  

Williams ACC, Fisher E, Hearn L, Eccleston C. Psychological therapies for the management of chronic pain (excluding headache) in adults. Cochrane Database Syst Rev. 2020;8:2384-8.

Dworkin RH, McDermott MP, Farrar JT, O'Connor AB, Senn S. Interpreting patient treatment response in analgesic clinical trials: implications for genotyping, phenotyping, and personalized pain treatment. Pain. 2014;155(3):457–60.

Yarnitsky D, Arendt-Nielsen L, Bouhassira D, Edwards RR, Fillingim RB, Granot M, et al. Recommendations on terminology and practice of psychophysical DNIC testing. Eur J Pain. 2010;14(4):339.

Gerhardt A, Hartmann M, Tesarz J, Janke S, Leisner S, Seidler G, et al. Subgroups of musculoskeletal pain patients and their psychobiological patterns - The LOGIN study protocol. BMC Musculoskelet Disord. 2012;13:136.

Andrejeva N, Baumeister D, Eich W, Tesarz J. Psychosoziale Faktoren in der Prävention von Schmerz. Der Schmerz. 2021;35:21–9. https://doi.org/10.1007/s00482-020-00523-4 .

Flor H. New developments in the understanding and management of persistent pain. Curr Opin Psychiatry. 2012;25(2):109–13.

Dworkin RH, Turk DC, Wyrwich KW, Beaton D, Cleeland CS, Farrar JT, et al. Interpreting the clinical importance of treatment outcomes in chronic pain clinical trials: IMMPACT recommendations. J Pain. 2008;9(2):105–21.

Turk DC, Rudy TE. Toward an empirically derived taxonomy of chronic pain patients: Integration of psychological assessment data. J Consult Clin Psychol. 1988;56(2):233–8.

Diers M, Yilmaz P, Rance M, Thieme K, Gracely R, Rolko C, et al. Treatment-related changes in brain activation in patients with fibromyalgia syndrome. Exp Brain Res. 2012;218(4):619–28.

Kamping S, Bomba IC, Kanske P, Diesch E, Flor H. Deficient modulation of pain by a positive emotional context in fibromyalgia patients. Pain. 2013;154(9):1846–55.

Flor H. Pain has an element of blank-a biobehavioral approach to chronicity. Pain. 2017;158(p):92–6.

Tesarz J, Gerhardt A, Treede R, Eich W. Altered pressure pain thresholds and increased wind-up in adult chronic back pain patients with a history of childhood maltreatment: a quantitative sensory testing study. Pain. 2016;157(8):1799–809.

Gerhardt A, Leisner S, Hartmann M, Janke S, Seidler G, Eich W, et al. Eye movement desensitization and reprocessing vs. treatment-as-usual for non-specific chronic back pain patients with psychological trauma: a randomized controlled pilot study. Front. Psychiatry. 2016;7(201):1–10.

Google Scholar  

Bisson JI, Andrew M. Psychological treatment of post-traumatic stress disorder (PTSD). Cochrane Database Syst Rev. 2007;18(3):CD003388.

Tesarz J, Seidler GH, Eich W. Treatment of pain with EMDR (in German). Stuttgart: Klett-Cotta; 2015.

Tesarz J, Wicking M, Bernardy K, Seidler GH. EMDR therapy’s efficacy in the treatment of pain. J EMDR Prac Res. 391(4):337–44.

Thieme K, Turk DC, Flor H. Responder criteria for operant and cognitive-behavioral treatment of fibromyalgia syndrome. Arthritis Rheum. 2007;57(5):830–6.

Myin-Germeys I, Kasanova Z, Vaessen T, Vachon H, Kirtley O, Viechtbauer W, et al. Experience sampling methodology in mental health research: a contextual approach to psychiatry. World Psychiatry. 2018;17:123–32.

Myin-Germeys I, Steinhart H, Klippel A, Reininghaus U. Ecological momentary interventions in psychiatry. Curr Opin Psychiatry. 2016;29:258–63.

Reininghaus U. Ecological momentary interventions in psychiatry: the momentum for change in daily social context. Psychiatr Prax. 2018;45:59–61.

Flor H, Rudy TE, Birbaumer N. Zur Anwendbarkeit des West Haven-Yale Multidimensional Pain Inventory im deutschen Sprachraum. Schmerz. 1990;4(2):82–7.

Moher D, Chan AW. SPIRIT (standard protocol items: recommendations for interventional trials). Guidelines for Reporting Health Research: A User’s Manual. Hoboken: Wiley; 2014.

Eldridge S, Chan C, Campbell M, Bond C, Hopewell S, Thabane L, et al. CONSORT 2010 statement: extension to randomised pilot and feasibility trials. BMJ. 2016;355:2384-8.

Schmerz IPF. Improving outcomes in chronic musculoskeletal pain through a personalized medicine approach using cross-illness multilevel assessments and mechanism-based interventions: Klinik für Allgemeine Innere Medizin und Psychosomatik. Heidelberg: Universitätsklinikum Heidelberg; 2020.

Baek J, Lee S, Cho T, Kim S-W, Kim M, Yoon Y, et al. Neural circuits underlying a psychotherapeutic regimen for fear disorders. Nature. 2019;566(7744):339–43.

Tesarz J, Leisner S, Gerhardt A, Janke S, Seidler G, Eich W, et al. Effects of eye movement desensitization and reprocessing (EMDR) treatment in chronic pain patients: a systematic review. Pain Med. 2014;15(2):247–63.

Gerhardt A, Eich W, Janke S, Leisner S, Treede R-D, Tesarz J. Chronic widespread back pain is distinct from chronic local back pain: evidence from quantitative sensory testing, pain drawings, and psychometrics. Clin J Pain. 2016;32(7):568–79.

Nia NG, Afrasiabifar A, Behnammoghadam M. Comparing the effect of eye movement desensitization and reprocessing (EMDR) with guided imagery on pain severity in patients with rheumatoid arthritis. J Pain Res. 2018;11:2107–13.

Flor H, Birbaumer N. Erwerb chronischer Schmerzen. Aps J. 1994;3:127.

Flor H, Denke C, Schaefer M, Grüsser S. Effect of sensory discrimination training on cortical reorganisation and phantom limb pain. Lancet. 2001;357(9270):1763–4.

Foell J, Bekrater-Bodmann R, Diers M, Flor H. Mirror therapy for phantom limb pain: brain changes and the role of body representation. Eur J Pain. 2014;18(5):729–39.

Thieme K, Turk DC, Flor H. Responder criteria for operant and cognitive–behavioral treatment of fibromyalgia syndrome. Arthritis Care Res. 2007;57(5):830–6.

Thieme K, Turk DC, Gracely RH, Flor H. Differential psychophysiological effects of operant and cognitive behavioural treatments in women with fibromyalgia. Eur J Pain. 2016;20(9):1478–89.

Wei L, Hutson AD. A comment on sample size calculations for binomial confidence intervals. J Appl Stat. 2013;40(2):311–9.

Rozental A, Kottorp A, Boettcher J, Andersson G, Carlbring P. Negative effects of psychological treatments: an exploratory factor analysis of the negative effects questionnaire for monitoring and reporting adverse and unwanted events. PLoS One. 2016;11(6):e0157503.

Ladwig I, Rief W, Nestoriuc Y. Welche Risiken und Nebenwirkungen hat Psychotherapie? - Entwicklung des Inventars zur Erfassung Negativer Effekte von Psychotherapie (INEP). Verhaltenstherapie. 2014;24(4):252–63.

Williams AC de C, Fisher E, Hearn L, Eccleston C. Psychological therapies for the management of chronic pain (excluding headache) in adults. Cochrane Database Syst Rev. 2020;(8):Art. No.: CD007407. https://doi.org/10.1002/14651858.CD007407.pub4 .

Fairbank JCT, Pynsent PB. The Oswestry Disability Index. Spine (Phila Pa 1976). 2000;40:1609–17.

Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand. 1983;67(6):361–70.

Watson D, Clark LA, Tellegen A. Development and validation of brief measures of positive and negative affect: the PANAS Scales. J Pers Soc Psychol. 1988;54:1063–70.

Breyer B, Bluemke M. Deutsche Version der Positive and Negative Affect Schedule PANAS (GESIS Panel). Zusammenstellung sozialwissenschaftlicher Items und Skalen (ZIS); 2016.

Download references

Acknowledgements

We thank our colleagues Victoria Lucas, Dr. Med., and Nadeshda Andrejeva, Dr. Phil. for their assistance with the study.

Open Access funding enabled and organized by Projekt DEAL. This work is supported entirely by the German Federal Ministry of Education and Research (BMBF) (Grant no. 01EC1904A). The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing and publishing the report.

Author information

Authors and affiliations.

Department of General Internal Medicine and Psychosomatics, Heidelberg University, Heidelberg, Germany

E. Beiner, D. Baumeister, D. Buhai, W. Eich & J. Tesarz

Institute of Cognitive and Clinical Neuroscience, Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Heidelberg, Germany

M. Löffler, A. Löffler, C. Malone & H. Flor

Integrative Spinal Research Group, Department of Chiropractic Medicine, Balgrist University Hospital, University of Zürich, Zürich, Switzerland

University of Zürich, Zürich, Switzerland

Department of Public Mental Health; Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Heidelberg, Germany

A. Schick, L. Ader & U. Reininghaus

Experimental Radiation Oncology, Medical Faculty Mannheim, Heidelberg University, Heidelberg, Germany

A. Sirazitdinov & J. Hesser

Interdisciplinary Center for Clinical Trials, Johannes Gutenberg University Medical Center Mainz, Mainz, Germany

M. Hopp & C. Ruckes

You can also search for this author in PubMed   Google Scholar

Contributions

All authors contributed substantially to the conception and the design of the study. JT, DB and CR contributed to the data analysis plan. JT, DB and AN drafted the manuscript. All authors revised the manuscript critically for important intellectual content. All authors approved the version of the manuscript to be published and have agreed to be accountable for all aspects of the work.

Corresponding author

Correspondence to J. Tesarz .

Ethics declarations

Ethics approval and consent to participate.

This trial has undergone ethical scrutiny and has been approved by the Medical Faculty of the University of Heidelberg Ethics Committee (S-923/2019). Additionally, considering that the study will take place in routine general practice, we have obtained the ethical approval of the State Chamber of Physicians Baden-Wuerttemberg. Approvals from Hesse and Rhineland-Palatinate, where recruitment will start later, are pending. Written, informed consent to participate will be obtained from all participants.

Consent for publication

Competing interests.

The authors declare no competing interests. During the trial, all authors will comply with their respective institution’s policies on conflicts of interest.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1..

CONSORT 2010 statement: extension to randomized pilot and feasibility trials.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Beiner, E., Baumeister, D., Buhai, D. et al. The PerPAIN trial: a pilot randomized controlled trial of personalized treatment allocation for chronic musculoskeletal pain—a protocol. Pilot Feasibility Stud 8 , 251 (2022). https://doi.org/10.1186/s40814-022-01199-6

Download citation

Received : 27 December 2021

Accepted : 01 November 2022

Published : 09 December 2022

DOI : https://doi.org/10.1186/s40814-022-01199-6

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Chronic pain
• Precision therapy
• Personalized allocation
• Psychotherapy

Pilot and Feasibility Studies

ISSN: 2055-5784

• Submission enquiries: Access here and click Contact Us
• General enquiries: [email protected]

• Search Menu
• Advance Articles
• Author Guidelines
• Submission Site
• Open Access
• Self-Archiving Policy
• Why Publish
• Call for Papers
• About JAMIA Open
• About American Medical Informatics Association
• Editorial Board
• Advertising & Corporate Services
• Journals Career Network
• For Reviewers
• Journals on Oxford Academic
• Books on Oxford Academic

Article Contents

Lay summary, introduction, materials and methods, author contributions, supplementary material, conflict of interest statement, data availability.

• < Previous

Application of machine learning methods in clinical trials for precision medicine

• Article contents
• Figures & tables
• Supplementary Data

Yizhuo Wang, Bing Z Carter, Ziyi Li, Xuelin Huang, Application of machine learning methods in clinical trials for precision medicine, JAMIA Open , Volume 5, Issue 1, April 2022, ooab107, https://doi.org/10.1093/jamiaopen/ooab107

• Permissions Icon Permissions

A key component for precision medicine is a good prediction algorithm for patients’ response to treatments. We aim to implement machine learning (ML) algorithms into the response-adaptive randomization (RAR) design and improve the treatment outcomes.

We incorporated 9 ML algorithms to model the relationship of patient responses and biomarkers in clinical trial design. Such a model predicted the response rate of each treatment for each new patient and provide guidance for treatment assignment. Realizing that no single method may fit all trials well, we also built an ensemble of these 9 methods. We evaluated their performance through quantifying the benefits for trial participants, such as the overall response rate and the percentage of patients who receive their optimal treatments.

Simulation studies showed that the adoption of ML methods resulted in more personalized optimal treatment assignments and higher overall response rates among trial participants. Compared with each individual ML method, the ensemble approach achieved the highest response rate and assigned the largest percentage of patients to their optimal treatments. For the real-world study, we successfully showed the potential improvements if the proposed design had been implemented in the study.

In summary, the ML-based RAR design is a promising approach for assigning more patients to their personalized effective treatments, which makes the clinical trial more ethical and appealing. These features are especially desirable for late-stage cancer patients who have failed all the Food and Drug Administration (FDA)-approved treatment options and only can get new treatments through clinical trials.

In a typical controlled clinical trial, patients are equally randomized to receive different treatments. However, it is possible that one treatment demonstrates advantages over others during the trial. Utilizing that information can benefit subsequent patients. This is why response-adaptive randomization (RAR), which allows uneven treatment assignment probabilities based on existing knowledge, has become popular recently. A key component of RAR is a good prediction algorithm for patients’ response to treatments. Previous works have explored using machine learning (ML) to predict treatment response, but few incorporated ML methods into RAR. This study implements 9 commonly used ML methods into RAR trial designs. We further present an ML-ensemble RAR design that builds upon the majority consensus of the 9 ML methods’ predictions. Extensive simulation studies and real-world applications show that using ML methods in RAR leads to the assignment of more patients to their optimal treatments, increasing the overall response rate. The proposed method will become a useful tool for future clinical trial design in the era of precision medicine.

It is known that patients respond differently to the same treatments. 1 The demand for selecting the optimal treatment for each and every patient has resulted in a rapidly developing field called precision medicine, also known as personalized medicine. 2 This field aims to provide guidance to select the most effective treatment based on distinctive patient biomarkers. As clinical trials also evolve in the age of precision medicine, there is a substantial need for novel trial designs to deliver more ethical and precise care. Compared with classical nonadaptive trials, adaptive trials have become popular among clinicians as they integrate accumulating patient data to modify the parameters of the trial protocol, provide personalized treatment assignment, and ultimately optimize patients’ outcomes. For example, the adaptive designs in phase 2/3 clinical trials take advantage of the interim treatment response data during the course of the trial and allocate more patients to the presumably more effective treatments. 3

Among different adaptive designs, one common adaptation is response-adaptive randomization (RAR). It refers to the adjustments of treatment allocations based on intermediate patient responses and new patients’ characteristics collected during the clinical trials. This RAR design is useful when the interaction between biomarkers and treatments are only putative or not known at the beginning of a trial, and it is also practical when there are multiple treatments to be considered. Its ultimate objective is to provide more patients with their personalized optimal therapies according to their biomarker profiles. The starting point of RAR can be traced back to Thompson, 4 who proposed employing a posterior probability estimated from the interim data to assign patients to the more effective treatment. Following his idea, the application of Bayesian methods with an inherent adaptive nature has boomed in area of RAR designs. 5–13

Currently, there are several major successes in applying Bayesian RAR concepts in clinical trials, from protocol development through legitimate registration. The BATTLE-1 trial (Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination) for patients with advanced non–small cell lung cancer (NSCLC) and the I-SPY 2 trial (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis) for patients with breast cancer in the setting of neoadjuvant chemotherapy are 2 biomarker-based, Bayesian RAR clinical trials. 14 , 15 However, Bayesian RAR designs have a number of challenges and limitations. Due to the modeling restrictions, Bayesian RAR methods usually consider only a very small number of biomarkers. With complex diseases or symptoms, hundreds or even thousands of biomarkers may need to be considered at the same time for treatment assignment. 16 Also, some Bayesian RAR methods adjust the design by separating the cohort based on the existence of biomarker(s), and thus these methods rely heavily on how well the biomarker(s) interact(s) with the treatments. If the biomarker is chosen incorrectly, it is possible to make wrong adjustments afterwards. 1 , 17

As the development of modern sequencing technology, clinicians have faced a massive volume of high dimensional data with a complex, nonlinear structure. How to build an effective and scalable algorithm for randomization becomes a fundamental question for the research of RAR trial designs. Machine learning (ML) methods have been applied to solve many real-world problems and have successfully demonstrated their strengths in processing large data sets, as well as capturing nonlinear data structures. With the expectations and resources to analyze this large amount of complex healthcare data, ML methods have established their supremacy in disease prediction, 18 disease classification, 19 imaging diagnosis, 20 drug manufacturing, 21 medication assignment, 22 and genomic feature identification tasks. 23

Although several supervised ML approaches have been applied to drug response prediction, 10 , 24–34 little of the work has explored incorporating ML methods into RAR trial designs. In this study, we implemented 9 ML algorithms into RAR designs and further presented an ML-ensemble RAR design combining these 9 ML algorithms. Specifically, ML methods help to match patient biomarker profiles with prediction of treatment outcomes and, in turn, have determined treatment allocation for future patients. These ML methods are able to address large data and complex structures. We have successfully demonstrated, in both simulation study and a real-world example, that ML-based RAR designs have higher response rates as there are more patients receiving effective treatments. The ensemble method outperformed all other single ML methods.

Adaptive design: response-adaptive randomization

In clinical trial design, adaptive design means making changes to the trial protocol after the trial has started and some data have been collected. These changes are based on the information from the collected data, including (1) the total sample size, (2) interim analyses, (3) patient allocation to different treatment arms, and more. 35 For (3), it refers to the RAR design in which the treatment allocation probability varies in order to favor the treatment estimated to be more effective and to increase the response rate in patients. The initial concept can be traced back to Thompson 4 and Robbins, 36 and led to others. Some famous RAR trials include the extracorporeal membrane oxygenation (ECMO) trial, which tested the efficacy of ECMO in patients with severe acute respiratory distress syndrome (ARDS), 37 and the first large-scale double-blind, placebo-controlled study which tested the superiority of fluoxetine over placebo in children and adolescents with depression. 38 A general scheme of the RAR designs is shown in Figure 1 .

Response-adaptive randomization (RAR) design. The number of adaptive randomization is adjustable per application.

Benchmark design: equal randomization

Randomization as a standard means for addressing the selection bias in treatment assignments has been extensively used in clinical trials. 35 It helps to achieve balance among treatment groups and accounts for the genuine uncertainty about which treatment is better at the beginning of the trial. Randomly assigning patients to treatment arms on a 1:1 basis is known as equal randomization (ER). Friedman et al 39 (p. 41) presented that equal allocation in principle maximizes statistical power and is consistent with the concept of equipoise that should exist before the trial starts. Here, we used the ER design as a benchmark randomization design to evaluate the performance of ML-based RAR designs.

Allocation rule

Ml algorithms and a ml ensemble.

We selected 9 mainstream ML algorithms and implemented them in the RAR design to predict treatment response. The prediction models were built using the best-fitting parameters for each model, which were obtained by the grid search method with a 10-fold cross-validation. 40 , 41 Grid search is a standard method which allows us to try a variety of tuning parameter combinations for the model within a reasonable amount of time. The 10-fold cross-validation performs the fitting process for a total of 10 times with randomly selected nine-tenths of the data (90%) to train the model in each fit and the rest of the data to validate. By doing this, we avoid bias from using a random single split. The selected model will generalize better to all of the samples in the dataset. Combining grid search with cross-validation, we evaluate the performance of each parameter combination and select the best parameters for each ML model. Here we conducted this hyperparameter tuning procedure in R using the “Caret” package 42 ; similar techniques are available in the scikit-learn Python ML library. 43 These selected ML algorithms can be roughly divided into 2 categories:

Parametric models: logistic regression, 44 LASSO regression, 45 and Ridge regression. 46

Nonparametric models: gradient boosting machine (GBM), 47 random forest (RF), 48 support vector machine (SVM), 49 Naive Bayes, 50 k-nearest neighbors (KNN), 51 and artificial neural networks (NNs). 52

For logistic regression, Ridge regression and Lasso regression, they are all considered parametric models. In detail, logistic regression assumes the linearity of independent variables and log odds. It is a particular form of GLM. 24 Ridge regression and LASSO regression assume that there is a linear relationship between the “dependent” variable and the explanatory variables. They are 2 regularization methods of GLM to prevent an over-fitting issue by adding penalties on the predictor variables that are less significant. 46 , 53

KNN, which classifies data points based on the points that are most similar to it, is a typical nonparametric model such that there is no assumption for underlying data distribution, and the number of parameters grows with the size of the data. 51 With NNs, however, there has been some debate regarding whether they belong to parametric or nonparametric methods. NNs typically consist of 3 layers: input layer, hidden layer, and output layer. Here we classify NNs as a nonparametric method, as the network architecture grows adaptively to match the complexity of given data. 52

Both GBM and RF are nonparametric methods that consist of sets of decision trees. Specifically, GBM builds one tree at a time and each new tree helps to correct errors made by previously trained tree by adding weights to the observations with the worst prediction from the previous iteration; RF trains each tree independently using a random sample of the data, and the results are aggregated in the end. 47 , 48

NB and SVM can be either parametric or nonparametric depending on whether they use kernel tricks. For the NB classifier, it becomes nonparametric if using a kernel density estimation (KDE) to obtain a more realistic estimate of the probability of an observation belonging to a class. 50 And for SVM, the basic idea is finding a hyperplane that best divides a dataset into 2 classes. It is considered a nonparametric when using the kernel trick to find this hyperplane. This is because the kernel is constructed by computing the pair-wise distances between the training points, and the complexity of the model grows with the size of the dataset. 49

Apart from comparing with the ER “benchmark” design, the current study also examined whether the ML ensemble could assign more patients to the best available treatment beyond other ML methods in adaptive design with the same assessment of individuals.

Inverse probability of treatment weighting

Evaluation metrics.

A low individual loss value suggests that the majority of patients have received the treatment and will respond at least as well as the real model’s optimal therapy.

We used simulation studies to evaluate the proposed methods.

Seven scenarios of different treatment main effects ( ⁠ α =0, 0.5, 0.7, 1, 1.3, 1.5, 1.7) and a fixed treatment-biomarker interaction ( ⁠ γ = 0.5 ⁠ ) were considered. We conducted 1000 Monte Carlo simulations for each scenario and compared the results obtained by the ML-based and ML-ensemble RAR designs with the results from the ER design.

Response rate and optimal treatment percentage

The response rate results and the percentage of receiving the optimal treatment are shown in Figure 2 . Overall, the performance of ML-based RAR designs is better than the performance of the ER design. When the treatment main effect is zero, the differences for both response rate and the optimal treatment percentage between ML-based RAR designs and the ER design are not significant. As the treatment effect increases, these differences become more obvious. Among these 9 ML algorithms, the neural network has the highest response rate and the highest proportion of patients receiving their optimal treatments. Additionally, the ensemble method combining these 9 ML methods outperforms all other methods and achieves an approximate 5% higher response rate and a more than 20% larger optimal treatment percentage compared to the ER design.

Simulation result: response rate (left), percentage of patients receiving their optimal treatments (right). The treatment-biomarker interaction, γ is fixed at 0.5. Boxplots display the median (middle line), the interquartile range (hinges), and 1.5 times the interquartile range (lower and upper whiskers) based on 1000 times simulation. The mean (over 1000 simulations) response rate ranges from 0.53 to 0.69, and the mean of optimal treatment percentages ranges from 0.50 to 0.71.

Individual loss, ATE, and power

The individual loss and the ATE results are shown in Figure 3 . The interpretation of the individual loss results coincides with the previous response rate results and the optimal treatment percentage results such that the ML-ensemble RAR design has the lowest individual loss value among all scenarios, which is preferred in the trial. The ATE has been adjusted by the IPTW method to account for confounding effect of using observational data. The logistic regression method now has the highest ATE, followed by the NN method. The ensemble method has a relatively low ATE, but it is higher than the ER method when the treatment main effect becomes larger. This shows that the average effect of changing the entire population from untreated to treated using RAR designs is better than that of using the ER design.

Simulation result: individual loss (left), average treatment effect (ATE, right). The treatment-biomarker interaction, γ is fixed at 0.5. Boxplots display the median (middle line), the interquartile range (hinges), and 1.5 times the interquartile range (lower and upper whiskers) based on 1000 times simulation. The mean (over 1000 simulations) individual loss ranges from 0.04 to 0.10, and the mean ATE ranges from -0.12 to 0.30.

The power results are shown in Figure 4 . The power is also weighted by the IPTW method to address potential bias. For the power analysis, the Type I error is controlled at 0.05. Several papers have shown in their simulation studies that the correlation among treatment assignments was inevitable when performing inference on the data from RAR design-implemented studies. 56 , 57 This correlation can increase the binomial variability and lower the power. In our simulation, the RAR design using the NN method has the lowest power, followed by using the logistic regression method. However, other ML-based RAR designs have comparable or even higher power than that of the ER design. The ensemble method has a relatively low power, but it is still better than the NN method.

Simulation result: power. The treatment-biomarker interaction, γ is fixed at 0.5. The Type I error is controlled at 0.05. The power ranges from 0.04 to 0.97.

Real-world example

We analyzed a publicly available acute myeloid leukemia (AML) dataset from Kornblau et al 58 where most of the clinical biomarkers are expression levels of cellular proteins. Kornblau et al sequenced protein expressions in leukemia-enriched cells from 256 newly diagnosed AML patients with a primary goal of eventually establishing a proteomic-based categorization of AML. The treatment and the response variables were carefully adjusted to binary variables. Specifically, the treatments were binarized to high-dose ara-C (HDAC)–based treatments and non-HDAC treatments; the responses were binarized to complete response (CR) and non-CR.

We first performed a feature selection to decide what interaction terms should be included in the model. We used each protein-treatment interaction term to build the generalized linear model (GLM) model and reported the p-value for each interaction to assess whether it has strong correlation with the dependent variable/the treatment response. The top 10 proteins whose interaction variables have the smallest P -values were selected. We then performed a gene network analysis on the genes that code for these proteins using GeneMANIA ( http://genemania.org ). 26 This analysis helps to illustrate the hidden interaction and network of the corresponding genes. Additionally, it shows other genes that have been reported to associate with the input 10 genes, using extensive existing knowledge such as protein and genetic interactions, pathways, co-expression, co-localization, and protein domain similarity. The results are presented in Figure 5 . The top 10 genes corresponding to the biomarkers identified in our study are highlighted with red circles.

AML data: the gene network analysis. The input 10 genes, namely the genes coding for top 10 proteins that significantly interacted with the treatment, were highlighted using red circles. Other genes that were presumably involved in AML were returned by GeneMANIA.

Using a cut-off P -value of 0.1 among 71 proteins, the expression levels of 3 of them were found to have the most significant interactions with the treatment, that is, the strongest correlation with the treatment outcomes: phosphothreonine 308 of Akt (Thr 308 p-Akt), the mechanistic target of rapamycin (mTOR), and signal transducer and activator of transcription 1 (STAT1). Studies have shown that these 3 proteins play critical roles in human AML. The level of Thr 308 p-Akt is associated with high-risk cytogenetics and predicts poor overall survival for AML patients. 27 In AML, the mTOR signaling pathway is deregulated and activated as a consequence of genetic and cytogenetic abnormalities. The mTOR inhibitors are often used to target aberrant mTOR activation and signaling. 59 , 60 The STAT1 transcription factor is constitutively activated in human AML cell lines and might contribute to the autonomous proliferation of AML blasts. The inhibition of this pathway can be of great interest for AML treatments. 61 , 62 Hence, we chose these 3 proteins to build ML models in our proposal.

The whole dataset (256 observations) was randomly shuffled and divided into 2 equal-sized blocks: block 1 and block 2. Each block was taken in turn as either the training set or the testing set. The results were aggregated after 100 repetitions. Since this clinical trial is already completed and it is not possible to get actual treatment responses using our methods, we separated the enrolled patients into 2 groups: a consistent group whose real treatments are the same as the treatments using the ML-based RAR designs and an inconsistent group whose real treatments are different from the treatments using the ML-based RAR designs. We compared the response rates in these 2 groups to elucidate the potential gain if the proposed RAR had been implemented. The results of each method are shown in Figure 6 . In the consistent group, the response patient percentages are at least 10% higher than 50%; while the response patient percentages in the inconsistent group are all lower than 50%, that is, we observe higher response rates in the consistent group. This means that patients in the inconsistent group may likely benefit from the RAR method we developed.

AML data result: the response percentage. Patients in the consistent group (left) were assigned to the same treatments using our ML-based RAR designs, while patients in the inconsistent group (right) were assigned to different treatments using our ML-based RAR designs. The 50% response percentage is marked with a black dashed line.

Patients are accrued in groups sequentially. RAR designs determine the treatment allocation for new groups of patients based on the accrued information of how previous groups of patients responded to their treatments. The number of RAR implementations, k ⁠ , should be predefined. The choice of k may depend on the total sample size, trial length, and other logistics and practical considerations. Our simulation study used k = 2 for a total sample size of 300 ⁠ . In the real data analysis with a smaller sample size of 256 subjects, we used k = 1 ⁠ .

We developed novel methods for RAR designs by incorporating 9 ML methods to predict treatment response and assign treatments accordingly. We showed that our ML-based RAR designs can effectively improve treatment response rates among patients. We further proposed an ensemble approach based on the consensus of the 9 ML methods to improve the prediction and decision making. Our proposed ML-ensemble RAR design builds on the predictive ability of 9 ML methods and can further improve predication accuracy and patient outcome. Specifically, suppose m out of 9 models indicate that treatment A is better than treatment B for patient i ⁠ , then we let p i A = m / 9 for 2 ≤ m ≤ 7 ⁠ , let p i A = 0.85 for m ≥ 7 ⁠ , and p i A = 0.15 for m ≤ 2 ⁠ . For m ≥ 7 ⁠ , we keep the assignment probability as a constant of 0.85 because we still want to reserve some randomness in the trial. These settings can be tuned based on prior knowledge of the treatment selections.

We also tried the combination of NN and GLM algorithms as another binary-combination method and conducted additional simulations. Since these 2 models may not always be in consensus regarding optimal treatment selection for each individual, we took the average of the treatment assigning probabilities of the NN and GLM methods. Similar as we have done previously for the 9 ML methods and the ensemble approach, we evaluated its performance through the overall response rate in simulated trials, the percentage of patients receiving their individually optimal treatment, and the average individual loss for all trial participants. We have provided the results in the Supplementary Figure S5 . Although the performance of this combination is slightly better than using either NN or GLM alone, it is still substantially worse than that of the ensemble using 9 ML algorithms, especially when the treatment main effect is high.

While we only considered settings of 2 treatment options in this work, ML-based RAR design can extend to multiple targeted treatments. Given L treatments, the l th treatment allocating probability of patient i is shown as p i l = π i l / ∑ l = 1 L π i l , l = 1 , … , L ⁠ , where π i l denotes the response probability of l t h treatment for patient i predicted by the ML algorithm. For example, NN can naturally adapt to a multiclass classification problem by replacing the binary cross-entropy loss to a categorical cross-entropy loss. 63

Although our work can effectively improve the treatment outcomes in the clinical trial, there are a few limitations that we would like to point out as directions for further research. First, equal weight was given to each of the 9 ML algorithms in the ensemble method. However, it is likely that different ML methods have distinct prediction accuracy at different scenarios. Incorporating such information by attaching different weights for different ML algorithms in the ensemble method could potentially lead to better adaptation to the data and may provide more precise treatment suggestions for personalized medicine. Second, although our method has been extensively evaluated using simulated and real data, we did not consider the setting with high-dimensional data, for example, the data from omics experiments. With the development of modern sequencing technology, more clinical trials seek to include such information in clinical decision making and trial design. With high-dimensional data, there are more challenges, such as adding appropriate feature selection steps, etc. Moreover, our current model did not consider the situation when complex interactions between treatment and individualized biomarkers exist in the dataset. When this problem is of interest, we might resort to other models that are specifically designed to address the heterogeneous treatment effect caused by these interactions, such as the honest causal forest model, 64 that are specifically designed to address the heterogeneous treatment effect caused by these interactions.

ML methods have successfully demonstrated their superior prediction performance in many applications, but have not been applied to conduct RAR in clinical trials. In this study, we developed novel methods for RAR designs by incorporating ML algorithms to predict treatment response and assign treatments accordingly. We showed that the ML-based RAR designs have better performance than that of the traditional ER design. And the ensemble approach demonstrated better results than the ER design at the greatest extent. As the ML field is getting mature and abundant packages are available on different programming software, our method is easy to implement in current clinical trial systems.

The research of XH was partially supported by the US National Institutes of Health grants U54CA096300, U01CA253911, and 5P50CA100632, and the Dr. Mien-Chie Hung and Mrs. Kinglan Hung Endowed Professorship.

XH, ZL, and YW conceived the concept of the study and designed the method. YW implemented the method, performed the experiments, and drafted the initial manuscript. BC interpreted the real-word data for the work. All authors edited and approved the final manuscript.

Supplementary material is available at JAMIA Open online.

None declared.

Simulation data can be reproduced by the R script that has been deposited in the online Dryad repositor. 65 The AML dataset used for real-world illustration can be downloaded from https://bioinformatics.mdanderson.org/public-datasets/supplements/ under “RPPA Data in AML”. 58

Antoniou M , Kolamunnage-Dona R , Jorgensen AL. Biomarker-guided non-adaptive trial designs in phase II and Phase III: a methodological review . JPM 2017 ; 7 ( 1 ): 1 .

Google Scholar

Ginsburg GS , Phillips KA. Precision medicine: from science to value . Health Aff (Millwood) 2018 ; 37 ( 5 ): 694 – 701 .

Maca J , Bhattacharya S , Dragalin V , Gallo P , Krams M. Adaptive seamless phase II/III designs—background, operational aspects, and examples . Drug Information J 2006 ; 40 ( 4 ): 463 – 73 .

Thompson WR. On the likelihood that one unknown probability exceeds another in view of the evidence of two samples . Biometrika 1933 ; 25 ( 3-4 ): 285 – 94 .

Bello GA , Sabo RT. Outcome-adaptive allocation with natural lead-in for three-group trials with binary outcomes . J Stat Comput Simul 2016 ; 86 ( 12 ): 2441 – 9 .

Zhou X , Liu S , Kim ES , Herbst RS , Lee JJ. Bayesian adaptive design for targeted therapy development in lung cancer—a step toward personalized medicine . Clin Trials 2008 ; 5 ( 3 ): 181 – 93 .

Williamson SF , Jacko P , Villar SS , Jaki T. A Bayesian adaptive design for clinical trials in rare diseases . Comput Stat Data Anal 2017 ; 113 : 136 – 53 .

Trippa L , Lee EQ , Wen PY , et al.  Bayesian adaptive randomized trial design for patients with recurrent glioblastoma . J Clin Oncol 2012 ; 30 ( 26 ): 3258 – 63 .

Brannath W , Zuber E , Branson M , et al.  Confirmatory adaptive designs with Bayesian decision tools for a targeted therapy in oncology . Stat Med 2009 ; 28 ( 10 ): 1445 – 63 .

Lee JJ , Xuemin G , Suyu L. Bayesian adaptive randomization designs for targeted agent development . Clin Trials 2010 ; 7 ( 5 ): 584 – 96 .

Lee JJ , Chu CT. Bayesian clinical trials in action . Stat Med 2012 ; 31 ( 25 ): 2955 – 72 .

Stallard N , Whitehead J , Cleall S. Decision-making in a phase II clinical trial: a new approach combining Bayesian and frequentist concepts . Pharm Stat 2005 ; 4 ( 2 ): 119 – 28 .

Dong G , Shih WJ , Moore D , Quan H , Marcella S. A Bayesian-frequentist two-stage single-arm phase II clinical trial design . Stat Med 2012 ; 31 ( 19 ): 2055 – 67 .

Kim ES , Herbst RS , Wistuba II , et al.  The BATTLE trial: personalizing therapy for lung cancer . Cancer Discov 2011 ; 1 ( 1 ): 44 – 53 .

Barker AD , Sigman CC , Kelloff GJ , Hylton NM , Berry DA , Esserman LJ. I-SPY 2: an adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy . Clin Pharmacol Ther 2009 ; 86 ( 1 ): 97 – 100 .

Nevozhay D , Adams RM , Murphy KF , Josi K , Balzsi G. Negative autoregulation linearizes the dose-response and suppresses the heterogeneity of gene expression . Proc Natl Acad Sci U S A 2009 ; 106 ( 13 ): 5123 – 8 .

Lee JJ , Liu S , Gu X. Bayesian adaptive randomization designs for targeted agent development . Clin Cancer Res 2008 ; 14 (19 Suppl ): PL06 .

Ma F , Gao J , Suo Q , You Q , Zhou J , Zhang A. Risk prediction on electronic health records with prior medical knowledge. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. London, UK: Association for Computing Machinery; 2018 : 1910 – 9 .

Fakoor R, Ladhak F, Nazi A, Huber M. Using deep learning to enhance cancer diagnosis and classification . In: Proceedings of the International Conference on Machine Learning; 2013 : 3937 – 49 .

Liu S, Liu S, Cai W, Pujol S, Kikinis R, Feng D. Early diagnosis of Alzheimer's disease with deep learning . In: 2014 IEEE 11th International Symposium on Biomedical Imaging (ISBI); 2014 : 1015 – 8 .

Pantuck A , Lee D-K , Kee T , et al.  Modulating BET bromodomain inhibitor ZEN‐3694 and enzalutamide combination dosing in a metastatic prostate cancer patient using CURATE.AI, an artificial intelligence platform . Adv Ther 2018 ; 1 ( 6 ): 1800104 .

Liang Z , Zhang G , Huang X , Hu QV. Deep learning for healthcare decision making with EMRs. IEEE Int Conf Bioinform Biomed 2014 : 556 – 9 . doi 10.1109/BIBM.2014.6999219.

Zhou J , Troyanskaya OG. Predicting effects of noncoding variants with deep learning-based sequence model . Nat Methods 2015 ; 12 ( 10 ): 931 – 4 .

Tibshirani R , Hastie T , Friedman J. Regularized paths for generalized linear models via coordinate descent . J Stat Soft 2010 ; 33 ( 1 ): 1–22.

Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies . Multivariate Behav Res 2011 ; 46 ( 3 ): 399 – 424 .

Warde-Farley D , Donaldson SL , Comes O , et al.  The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene function . Nucleic Acids Res 2010 ; 38 ( Web Server issue ): W214 – 20 .

Gallay N , Santos CD , Cuzin L , et al.  The level of AKT phosphorylation on threonine 308 but not on serine 473 is associated with high-risk cytogenetics and predicts poor overall survival in acute myeloid leukemia . Blood 2008 ; 112 ( 11 ): 1503 .

Chekroud AM , , Zotti RJ , , Shehzad Z , et al.  Cross-trial prediction of treatment outcome in depression: a machine learning approach . Lancet Psychiatry 2016 ; 3 ( 3 ): 243 – 50 .

Shimabukuro DW , , Barton CW , , Feldman MD , , Mataraso SJ , , Das R. Effect of a machine learning-based severe sepsis prediction algorithm on patient survival and hospital length of stay: a randomised clinical trial . BMJ Open Respir Res 2017 ; 4 ( 1 ): e000234 .

Zazzi M , , Incardona F , , Rosen-Zvi M , et al.  Predicting response to antiretroviral treatment by machine learning: the EuResist project . Intervirology 2012 ; 55 ( 2 ): 123 – 7 .

Patel MJ , , Andreescu C , , Price JC , , Edelman KL , , Reynolds CF , , Aizenstein HJ. Machine learning approaches for integrating clinical and imaging features in late-life depression classification and response prediction . Int J Geriatr Psychiatry 2015 ; 30 ( 10 ): 1056 – 67 .

Schmidt-Erfurth U , , Bogunovic H , , Sadeghipour A , et al.  Machine learning to analyze the prognostic value of current imaging biomarkers in neovascular age-related macular degeneration . Ophthalmol Retina 2018 ; 2 ( 1 ): 24 – 30 .

Rajkomar A , , Dean J , , Kohane I. Machine learning in medicine . N Engl J Med 2019 ; 380 ( 14 ): 1347 – 58 .

Vamathevan J , , Clark D , , Czodrowski P , et al.  Applications of machine learning in drug discovery and development . Nat Rev Drug Discov 2019 ; 18 ( 6 ): 463 – 77 .

Rosenberger WF , Lachin JM. Randomization in Clinical Trials: Theory and Practice . Hoboken, NJ : John Wiley & Sons; 2015 .

Google Preview

Robbins H. Some aspects of the sequential design of experiments . Bull Amer Math Soc 1952 ; 58 ( 5 ): 527 – 35 .

Combes A , Hajage D , Capellier G ; EOLIA Trial Group, REVA, and ECMONet , et al.  Extracorporeal membrane oxygenation for severe acute respiratory distress syndrome . N Engl J Med 2018 ; 378 ( 21 ): 1965 – 75 .

Emslie G , Rush A , Weinberg W , et al.  A double-blind, randomized, placebo-controlled trial of fluoxetine in children and adolescents with depression . Arch Gen Psychiatry 1997 ; 54 ( 11 ): 1031 – 7 .

Friedman LM , Furberg C , DeMets DL. Fundamentals of Clinical Trials . Boston, MA : J. Wright, PSG Inc .; 1981 .

Stone M. Cross-validatory choice and assessment of statistical predictions . J R Stat Soc Ser B (Methodol) 1974 ; 36 ( 2 ): 111 – 33 .

Bergstra J , Bengio Y. Random search for hyper-parameter optimization . J Mach Learn Res 2012 ; 13 ( 10 ): 281 – 305 .

Kuhn M. caret: Classification and Regression Training; 2015 . [Database] https://github.com/topepo/caret/ Accessed October 9, 2021.

Pedregosa F , Varoquaux G , Gramfort A , et al.  Scikit-learn: machine learning in Python . J Mach Learn Res 2011 ; 12 : 2825 – 30 .

Cramer JS. The Origins of Logistic Regression . Tinbergen Institute Discussion Papers 02-119/4. Amsterdam, Netherlands : Tinbergen Institute; 2002 .

Tibshirani R. Regression shrinkage and selection via the Lasso . J R Stat Soc Ser B (Methodol) 1996 ; 58 ( 1 ): 267 – 88 .

Hoerl AE , Kennard RW. Ridge regression: biased estimation for nonorthogonal problems . Technometrics 1970 ; 12 ( 1 ): 55 – 67 .

Friedman JH. Greedy function approximation: a gradient boosting machine . Ann Statist 2001 ; 29 ( 5 ): 1189 – 232 .

Ho TK. Random Decision Forests . In: Proceedings of 3rd international conference on document analysis and recognition. Vol. 1. Manhattan, NY: IEEE Oxford University Press ; 1995 : 278 – 82 .

Boser BE , Guyon IM , Vapnik VN. A Training Algorithm for Optimal Margin Classifiers . In: Proceedings of the fifth annual workshop on Computational learning theory (COLT '92). New York, NY: Association for Computing Machinery (ACM); 1992 : 338 – 45 .

John GH , Langley P. Estimating Continuous Distributions in Bayesian Classifiers . In: Proceedings of the Eleventh conference on Uncertainty in artificial intelligence (UAI'95). San Francisco, CA : Morgan Kaufmann Publishers; 1995 : 338 – 45 .

Altman NS. An introduction to Kernel and nearest-neighbor nonparametric regression . Am Statist 1992 ; 46 ( 3 ): 175 – 85 .

Grossberg S. Nonlinear neural networks: principles, mechanisms, and architectures . Neural Networks 1988 ; 1 ( 1 ): 17 – 61 .

Imbens G. Nonparametric estimation of average treatment effects under exogeneity: a review . Rev Econ Stat 2004 ; 86 (1): 4 – 29 .

Rosenbaum PR. Model-based direct adjustment . J Am Stat Assoc 1987 ; 82 ( 398 ): 387 – 94 .

Lunceford JK , Davidian M. Stratification and weighting via the propensity score in estimation of causal treatment effects: a comparative study . Stat Med 2004 ; 23 ( 19 ): 2937 – 60 .

Melfi VF , Page C. Variability in adaptive designs for estimation of success probabilities . Project Euclid 1998 ; 34 : 106 – 14 .

Rosenberger WF , Stallard N , Ivanova A , Harper CN , Ricks ML. Optimal adaptive designs for binary response trials . Biometrics 2001 ; 57 ( 3 ): 909 – 13 .

Kornblau SM , Tibes R , Qiu YH , et al.  Functional proteomic profiling of AML predicts response and survival . Blood 2009 ; 113 ( 1 ): 154 – 64 .

Feng Y , Chen X , Cassady K , et al.  The role of mTOR inhibitors in hematologic disease: from bench to bedside . Front Oncol 2020 ; 10 : 611690 .

Sandhöfer N , Metzeler KH , Rothenberg M , et al.  Dual PI3K/mTOR inhibition shows antileukemic activity in MLL-rearranged acute myeloid leukemia . Leukemia 2015 ; 29 ( 4 ): 828 – 38 .

Gouilleux-Gruart V , Gouilleux F , Desaint C , et al.  STAT-related transcription factors are constitutively activated in peripheral blood cells from acute leukemia patients . Blood 1996 ; 87 ( 5 ): 1692 – 97 .

Weber-Nordt RM , Egen C , Wehinger J , et al.  Constitutive activation of STAT proteins in primary lymphoid and myeloid leukemia cells and in Epstein-Barr Virus (EBV)–related lymphoma cell lines . Blood 1996 ; 88 ( 3 ): 809 – 16 .

Goodfellow I , Bengio Y , Courville A. Deep Learning . Cambridge, MA : MIT Press; 2016 .

Athey S , Tibshirani J , Wager S. Generalized Random Forests; 2018 . https://grf-labs.github.io/grf/ Accessed July 14, 2021.

Wang Y , Huang XH , Li Z. Application of machine learning methods in clinical trials for precision medicine . Dryad 2021 .

Supplementary data

Email alerts, citing articles via.

• Advertising and Corporate Services

Affiliations

• Online ISSN 2574-2531
• Copyright © 2024 International American Medical Informatics Association
• About Oxford Academic
• Publish journals with us
• University press partners
• What we publish
• New features  
• Open access
• Institutional account management
• Rights and permissions
• Get help with access
• Accessibility
• Advertising
• Media enquiries
• Oxford University Press
• Oxford Languages
• University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

• Copyright © 2024 Oxford University Press
• Cookie settings
• Cookie policy
• Privacy policy
• Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

Subscribe to the PwC Newsletter

Join the community, edit social preview.

Add a new code entry for this paper

Remove a code repository from this paper, mark the official implementation from paper authors, add a new evaluation result row, remove a task.

Add a method

Remove a method, edit datasets, personalized assignment to one of many treatment arms via regularized and clustered joint assignment forests.

1 Nov 2023  ·  Rahul Ladhania , Jann Spiess , Lyle Ungar , Wenbo Wu · Edit social preview

We consider learning personalized assignments to one of many treatment arms from a randomized controlled trial. Standard methods that estimate heterogeneous treatment effects separately for each arm may perform poorly in this case due to excess variance. We instead propose methods that pool information across treatment arms: First, we consider a regularized forest-based assignment algorithm based on greedy recursive partitioning that shrinks effect estimates across arms. Second, we augment our algorithm by a clustering scheme that combines treatment arms with consistently similar outcomes. In a simulation study, we compare the performance of these approaches to predicting arm-wise outcomes separately, and document gains of directly optimizing the treatment assignment with regularization and clustering. In a theoretical model, we illustrate how a high number of treatment arms makes finding the best arm hard, while we can achieve sizable utility gains from personalization by regularized optimization.

Code Edit Add Remove Mark official

Tasks edit add remove, datasets edit, results from the paper edit add remove, methods edit add remove.

Advertisement

Can Artificial Intelligence Improve Psychotherapy Research and Practice?

• Published: 26 July 2020
• Volume 47 , pages 852–855, ( 2020 )

Cite this article

• Rachel L. Horn   ORCID: orcid.org/0000-0002-0646-3757 1 &
• John R. Weisz   ORCID: orcid.org/0000-0002-5560-6814 1  

2188 Accesses

16 Citations

Explore all metrics

Leonard Bickman’s article on the future of artificial intelligence (AI) in psychotherapy research paints an encouraging picture of the progress to be made in this field. We support his perspective, but we also offer some cautionary notes about the boost AI can provide. We suggest that AI is not likely to transform psychotherapy research or practice to the degree seen in pharmacology and medicine because the factors that contribute to treatment response in these realms differ so markedly from one another, and in ways that do not favor advances in psychotherapy. Despite this limitation, it seems likely that AI will have a beneficial impact, improving empirical analysis through data-driven model development, tools for addressing the limitations of traditional regression methods, and novel means of personalizing treatment. In addition, AI has the potential to augment the reach of the researcher and therapist by expanding our ability to gather data and deliver interventions beyond the confines of the lab or clinical office.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price includes VAT (Russian Federation)

Instant access to the full article PDF.

Rent this article via DeepDyve

Institutional subscriptions

Similar content being viewed by others

Improving Mental Health Services: A 50-Year Journey from Randomized Experiments to Artificial Intelligence and Precision Mental Health

Leonard Bickman

A Drifters Handbook

Artificial Intelligence and Humanistic Medicine: A Symbiosis

Andrews, G., Basu, A., Cuijpers, P., Craske, M. G., McEvoy, P., English, C. L., et al. (2018). Computer therapy for the anxiety and depression disorders is effective, acceptable and practical health care: An updated meta-analysis. Journal of Anxiety Disorders, 55 , 70–78. https://doi.org/10.1016/j.janxdis.2018.01.001 .

Article   CAS   PubMed   Google Scholar  

Chung, Y., Addington, J., Bearden, C. E., Cadenhead, K., Cornblatt, B., Mathalon, D. H., … Walker, E. (2018). Use of machine learning to determine deviance in neuroanatomical maturity associated with future psychosis in youths at clinically high risk. JAMA Psychiatry, 75 (9), 960–968. https://doi.org/10.1001/jamapsychiatry.2018.1543 .

DeRubeis, R. J. (2019). The history, current status, and possible future of precision mental health. Behaviour Research and Therapy, 123 , 103506. https://doi.org/10.1016/j.brat.2019.103506 .

Article   PubMed   Google Scholar  

DeRubeis, R. J., Cohen, Z. D., Forand, N. R., Fournier, J. C., Gelfand, L. A., & Lorenzo-Luaces, L. (2014). The personalized advantage index: Translating research on prediction into individualized treatment recommendation. A demonstration. PLoS ONE, 9 (1), e83875. https://doi.org/10.1371/journal.pone.0083875 .

Article   CAS   PubMed   PubMed Central   Google Scholar  

Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): A randomized controlled trial. JMIR Mental Health, 4 (2), e19. https://doi.org/10.2196/mental.7785 .

Article   PubMed   PubMed Central   Google Scholar  

Gunduz, N., & Fokoué, E. (2015). Robust classification of high dimension low sample size data . http://arxiv.org/abs/1501.00592 .

Kleiman, E. M., Turner, B. J., Fedor, S., Beale, E. E., Huffman, J. C., & Nock, M. K. (2017). Examination of real-time fluctuations in suicidal ideation and its risk factors: Results from two ecological momentary assessment studies. Journal of Abnormal Psychology, 126 (6), 726–738. https://doi.org/10.1037/abn0000273 .

Lorenzo-Luaces, L., DeRubeis, R., Van Straten, A., & Tiemens, B. (2017). A prognostic index (PI) as a moderator of outcomes in the treatment of depression: A proof of concept combining multiple variables to inform risk-stratified stepped care models. Journal of Affective Disorders, 213 , 78–85. https://doi.org/10.1016/j.jad.2017.02.010 .

Ng, M. Y., & Weisz, J. R. (2016). Building a science of personalized intervention for youth mental health. Journal of Child Psychology and Psychiatry, 57 (3), 216–236. https://doi.org/10.1111/jcpp.12470 .

Schleider, J. L., & Weisz, J. R. (2017). Little treatments, promising effects: Meta-analysis of single-session interventions for youth psychiatric problems. Journal of the American Academy of Child and Adolescent Psychiatry, 56 , 107–115. https://doi.org/10.1016/j.jaac.2016.11.007 .

van Bronswijk, S. C., DeRubeis, R. J., Lemmens, L. H., Peeters, F. P., Keefe, J. R., Cohen, Z. D., et al. (2019). Precision medicine for long-term depression outcomes using the personalized advantage index approach: Cognitive therapy or interpersonal psychotherapy? Psychological Medicine . https://doi.org/10.1017/S0033291719003192 .

Weisz, J. R., Kuppens, S., Ng, M. Y., Vaughn-Coaxum, R. A., Ugueto, A. M., Eckshtain, D., et al. (2019). Are psychotherapies for young people growing stronger? Tracking trends over time for youth anxiety, depression, ADHD, and conduct problems. Perspectives on Psychological Science, 14 (2), 216–237. https://doi.org/10.1177/174569161880543 .

Download references

Author information

Authors and affiliations.

Department of Psychology, Harvard University, 33 Kirkland St, Cambridge, MA, 02138, USA

Rachel L. Horn & John R. Weisz

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Rachel L. Horn .

Ethics declarations

Conflict of interest.

We have no known conflict of interest to disclose.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

This article is part of the Festschrift for Leonard Bickman Special Issue on The Future of Children's Mental Health Services.

Rights and permissions

Reprints and permissions

About this article

Horn, R.L., Weisz, J.R. Can Artificial Intelligence Improve Psychotherapy Research and Practice?. Adm Policy Ment Health 47 , 852–855 (2020). https://doi.org/10.1007/s10488-020-01056-9

Download citation

Published : 26 July 2020

Issue Date : September 2020

DOI : https://doi.org/10.1007/s10488-020-01056-9

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Artificial intelligence
• Psychotherapy
• Methodological limitations
• Machine learning
• Find a journal
• Publish with us
• Track your research

Bayesian predictive modeling for genomic based personalized treatment selection

Affiliation.

• 1 Department of Biostatistics, The University of Texas MD Anderson Cancer Center, Houston, Texas 77030, U.S.A.
• PMID: 26575856
• PMCID: PMC4870163
• DOI: 10.1111/biom.12448

Efforts to personalize medicine in oncology have been limited by reductive characterizations of the intrinsically complex underlying biological phenomena. Future advances in personalized medicine will rely on molecular signatures that derive from synthesis of multifarious interdependent molecular quantities requiring robust quantitative methods. However, highly parameterized statistical models when applied in these settings often require a prohibitively large database and are sensitive to proper characterizations of the treatment-by-covariate interactions, which in practice are difficult to specify and may be limited by generalized linear models. In this article, we present a Bayesian predictive framework that enables the integration of a high-dimensional set of genomic features with clinical responses and treatment histories of historical patients, providing a probabilistic basis for using the clinical and molecular information to personalize therapy for future patients. Our work represents one of the first attempts to define personalized treatment assignment rules based on large-scale genomic data. We use actual gene expression data acquired from The Cancer Genome Atlas in the settings of leukemia and glioma to explore the statistical properties of our proposed Bayesian approach for personalizing treatment selection. The method is shown to yield considerable improvements in predictive accuracy when compared to penalized regression approaches.

Keywords: Bayesian analysis; Genomics; Partial exchangeability; Personalized medicine; Predictive probability; Unsupervised clustering.

© 2015, The International Biometric Society.

Publication types

• Research Support, N.I.H., Extramural
• Bayes Theorem*
• Biometry / methods
• Computer Simulation
• Data Interpretation, Statistical
• Diagnosis, Computer-Assisted
• Gene Expression Profiling
• Glioma / genetics
• Leukemia / genetics
• Models, Statistical*
• Precision Medicine / methods*
• Therapy, Computer-Assisted / statistics & numerical data*

Grants and funding

• P30 CA016672/CA/NCI NIH HHS/United States

Unlock Your Potential with Expert Guidance

Student Assignment Support

Precision medicine: revolutionizing healthcare through personalized treatment.

Published by

Archismita Roy

In recent years, the field of precision medicine has emerged as a beacon of hope in healthcare, promising to revolutionize treatment approaches by tailoring interventions to individual patients based on their unique genetic makeup, lifestyle, and environmental factors. This rapidly evolving field holds immense promise for improving patient outcomes, enhancing treatment efficacy, and reducing adverse effects. In this essay, we explore the ethical considerations, technological advancements, and potential impact of precision medicine on specific diseases or treatment approaches.

Understanding Precision Medicine

Precision medicine, also known as personalized or individualized medicine, represents a paradigm shift in healthcare. Unlike traditional one-size-fits-all approaches, precision medicine recognizes that each patient is unique and may respond differently to treatments based on genetic variations, molecular profiles, and other personalized factors.

At the heart of precision medicine lies the integration of advanced technologies, such as genomics, proteomics, and bioinformatics, to analyze vast amounts of data and derive actionable insights. By leveraging these tools, healthcare providers can tailor interventions to target specific disease mechanisms and optimize treatment outcomes.

Ethical Considerations in Precision Medicine

While precision medicine offers tremendous potential for improving patient care, it also raises important ethical considerations. One of the key concerns revolves around data privacy and security. As precision medicine relies heavily on genomic and health data, ensuring the confidentiality and protection of patient information is paramount. Healthcare providers must implement robust safeguards to safeguard patient privacy and prevent unauthorized access or misuse of sensitive data.

Moreover, the equitable access to precision medicine technologies and therapies is another ethical consideration. Disparities in access to healthcare services and genetic testing may exacerbate existing inequalities, leading to differential treatment outcomes among different populations. Healthcare policymakers and stakeholders must work collaboratively to address these disparities and ensure that all patients have access to the benefits of precision medicine, regardless of their socioeconomic status or geographic location.

Technological Advancements Driving Precision Medicine

Advancements in technology have played a pivotal role in driving the rapid progress of precision medicine. The advent of high-throughput sequencing technologies, such as next-generation sequencing (NGS), has revolutionized genomic analysis, enabling researchers to sequence entire genomes or specific gene regions with unprecedented speed and accuracy.

Furthermore, advancements in computational biology and artificial intelligence (AI) have facilitated the analysis of large-scale genomic and clinical datasets, uncovering hidden patterns and associations that may inform treatment decisions. Machine learning algorithms can identify biomarkers, predict treatment responses, and stratify patients into subgroups based on their molecular profiles, guiding personalized treatment approaches.

Impact of Precision Medicine on Disease Management

Precision medicine has the potential to transform the management of various diseases across multiple therapeutic areas. For example:

• Oncology : In cancer treatment, precision medicine enables oncologists to identify specific genetic mutations driving tumor growth and tailor therapies to target these mutations. Precision oncology approaches, such as targeted therapies and immunotherapies, have shown remarkable efficacy in improving survival rates and reducing side effects in patients with various cancer types.
• Cardiology : In cardiology, precision medicine aims to identify genetic predispositions to cardiovascular diseases and tailor interventions to mitigate individual risk factors. Genetic testing can identify patients with hereditary cardiovascular conditions, such as familial hypercholesterolemia, allowing for early intervention and personalized treatment strategies.
• Neurology : In neurology, precision medicine holds promise for the treatment of neurodegenerative disorders, such as Alzheimer’s disease and Parkinson’s disease. By elucidating the genetic and molecular mechanisms underlying these conditions, researchers can develop targeted therapies to slow disease progression and improve cognitive function in affected individuals.

Future Directions and Challenges

While the potential of precision medicine is vast, several challenges lie ahead in realizing its full impact. One of the primary challenges is the interpretation and integration of complex genomic and clinical data into clinical practice. Healthcare providers must develop standardized protocols and guidelines for interpreting genetic test results and translating them into actionable treatment recommendations.

Furthermore, the cost-effectiveness and reimbursement of precision medicine therapies remain a concern. Many precision medicine interventions, such as genetic testing and targeted therapies, are associated with high costs, limiting their accessibility to patients. Healthcare payers and policymakers must address these cost barriers and establish reimbursement mechanisms to ensure equitable access to precision medicine technologies.

Despite these challenges, the future of precision medicine appears promising. With ongoing advancements in technology, collaboration among stakeholders, and a commitment to ethical principles, precision medicine has the potential to revolutionize healthcare delivery, improve patient outcomes, and usher in a new era of personalized medicine.

Precision medicine represents a transformative approach to healthcare, offering personalized treatment strategies tailored to individual patients’ unique characteristics. By leveraging advanced technologies, such as genomics and AI, precision medicine holds promise for improving treatment efficacy, reducing adverse effects, and ultimately enhancing patient outcomes across various disease areas. However, addressing ethical considerations, overcoming technological challenges, and ensuring equitable access to precision medicine therapies are critical steps in realizing its full potential. With continued innovation and collaboration, precision medicine has the power to revolutionize healthcare delivery and usher in a new era of personalized medicine.

Share this:

Leave a comment cancel reply, dear students,.

We are thrilled to extend a warm welcome to you at Student Assignment Support. As you embark on your academic journey, we understand the challenges that come with managing coursework, deadlines, and the pressure to excel in your studies. That’s why we’re here to offer you a helping hand.

Let’s connect

Join the fun.

Stay updated with our latest tutorials and ideas by joining our newsletter.

Type your email…

Recent posts

Maximizing Efficiency: The Role of Citation Generators in Referencing

AI Detector: Revolutionizing Proofreading Services

Mastering Android Studio: A Comprehensive Tutorial

Exploring the World of BSc Nursing: A Comprehensive Guide

Understanding Influenza Disease: A Comprehensive Overview

Unlocking Economic Potential: The Importance of Financial Inclusion

• Business Plan
• Dissertation
• Finance audit
• Medical Project
• Nursing Project
• Online Help
• Proofreading
• Referencing
• Uncategorized

• Already have a WordPress.com account? Log in now.
• Subscribe Subscribed
• Copy shortlink
• Report this content
• View post in Reader
• Manage subscriptions
• Collapse this bar

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Elsevier - PMC COVID-19 Collection

Personalized Medicine of Alzheimer’s Disease

Alzheimer’s disease (AD) is a major problem of health and disability, with a relevant economic impact on society (e.g., €177 billion in Europe). Despite important advances in pathogenesis, diagnosis, and treatment, The primary causes of AD remain elusive, accurate biomarkers are not well characterized, and available pharmacological treatments are not cost-effective. As a complex disorder, AD is polygenic and multifactorial: hundreds of defective genes distributed across the human genome may contribute to its pathogenesis (with the participation of diverse environmental factors, cerebrovascular dysfunction, and epigenetic phenomena) and lead to amyloid deposition, neurofibrillary tangle formation, and premature neuronal death. Future perspectives for the global management of AD predict that structural and functional genomics and proteomics may help in the search for reliable biomarkers, and that pharmacogenomics may be an option in optimizing drug development and therapeutics.

27.1. Overview

Since the identification of its pathogenic features by Alois Alzheimer in 1906, more than 90,000 papers have been published on Alzheimer’s disease (AD) to date (2.5 million references on cancer since 1818; 1.6 million on cardiovascular disorders since 1927; and 1.01 million on central nervous system disorders since 1893) [1] . The number of people affected by dementia is becoming a public and socioeconomic concern in many countries all over the world, independent of economic conditions. The growth of the elderly population is a common phenomenon in both developed and developing countries, bringing about future challenges in terms of health policy and disability rates.

In the United States, rates for the leading causes of death are heart disease (200.2 per 100,000), cancer (180.7 per 100,000), and stroke (43.6 per 100,000). AD is the fifth leading cause of death in people older than 65 years of age, representing 71,600 deaths per year. AD affects approximately 5.4 million individuals in the United States and is estimated to affect up to 16 million by 2050 [2] . Disability caused by senility and dementia affects 9.2 per 1000 in the population aged 65–74 years, 33.5 per 1000 in those within the 75–84 range, and 83.4 per 1000 in the population over 85 years [3] , [4] . In low- to middle-income countries, dementia makes the largest contribution to disability, with a median population-attributable prevalence fraction of 25.1%, followed by stroke (11.4%), limb impairment (10.5%), arthritis (9.9%), depression (8.3%), eyesight problems (6.8%), and gastrointestinal impairments (6.5%) [5] .

In Western countries, AD is the most prevalent form of dementia (45–60%), followed by vascular dementia (30–40%), and mixed dementia (10–20%), which in people older than 85 years of age may account for more than 80% of cases.

The different forms of dementia pose several challenges to society and to the scientific community: (1) they represent an epidemiological problem and a socioeconomic, psychological, and family burden; (2) most of them have an obscure/complex pathogenesis; (3) their diagnosis is not easy and lacks specific biomarkers; and (4) their treatment is difficult and inefficient.

In terms of economic burden, approximately 10–20% of direct costs are associated with pharmacological treatment, with a gradual increase that parallels the severity of the disease. A Canadian study [6] shows that the mean total cost to treat patients with very mild AD is $367 per month, compared with $4063 per month for patients with severe or very severe AD. Only 20–30% of patients with dementia respond appropriately to conventional drugs, and the onset of adverse drug reactions imposes the need for other drugs to neutralize side effects, thus multiplying the initial cost of the pharmacological treatment and the health risk for the patients [7] . Wimo et al. [8] studied the economic impact of dementia in Europe in the EU-funded Eurocode project and found that the total cost of dementia in EU27 countries in 2008 was estimated to be €160 billion (€22,000 per dementia patient per year), of which 56% were costs of informal care. The corresponding costs for the whole of Europe were €177 billion. Informal caregiver costs were the largest cost component, accounting for about half to just over 60% of total societal costs, depending on the country and AD severity [9] .

In addition (and related) to the problem of direct and indirect costs for the management of dementia, there is an alarming abuse of inappropriate psychotropic drug consumption worldwide. Antipsychotic medications are taken by more than 30% of elderly patients with dementia [10] , and conventional antipsychotics are associated with a higher risk of all-cause mortality among nursing home residents [11] .

Abuse, misuse, self-prescription, and uncontrolled medical prescription of CNS drugs are becoming major problems with unpredictable consequences for brain health. The pharmacological management of dementia is an issue of special concern because of the polymedication required to modulate its symptomatic complexity where cognitive decline, behavioral changes, and psychomotor deterioration coexist. In parallel, a growing body of fresh knowledge is emerging on the pathogenesis of dementia, together with data on the neurogenomics and pharmacogenomics of CNS disorders. The incorporation of this new armamentarium of molecular pathology and genomic medicine into daily medical practice, together with educational programs for the correct use of drugs, must help researchers and clinicians to (1) understand AD pathogenesis; (2) establish an early diagnosis; and (3) optimize therapeutics either as a preventive strategy or as formal symptomatic treatment [7] , [12] .

27.2. Toward a Personalized Medicine for Dementia and Neurodegenerative Disorders

Common features of neurodegenerative disorders include the following:

• • Polygenic/complex disorders in which genetic, epigenetic, and environmental factors are involved
• • Deterioration of higher activities of the CNS
• • Multifactorial dysfunction in several brain circuits
• • Accumulation of toxic proteins in the nervous tissue

For instance, the neuropathological hallmarks of AD (amyloid deposition in senile plaques, neurofibrillary tangle formation, and neuronal loss) are merely the phenotypic expression of a pathogenic process in which different gene clusters and their products are potentially involved [7] , [12] .

A large number of the genes that form the structural architecture of the human genome are expressed in the brain in a time-dependent manner along the lifespan. The cellular complexity of the CNS (10 3 different cell types) and synapses (each of the 10 11 neurons in the brain having around 10 3 –10 4 synapses with a complex multiprotein structure integrated by 10 3 different proteins) requires very powerful technology for gene expression profiling, which is still in its very early stages and is not devoid of technical obstacles and limitations [13] . Transcripts of 16,896 genes have been measured in different CNS regions. Each region possesses its own unique transcriptome fingerprint that is independent of age, gender, and energy intake. Fewer than 10% of genes are affected by age, diet, or gender, with most of these changes occurring between middle and old age. Gender and energy restriction have robust influences on the hippocampal transcriptome of middle-aged animals. Prominent functional groups of age- and energy-sensitive genes are those encoding proteins involved in DNA damage responses, mitochondrial and proteasome functions, cell fate determination, and synaptic vesicle trafficking [14] .

The introduction of novel procedures in an integral genomic medicine protocol for CNS disorders and dementia is imperative in drug development and in clinical practice in order to improve diagnostic accuracy and to optimize therapeutics. Personalized strategies, adapted to the complexity of each case, are essential to depict a clinical profile based on specific biomarkers correlating with individual genomic profiles [7] , [15] .

Our understanding of the pathophysiology of CNS disorders and dementia has advanced dramatically during the last 30 years, especially in terms of their molecular pathogenesis and genetics. The drug treatment of CNS disorders has also made remarkable strides with the introduction of many new drugs for the treatment of schizophrenia, depression, anxiety, epilepsy, Parkinson’s disease, and AD, among many other quantitatively and qualitatively important neuropsychiatric disorders.

Improvement in terms of clinical outcome, however, has fallen short of expectations, with up to one-third of patients continuing to experience clinical relapse or unacceptable medication-related side effects in spite of efforts to identify optimal treatment regimes with one or more drugs. Potential reasons for this historical setback might be: (1) that the molecular pathology of most CNS disorders is still poorly understood; (2) that drug targets are inappropriate, not fitting into the real etiology of the disease; (3) that most treatments are symptomatic but not antipathogenic; (4) that the genetic component of most CNS disorders is poorly defined; and (5) that the understanding of genome–drug interactions is very limited [7] , [12] .

The optimization of CNS therapeutics requires the establishment of new postulates regarding (1) the costs of medicines, (2) the assessment of protocols for multifactorial treatment in chronic disorders, (3) the implementation of novel therapeutics addressing causative factors, and (4) the establishment of pharmacogenomic strategies for drug development [12] . Personalized therapeutics based on individual genomic profiles implies the characterization of five types of gene clusters:

• • Genes associated with disease pathogenesis
• • Genes associated with the mechanism of action of drugs
• • Genes associated with drug metabolism (phase I and II reactions)
• • Genes associated with drug transporters
• • Pleiotropic genes involved in multifaceted cascades and metabolic reactions

27.3. Genomics of Alzheimer’s Disease

More than 3000 genes distributed across the human genome have been screened for association with AD during the past 30 years [16] . In the Alzgene database [17] there are 695 genes potentially associated with AD, of which the top ten are (in decreasing order of importance): APOE (19q13.2), BIN1 (2q14), CLU (8p21–p12), ABCA7 (19p13.3), CR1 (1q32), PICALM (11q14), MS4A6A (11q12.1), CD33 (19q13.3), MS4A4E (11q12.2), and CD2AP (6p12). Potentially defective genes associated with AD represent about 1.39% (35,252.69 Kb) of the human genome, which is integrated by 36,505 genes (3,095,677.41 Kb). The highest number of AD-related defective genes concentrate on chromosomes 10 (5.41%; 7337.83 Kb), 21 (4.76%; 2289.15 Kb), 7 (1.62%; 2584.26 Kb), 2 (1.56%; 3799.67 Kb), 19 (1.45%; 854.54 Kb), 9 (1.42%; 2010.62 Kb), 15 (1.23%; 1264.4 Kb), 17 (1.19%; 970.16 Kb), 12 (1.17%; 1559.9 Kb), and 6 (1.15%; 1968.22 Kb), with the highest proportion (related to the total number of genes mapped on a single chromosome) located on chromosome 10 and the lowest on chromosome Y [18] ( Figure 27.1 ).

Distribution of AD-related genes in the human genome.

The genetic and epigenetic defects identified in AD can be classified into four major categories: Mendelian mutations; susceptibility SNP; mtDNA mutations; and epigenetic changes. Mendelian mutations affect genes directly linked to AD, including 32 mutations in the amyloid beta precursor protein ( APP ) gene (21q21)(AD1), 165 mutations in the presenilin 1 ( PSEN1 ) gene (14q24.3)(AD3), and 12 mutations in the presenilin 2 ( PSEN2 ) gene (1q31–q42) (AD4) [16] , [17] , [18] , [19] , [20] . PSEN1 and PSEN2 are important determinants of γ-secretase activity responsible for proteolytic cleavage of APP and NOTCH receptor proteins. Mendelian mutations are very rare in AD (1:1000). Mutations in exons 16 and 17 of the APP gene appear with a frequency of 0.30% and 0.78%, respectively, in AD patients. Likewise, PSEN1 , PSEN2 , and microtubule-associated protein Tau ( MAPT )(17q21.1) mutations are present in less than 2% of cases. Mutations in these genes confer specific phenotypic profiles to patients with dementia: amyloidogeneic pathology associated with APP , PSEN1 , and PSEN2 mutations and tauopathy associated with MAPT mutations represent the two major pathogenic hypotheses for AD [16] , [17] , [18] , [19] , [20] , [21] .

Multiple polymorphic risk variants can increase neuronal vulnerability to premature death (see Appendix A). Among these susceptibility genes, the apolipoprotein E ( APOE ) gene (19q13.2)(AD2) is the most prevalent as a risk factor for AD, especially in those subjects harboring the APOE-4 allele ( Figure 27.2 ), whereas carriers of the APOE-2 allele might be protected against dementia. APOE -related pathogenic mechanisms are also associated with brain aging and with the neuropathological hallmarks of AD [16] .

Distribution and frequency of APOE genotypes in AD and vascular dementia.

27.4. Pathogenic Events

The dual amyloidogenic-tauopathic theory of AD has dominated the pathogenic universe of AD-related neurodegeneration (and divided the research community) for the past 50 years, nourished by the presence of APP , PSEN1 , PSEN2 , and MAPT mutations in a very small number of cases with early-onset AD. Nevertheless, this theory does not explain AD pathogenesis in full, and consequently novel (or complementary) theories have been emerging recently and during the past decades. A summary of the pathogenic events in AD is given in the following sections.

27.4.1. Genomic Defects

As a complex polygenic/multifactorial disorder, in which hundreds of polymorphic variants of risk might be involved (Appendix A, Figure 27.1 ), AD fulfils the “golden rule” of complex disorders, according to which the larger the number of genetic defects distributed in the human genome, the earlier the onset of the disease and the poorer its therapeutic response to conventional treatments; conversely, the smaller the number of pathogenic SNPs, the later the onset of the disease and the better its therapeutic response to different pharmacological interventions [12] , [16] , [22] , [23] , [24] , [25] , [26] , [27] , [28] . Genetic variation associated with different diseases interferes with microRNA-mediated regulation by creating, destroying, or modifying microRNA (miRNA) binding sites. miRNA-target variability is a ubiquitous phenomenon in the adult human brain which may influence gene expression in physiological and pathological conditions. AD-related SNPs interfere with miRNA gene regulation and affect AD susceptibility. Significant interactions include target SNPs present in seven genes related to AD prognosis with the miRNAs miR-214, -23a and -23b, -486-3p, -30e*, -143, -128, -27a and -27b, -324-5p, and -422a. The dysregulated miRNA network contributes to aberrant gene expression in AD [29] , [30] , [31] .

27.4.2. Epigenetic Phenomena

Epigenetic factors have emerged as important mediators of development and aging, gene–gene and gene–environmental interactions, and the pathophysiology of complex disorders. Major epigenetic mechanisms (DNA methylation, histone modifications and chromatin remodeling, and noncoding RNA regulation) may contribute to AD pathology [30] , [31] .

27.4.3. Cerebrovascular Dysfunction

Vascular and metabolic dysfunctions are key components in AD pathology throughout the course of disease. Although common denominators between vascular and metabolic dysfunction are oxidative stress and Aβ [32] , genetic factors and cardiovascular risk factors may also account for the cerebrovascular damage present in AD [33] . Inherited polymorphisms of the vascular susceptibility gene Ninjurin2 ( NINJ2 ) are associated with AD risk [34] . Endothelial dysfunction has been implicated as a crucial event in the development of AD.

Breakdown of the blood–brain barrier (BBB) as a result of disruption of tight junctions and transporters leads to increased leukocyte transmigration and is an early event in the pathology of many CNS disorders. BBB breakdown leads to neuroinflammation and oxidative stress, with mitochondrial dysfunction. The high concentration of mitochondria in cerebrovascular endothelial cells might account for the sensitivity of the BBB to oxidant stressors [35] , [36] .

Chronic brain hypoperfusion may be sufficient to induce premature neuronal death and dementia in vulnerable subjects [16] , [23] , [24] , [25] , [37] , [38] , [39] . APOE-related changes in cortical oxygenation and hemoglobin consumption are evident, as revealed by brain optical topography analysis, and reflect that APOE-4 carriers exhibit deficient brain hemodynamics and a poorer panneocortical oxygenation than do APOE-3 or APOE-2 carriers [18] . Hypoperfusion in frontal, parietal, and temporal regions is a common finding in AD. White matter hyperintensities (WMH) correlate with age and with disease severity [40] .

Cerebral amyloid angiopathy (CAA) accounts for the majority of primary lobal intracerebral hemorrhages (ICH) among the elderly, and represents the cause of 20% of spontaneous ICHs in patients over 70 years of age. The basis for this disease process is the deposition and formation of eventually destructive amyloid plaques in the walls of brain vessels, predominantly arterial but not excluding venules and capillaries. CAA and CAA-associated microhemorrhages may also participate in the pathogenesis of AD [41] . Aβ deposition in asymptomatic elderly individuals is associated with lobar MH (LMH).

LMH is present in 30.8% of AD, 35.7% of MCI, and 19.1% of controls [42] . Neurovascular dysfunction in AD leads to reduced clearance across the BBB and accumulation of neurotoxic Aβ peptides in the brain. The ABC transport protein P-glycoprotein (P-gp, ABCB1) is involved in the export of Aβ from the brain into the blood. P-gp , LRP1 , and RAGE mRNA expression is reduced in mice treated with Aβ 1–42 . In addition to the age-related decrease in P-gp expression, Aβ 1–42 itself downregulates the expression of P-gp and other Aβ transporters, which could exacerbate the intracerebral accumulation of Aβ and thereby accelerate neurodegeneration in AD and cerebral β-amyloid angiopathy [43] .

27.4.4. Phenotypic Expression of Amyloid Deposits and Neurofibrillary Tangles

β-Amyloid deposits in senile and neuritic plaques and hyperphosphorylated tau proteins in neurofibrillary tangles (NFT) are extracellular and intracellular expressions, respectively, of the AD neuropathological phenotype, together with selective neuronal loss in hippocampal and neocortical regions. Aβ plaque in the brain is the primary (postmortem) diagnostic criterion of AD. The main component of senile plaques is Aβ, a 39–43 amino acid peptide, generated by the proteolytic cleavage of amyloid precursor protein (APP) by the action of beta- and gamma-secretases. Aβ is neurotoxic, and this neurotoxicity is related to its aggregation state [16] , [17] , [18] , [19] , [20] , [21] .

27.4.5. Neuronal Apoptosis

Neuronal loss is a pathognomonic finding in AD and the final common path of multiple pathogenic mechanisms leading to neurodegeneration in dementia. Atrophy of the medial temporal lobe, especially the hippocampus and the parahippocampal gyrus, is considered to be AD’s most predictive structural brain biomarker. The medial and posterior parts of the parietal lobe seem to be preferentially affected, compared to the other parietal lobe parts [18] .

27.4.6. Neurotransmitter Deficits

An imbalance of different neurotransmitters (glutamate, acetylcholine, noradrenaline, dopamine, serotonin, and some neuropeptides) has been proposed as the neurobiological basis of behavioral symptoms in AD. Altered reuptake of neurotransmitters by vesicular glutamate transporters (VGLUTs), excitatory amino acid transporters (EAATs), the vesicular acetylcholine transporter (VAChT), the serotonin reuptake transporter (SERT), or the dopamine reuptake transporter (DAT) are involved in the neurotransmission imbalance in AD. Protein and mRNA levels of VGLUTs, EAAT1-3, VAChT, and SERT are reduced in the disease [44] .

27.4.7. Oxidative Stress

Oxidative damage is a classic pathogenic mechanism of neurodegeneration [36] , [45] . It is greater in brain tissue from patients with AD than age-matched controls. Tayler et al. [46] studied the timing of this damage in relation to other pathogenic AD processes. Antioxidant capacity is elevated in AD and directly related to disease severity as indicated by the Braak tangle stage and the amount of insoluble Aβ. Accumulation of Aβ has been shown in brain mitochondria of AD patients and in AD transgenic mouse models. The presence of Aβ in mitochondria leads to free radical generation and neuronal stress.

A novel mitochondrial Aβ-degrading enzyme, presequence protease (Pre), has been identified in the mitochondrial matrix. hPreP activity is decreased in AD human brains and in the mitochondrial matrix of AD transgenic mouse brains (TgmAβPP and TgmAβPP/ABAD). Mitochondrial fractions isolated from AD brains and TgmAβPP mice have higher levels of 4-hydroxynonenal, an oxidative product. Cytochrome c oxidase activity is significantly reduced in the AD mitochondria. Decreased PreP proteolytic activity, possibly due to enhanced ROS production, may contribute to Aβ accumulation in mitochondria, leading to mitochondrial toxicity and neuronal death in AD [47] .

27.4.8. Cholesterol and Lipid Metabolism Dysfunction

Cholesterol seems to be intimately linked with the generation of amyloid plaques, which are central to AD pathogenesis. APOE variants are determinants in cholesterol metabolism and diverse forms of dyslipoproteinemia [12] , [48] . Cholesterol protects the Aβ-induced neuronal membrane disruption and inhibits beta-sheet formation of Aβ on the lipid bilayer [49] . Jones et al. [50] found a significant over-representation of association signals in pathways related to cholesterol metabolism and the immune response in both of the two largest genome-wide association studies for LOAD.

27.4.9. Neuroinflammation and Immunopathology

Several genes associated with immune regulation and inflammation show polymorphic variants of risk in AD, and abnormal levels of diverse cytokins have been reported in the brain, CSF, and plasma of AD patients [16] , [23] . The activation of inflammatory cascades has been consistently demonstrated in AD pathophysiology, in which reactive microglia are associated with Aβ deposits and clearance. Resident microglia fail to trigger an effective phagocytic response to clear Aβ deposits, although they mainly exist in an “activated” state. Oligomeric Aβ (oAβ) can induce more potent neurotoxicity when compared with fibrillar Aβ (fAβ). Aβ (1–42) fibrils, not Aβ (1–42) oligomers, increase microglial phagocytosis [51] . Among several putative neuroinflammatory mechanisms, the TNF-α signaling system has a central role in this process. In AD, TNF-α levels are altered in serum and CSF. The abnormal production of inflammatory factors may accompany the progression from mild cognitive impairment (MCI) to dementia. Abnormal activation of the TNF-α signaling system, represented by increased expression of sTNFR1, is associated with a higher risk of progression from MCI to AD [52] .

27.4.10. Neurotoxic Factors

Old and new theories suggest that different toxic agents, from metals (e.g., aluminium, copper, zinc, iron) to biotoxins and pesticides, might contribute to neurodegeneration. Dysfunctional homeostasis of transition metals is believed to play a role in AD pathogenesis [18] .

27.4.11. Other Players

Many novel pathogenic mechanisms potentially involved in AD neurodegeneration have been proposed in recent times. Moreover, there has been a revival of some old hypotheses. Examples of pathogenic players in AD, other than those just discussed, include the Ca 2+ hypothesis, insulin resistance, NGF imbalance, glycogen synthase kinase-3 (GSK-3), advanced glycation end products (AGEs) and their receptors (RAGE), the efflux transporter P-glycoprotein (P-gp), c-Abl tyrosine kinase, post-transcriptional protein alterations that compromise the proteasome system and the chaperone machinery (HSPB8–BAG3), autophagy as a novel Aβ-generating pathway, hypocretin (orexin), cathepsin B, Nogo receptor proteins, adipocytokines and CD34 + progenitor cells, CD147, impairment of synaptic plasticity (PSD-95), anomalies in neuronal cell division and apoptosis, stem cell factor (SCF), telomere shortening, deficiency in repair of nuclear and mitochondrial DNA damage, and microDNAs [18] .

27.5. Biomarkers and Comorbidity

AD’s phenotypic features represent the biomarkers to be used as diagnostic predictors and the expression of pathogenic events to be modified with an effective therapeutic intervention. Important differences have been found in the AD population (as compared with healthy subjects) in different biological parameters, including blood pressure, glucose, cholesterol and triglyceride levels, transaminase activity, hematological parameters, metabolic factors, thyroid function, brain hemodynamic parameters, and brain mapping activity [7] , [23] , [24] , [25] , [53] , [54] , [55] , [56] , [57] , [58] , [59] .

These clinical differences are clear signs of comorbidity rather than typical features of AD. Blood pressure values, glucose levels, and cholesterol levels are higher in AD than in healthy elderly subjects. Approximately 20% of AD patients are hypertensive, 25% are diabetics, 50% are hypercholesterolemic, and 23% are hypertriglyceridemic. More than 25% of patients exhibit high GGT activity, 5–10% show anemic conditions, 30–50% show an abnormal cerebrovascular function characterized by poor brain perfusion, and more than 60% have an abnormal electroencephalographic pattern, especially in frontal, temporal, and parietal regions, as revealed by quantitative EEG (qEEG) or computerized mapping [7] , [12] , [23] , [54] . Significant differences are currently seen between females and males, indicating the effect of gender on the phenotypic expression of the disease. In fact, the prevalence of dementia is 10–15% higher in females than in males from 65–85 years of age. All of these parameters are highly relevant when treating AD patients, because some of them reflect a concomitant pathology that also needs therapeutic consideration.

AD biomarkers can be differentiated into several categories: (1) neuropathological markers; (2) structural and functional neuroimaging markers; (3) neurophysiological markers (EEG, qEEG, brain mapping); (4) biochemical markers in body fluids (e.g., blood, urine, saliva, CSF); and (5) genomic markers (structural and functional genomics, proteomics, metabolomics).

27.5.1. Neuropathology

Plaques and tangles in the hippocampus and cortex are still considered the seminal findings in AD neuropathology and are the conventional means of establishing the boundary between amyloidopathies and tauopathies; however, both phenotypic markers are also present in normal brains, in more than 60% of cases with traumatic brain injury, and in many other brain disorders [60] .

27.5.2. Structural and Functional Neuroimaging

Structural and functional neuroimaging techniques (MRI, fMRI, PET, SPECT) are essential tools in the diagnosis of dementia, although the specificity of visual observations in degenerative forms of dementia is of doubtful value. Nevertheless, these procedures are irreplaceable for a differential diagnosis. There is a characteristic regional impairment in AD that involves mainly the temporo–parietal association cortices, the mesial temporal structures, and, to a more variable degree, the frontal association cortex. This pattern of functional impairment can provide a biomarker for diagnosis of AD and other neurodegenerative dementias at the clinical stage of mild cognitive impairment, and for monitoring its progression. Healthy young APOE ɛ4 carriers have smaller hippocampal volumes than APOE ɛ2 carriers.

The difference in hippocampal morphology is cognitively/clinically silent in young adulthood, but can render APOE ɛ4 carriers more prone to the later development of AD, possibly because of lower reserve cognitive capacity [61] . LOAD patients exhibit a selective parahippocampal white matter (WM) loss, while EOAD patients experience a more widespread pattern of posterior WM atrophy. The distinct regional distribution of WM atrophy reflects the topography of gray matter (GM) loss. ApoE ɛ4 status is associated with a greater parahippocampal WM loss in AD. The greater WM atrophy in EOAD than in LOAD fits with the evidence that EOAD is a more aggressive form of the disease [62] . FDG-PET is quantitatively more accurate than perfusion SPECT.

Regional metabolic and blood flow changes are closely related to clinical symptoms, and most areas involved in these changes also develop significant cortical atrophy. FDG-PET is complementary to amyloid PET, which targets a molecular marker that does not have a close relation to current symptoms. FDG-PET is expected to play an increasing role in diagnosing patients at an early stage of AD and in clinical trials of drugs aimed at preventing or delaying the onset of dementia [63] . Functional neuroimaging biomarkers are becoming popular, with the introduction of novel tracers for brain amyloid deposits. Amyloid deposition causes severe damage to neurons many years before onset of dementia via a cascade of several downstream effects.

Positron emission tomography (PET) tracers for amyloid plaque are desirable for early diagnosis of AD, particularly to enable preventative treatment once effective therapeutics is available. The amyloid imaging tracers flutemetamol, florbetapir, and florbetaben labeled with 18 F have been developed for PET. These tracers are currently undergoing formal clinical trials to establish whether they can be used to accurately image fibrillary amyloid, and to distinguish patients with AD from normal controls and those with other diseases that cause dementia [63] .

27.5.3. Neurophysiology

There is a renewed interest in the use of computerized brain mapping as a diagnostic aid and as a monitoring tool in AD [64] . Electroencephalography (EEG) studies in AD show an attenuation of average power within the alpha band (7.5–13 Hz) and an increase in power in the theta band (4–7 Hz) [65] . APOE genotypes influence brain bioelectrical activity in AD. In general, APOE-4 carriers tend to exhibit a slower EEG pattern from early stages [16] , [18] , [66] .

27.5.4. Biochemistry of Body Fluids

Other biomarkers of potential interest include cerebrospinal fluid (CSF) and peripheral levels of Aβ 42 , protein tau, histamine, interleukins, and some other novel candidate markers such as chitinase 3-like 1 (CHI3L1) protein [7] , [16] , [25] , [67] , [68] , [69] . The concentration of the 42-amino-acid form of Aβ (Aβ 1–42 ) is reduced in the CSF of AD patients, which is believed to reflect the AD pathology, with plaques in the brain acting as sinks. Novel C-truncated forms of Aβ (Aβ 1–14 , Aβ 1–15 , and Aβ 1–16 ) were identified in human CSF. The presence of these small peptides is consistent with a catabolic amyloid precursor protein cleavage pathway by β- followed by α-secretase. Aβ 1–14 , Aβ 1–15 , and Aβ 1–16 increase dose-dependently in response to γ-secretase inhibitor treatment, while Aβ 1–42 levels are unchanged [70] .

Kester et al. [71] investigated change over time in CSF levels of amyloid-beta 40 and 42 (Aβ 40 and Aβ 42 ), total tau (tau), tau phosphorylated at threonine 181 (ptau-181), isoprostane, neurofilaments heavy (NfH) and neurofilaments light (NfL). Aβ 42 , tau, and tau phosphorylated at threonine 181 differentiated between diagnosis groups, whereas isoprostane, NfH, and NfL did not. In contrast, effects of follow-up time were found only for nonspecific CSF biomarkers: levels of NfL decreased, and levels of isoprostane, Aβ 40 , and tau increased over time. An increase in isoprostane was associated with progression of mild cognitive impairment in AD and with cognitive decline. Contrary to AD-specific markers, nonspecific CSF biomarkers show change over time, which potentially can be used to monitor disease progression in AD.

27.5.5. Genomics and Proteomics

Structural markers are represented by SNPs in genes associated with AD, polygenic cluster analysis, and genome-wide studies (GWSs). Functional markers attempt to correlate genetic defects with specific phenotypes (genotype–phenotype correlations). In proteomic studies, several candidate CSF protein biomarkers have been assessed in neuropathologically confirmed AD, nondemented (ND) elderly controls, and non-AD dementias (NADD). Markers selected included apolipoprotein A-1 (ApoA1), hemopexin (HPX), transthyretin (TTR), pigment epithelium-derived factor (PEDF), Aβ 1–40 , Aβ 1–42 , total tau, phosphorylated tau, α-1 acid glycoprotein (A1GP), haptoglobin, zinc α-2 glycoprotein (Z2GP), and apolipoprotein E (ApoE). Concentrations of Aβ 1–42 , ApoA1, A1GP, ApoE, HPX, and Z2GP differed significantly among AD, ND, and NADD subjects. The CSF concentrations of these three markers distinguished AD from ND subjects with 84% sensitivity and 72% specificity, with 78% of subjects correctly classified.

By comparison, Aβ 1–42 alone gave 79% sensitivity and 61% specificity, with 68% of subjects correctly classified. For the diagnostic discrimination of AD from NADD, only the concentration of Aβ 1–42 was significantly related to diagnosis, with a sensitivity of 58% and a specificity of 86% [72] . Carrying the APOE-ɛ4 allele was associated with a significant decrease in CSF Aβ 1–42 concentrations in middle-aged and older subjects. In AD, Aβ 1–42 levels are significantly lower in APOEɛ4 carriers compared to noncarriers. These findings demonstrate significant age effects on CSF Aβ 1–42 and pTau181 across the lifespan, and also suggest that a decrease in Aβ 1–42 , but an increase in pTau181 CSF levels, is accelerated by the APOEɛ4 genotype in middle-aged and older adults with normal cognition [73] .

Han et al. [74] carried out a GWAS to better define the genetic backgrounds of normal cognition, mild cognitive impairment (MCI), and AD in terms of changes in CSF levels of Aβ 1–42 , T-tau, and P-tau181P. CSF Aβ 1–42 levels decreased with APOE gene dose for each subject group. T-tau levels tended to be higher among AD cases than among normal subjects. CYP19A1 “aromatase” (rs2899472), NCAM2 , and multiple SNPs located on chromosome 10 near the ARL5B gene demonstrated the strongest associations with Aβ 1–42 in normal subjects.

Two genes found to be near the top SNPs, CYP19A1 (rs2899472) and NCAM2 (rs1022442), have been reported as genetic factors related to the progression of AD. In AD subjects, APOE ɛ2/ɛ3 and ɛ2/ɛ4 genotypes were associated with elevated T-tau levels, and the ɛ4/ɛ4 genotype was associated with elevated T-tau and P-tau181P levels. Blood-based markers reflecting core pathological features of AD in presymptomatic individuals are likely to accelerate the development of disease-modifying treatments.

Thambisetty et al. [75] performed a proteomic analysis to discover plasma proteins associated with brain Aβ burden in nondemented older individuals. A panel of 18 2DGE plasma protein spots effectively discriminated between individuals with high and low brain Aβ. Mass spectrometry identified these proteins, many of which have established roles in Aβ clearance, including a strong signal from ApoE. A strong association was observed between plasma ApoE concentration, and Aβ burden in the medial temporal lobe. Targeted voxel-based analysis localized this association to the hippocampus and entorhinal cortex. APOE ɛ4 carriers also showed greater Aβ levels in several brain regions relative to ɛ4 noncarriers. Both peripheral concentration of the ApoE protein and the APOE genotype may be related to early neuropathological changes in brain regions vulnerable to AD pathology even in the nondemented elderly.

27.6. Therapeutic Strategies

Modern therapeutic strategies in AD are aimed at interfering with the main pathogenic mechanisms potentially involved in AD [7] , [12] , [16] , [18] , [23] , [24] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] ( Box 27.1 ). Starting in the early 1990s, the neuropharmacology of AD was dominated by acetylcholinesterase inhibitors, represented by tacrine, donepezil, rivastigmine, and galantamine [76] , [77] , [78] . Memantine, a partial NMDA antagonist, was introduced in the 2000s for the treatment of severe dementia [79] ; and the first clinical trials with immunotherapy, to reduce amyloid burden in senile plaques, were withdrawn due to severe ADRs [80] , [81] . After the initial promise of β- and γ-secretase inhibitors [82] , [83] and novel vaccines [84] , [85] devoid of severe side effects, during the past few years no relevant drug candidates have dazzled the scientific community with their capacity to halt disease progression; however, a large number of novel therapeutic strategies for the pharmacological treatment of AD have been postulated, with some apparent effects in preclinical studies (see Box 27.1 ).

Experimental Strategies for the Pharmacological Treatment of Alzheimer’s Disease

New cholinesterase inhibitors

Cholinergic receptor agonists

Monoamine regulators

Diverse natural compounds derived from vegetal sources:

• Alkaloids from the calabar bean ( Physostigma venenosum )
• Huperzine A from Huperzia serrata
• Galantamine from the snowdrop Galanthus woronowii
• Cannabinoids (cannabidiol) from Cannabis sativa
• Saffron ( Crocus sativus )
• Ginseng ( Panax species )
• Sage ( Salvia species )
• Lemon balm ( Melissa officinalis )
• Polygala tenuifolia
• Nicotine from Nicotiana species
• Grape seed polyphenolic extracts
• Fuzhisan, a Chinese herbal medicine
• Resveratrol
• Xanthoceraside
• Garlic ( Allium sativum )
• Linarin from Mentha arvensis and Buddleja davidii
• Carotenoids (e.g., retinoic acid, all-trans retinoic acid, lycopene and β-carotene)
• Curcumin from the rhizome of Curcuma longa
• Decursinol from the roots of Angelica gigas
• Bacopa monniera LINN (Syn. Brahmi)
• Phytoestrogens
• Walnut extract
• Erigeron annuus leaf extracts
• Epigallocatechin-3-gallate
• The brown algae ( Ecklonia cava )
• Gami-Chunghyuldan (standardized multiherbal medicinal formula)
• Punica granatum extracts

Plants of different origin:

• Yizhi Jiannao
• Drumstick tree ( Moringa oleifera )
• Ginkgo/Maidenhair tree ( Ginkgo biloba )
• Sicklepod ( Cassia obtisufolia )
• Sal Leaved Desmodium ( Desmodium gangeticum )
• Lemon Balm ( Melissa officinalis )
• Garden sage, common sage ( Salvia officinalis )

Immunotherapy and treatment options for tauopathies:

• Tau kinase inhibitors
• 2-Aminothiazoles
• Phosphoprotein phosphatase 2A (PP2A) inhibitors
• c-Jun N-terminal kinase (JNKs) inhibitors
• p38 MAP kinase inhibitors (CNI-1493)
• Harmine (β-carboline alkaloid)

Immunotherapy and Aβ breakers for AD-related amyloidopathy:

• Active and passive immunization

Secretase inhibitors (β- and γ-)

Neurosteroids

Phosphodiesterase inhibitors

Protein phosphatase methylesterase-1 inhibitors

Histone deacetylase inhibitors

mTOR inhibitors

Peroxisome proliferator-activated receptor agonists

P-glycoprotein regulators

Nuclear receptor agonists

Glycogen synthase kinase-3β (GSK-3β) regulators

Histamine H3 receptor inverse agonists

Kynurenine 3-monooxygenase inhibitors

Chaperones (small heat shock proteins (sHSPs); Hsp90 inhibitors and HSP inducers)

microRNAs (miRNAs) and gene silencing (RNA interference)(RNAi)

Miscellaneous strategies:

• Sodium fullerenolate
• Glucagon-like peptide -1 (GLP-1)
• Macrophage inflammatory protein-2 (MIP-2)
• Stromal cell-derived factor-1α (SDF-1α)
• Cyclooxygenase-1 and cyclooxygenase-2 inhibitors
• Bone morphogenetic protein 9 (BMP-9)
• Granulocyte colony stimulating factor (G-CSF)/AMD3100 (CXCR4 antagonist)
• Vitamins (A, B, C, D)
• ω-3 Polyunsaturated fatty acids (n-3 PUFAs)
• Docosahexaenoic acid (DHA, C22:6 n-3)
• Sphingosylphosphorylcholine
• Citidine-5-diphosphocholine (CDP-choline)
• Cathepsin B inhibitors
• Pituitary adenylate cyclase–activating polypeptide
• NAP (Davunetide)
• Transcription factor specificity protein 1 (Sp1) inhibitors (tolfenamic acid)
• 2-(2,6-Dioxopiperidin-3-yl)phthalimidine EM-12 dithiocarbamates
• N-substituted 3-(Phthalimidinp-2-yl)-2,6-dioxopiperidines
• 3-substituted 2,6-Dioxopiperidines
• Pyrrolo[3,2-e][1,2,4]triazolo[1,5-a]pyrimidine (SEN1176)
• Latrepirdine
• Leucettines
• Dihydropyridines (inhibitors of L-type calcium channels)
• Brain-penetrating angiotensin-converting enzyme (ACE) inhibitors
• NADPH oxidase inhibitors (Apocynin)
• Heterocyclic indazole derivatives (inhibitors of serum- and glucocorticoid-inducible-kinase 1 [ SGK1 ])
• IgG-single-chain Fv fusion proteins

27.6.1. Immunotherapy

There are two main modalities of immunotherapy for AD: (1) passive immunotherapy, with the administration of monoclonal Aβ-specific antibodies [86] ; and (2) active immunization with the Aβ 42 antigen [87] , [88] or Aβ-conjugated synthetic fragments bound to a carrier protein, thus avoiding potential problems associated with mounting a T-cell response directly against Aβ [89] . A new approach—delivering Aβ 42 in a novel immunogen-adjuvant manner consisting of sphingosine-1-phosphate (S1P)-containing liposomes, administered to APP/PS1 transgenic mice before and after the detection of AD-like pathology in the brain—has recently been developed [85] .

The results from this novel vaccine (EB101) indicate that active immunization significantly prevents and reverses the progression of AD-like pathology and also clears prototypical neuropathological hallmarks in transgenic mice. This new approach strongly induces T-cell, B-cell, and microglial immune response activation, avoiding the Th1 inflammatory reaction [90] .

The rationale for amyloid immunotherapy in AD [91] is based on the following assumptions:

• • β-amyloid plaques and their aggregated, proto-fibrillar, and oligomeric precursors contain immunologic neo-epitopes that are absent from the full-length amyloid precursor protein (APP), as well as from its soluble proteolytic derivatives restricted to brain tissue; consequently, β-amyloid-based immunotherapies designed to selectively target pathologic neo-epitopes present on Aβ oligomers, protofibrils, or fibrils should not cause autoimmune disease in unaffected tissues throughout the organism
• • β-amyloid buildup precedes neurodegeneration and functional loss, and either the prevention of its formation or its removal can be expected to result in the slowing or the prevention of neurodegeneration
• • β-amyloid can cause the formation of neurofibrillary tangles in vivo and in vitro. The removal of β-amyloid, or the prevention of its buildup, has the potential not only to correct β-amyloid-related toxicity, but also to prevent the formation of neurofibrillary tangles
• • Conformational changes of endogenously occurring proteins and the formation of insoluble aggregates are commonly associated with neurodegeneration and brain disease, so the removal or prevention of these pathologic protein aggregates is also a therapeutic goal in the principle of immunotherapy
• • Immunotherapy works in experimental animals and in initial clinical trials: both active immunization and passive antibody transfer consistently reduce brain β-amyloid load, improve β-amyloid-related memory impairments, and protect neurons against degeneration in many independent experiments using different mouse models and primates [90]

Since Aβ immunotherapy has a limited clearance effect of tau aggregates in dystrophic neurites, the development of an alternative therapy that directly targets pathological tau has become crucial. Increased levels of tau oligomers have been observed in the early stage of AD, prior to the detection of neurofibrillary tangles (NFT) formed by aggregation and accumulation of the microtubule-associated protein tau [92] . Several approaches have been taken to treat AD by targeting tau, such as the following:

• 1. The inhibition of tau hyperphosphorylation, by a kinase inhibitor of soluble aggregated tau formation, which also prevents related motor deficits [93] .
• 2. Activation of the proteolytic pathway, by the degrading action of calpain [94] and puromycin-sensitive aminopeptidase [95] .
• 3. The stabilization of microtubules, treating tauopathies by functionally binding and stabilizing microtubules with mt-binding protein tau [96] and paclitaxel, a drug proven effective in restoring affected axonal transport and motor impairments [97] .
• 4. Tau clearance by immunotherapy in this case, the tau active vaccination uses phosphorylated antigens of tau fragments associated with neurofibrillary tangles [98] that results in an efficient reduction of both soluble and insoluble tau active fragments, reducing phosphorylated NFTs in AD-like mouse brains.

Preclinical studies have shown clear evidence that Aβ immunization therapy provides protection and reverses the pathological effects of AD in transgenic mouse models [99] . This strategy seems to improve cognition performance [100] after Aβ 42 immunization, in addition to causing an effective reduction in Aβ pathology. A recent immunization study has proven that a fragment of the Aβ peptide bound to polylysines activates the immune response that diminishes AD-like pathology in APP transgenic mice. This result reinforces the notion that the immune-conjugate approach is an effective means of Aβ immunotherapy, and also that the entire Aβ peptide is not necessary for its efficacy. It is in accordance with the hypothesis that specific antibodies directed against the amino-terminal and/or central region of the amyloid peptide provide beneficial protection against amyloid pathology. Passive immunization studies have also been conducted with promising experimental results, showing that a humoral response alone, without Aβ cellular response, is sufficient to reduce the β-amyloid burden and reverse memory deficits [101] .

Among the drugs and vaccines currently under development to treat the pathological effects of AD, the most promising are bapineuzumab, solanezumab, CAD106, and EB101. Solanezumab is a monoclonal antibody raised against Aβ 13–28 that recognizes an epitope in the core of the amyloid peptide, binding selectively to soluble Aβ and with low affinity for the fibrillar Aβ form [102] . Thus, it presents fewer adverse events than does bapineuzumab, which binds to Aβ amyloid plaques more strongly than soluble Aβ [103] . There are a few other monoclonal antibodies against Aβ that have properties different from those of bapineuzumab, such as PF-04360365, which specifically targets the free carboxy-terminus of Aβ 1–40 , MABT5102A, which binds with equally high affinity to Aβ monomers, oligomers, and fibrils, and GSK933776A, which targets the N-terminus of Aβ.

Specific anti-Aβ antibodies are present in pooled preparations of intravenous immunoglobulin (IVIg or IGIV), which has already been approved by the FDA for the treatment of a variety of neurological conditions. Current results from these studies have shown that IVIg treatment may also be an efficacious alternative approach in the treatment of AD neuropathologies [90] , [104] .

Avoiding both the strong Th1 effects of the QS-21 adjuvant and the T-cell epitopes at the C-terminus of Aβ, CAD106 consists of a short N-terminal fragment of Aβ attached to a virus-like particle, with no additional adjuvant [105] . This therapeutic agent is currently in phase II trials. Affiris is testing two short 6-amino-peptides (AD01, AD02), administered with aluminum hydroxide as adjuvant, that mimic the free N-terminus of Aβ and therefore cause cross-reactivity with the native peptide in phase I trials [106] . In terms of prevention and therapeutic treatment, the EB101 vaccine showed for the first time the effectiveness of combining a liposomal immunogen-adjuvant with an Aβ antigen to induce an effective immunological response combined with an anti-inflammatory effect in preclinical studies using APP/PS1 transgenic mice [85] , [90] .

The EB101 vaccine immunization process has shown a marked positive effect as a preventive and therapeutic treatment, reducing amyloidosis-induced inflammation as an effective Th2 immunomodulator. Moreover, this vaccine proved to stimulate innate immunity and enable effective phagocytosis to clear amyloid and neurofibrillary tangles, which are among the major hallmarks of AD-like neuropathology observed. A few other vaccines are currently under development, and recent studies have opened up new perspectives in the immunization approach to AD pathology; in particular, gene-gun-mediated genetic immunization with the Aβ 42 gene [107] shows that self-tolerance can be broken in order to produce a humoral response to the Aβ 42 peptide with minimal cellular response.

27.7. Pharmacogenomics

AD patients may take 6–12 different drugs per day for the treatment of dementia-related symptoms, including memory decline (conventional antidementia drugs, neuroprotectants), behavioral changes (antidepressants, neuroleptics, sedatives, hypnotics), and functional decline. Such drugs may also be taken for the treatment of concomitant pathologies (epilepsy, cardiovascular and cerebrovascular disorders, parkinsonism, hypertension, dyslipidemia, anemia, arthrosis, etc). The co-administration of several drugs may cause side effects and ADRs in more than 60% of AD patients, who in 2–10% of cases require hospitalization. In more than 20% of patients, behavioral deterioration and psychomotor function can be severely altered by polypharmacy. The principal causes of these iatrogenic effects are (1) the inappropriate combination of drugs, and (2) the genomic background of the patient, which is responsible for his/her pharmacogenomic outcome.

Pharmacogenomics account for 30–90% of the variability in pharmacokinetics and pharmacodynamics. The genes involved in the pharmacogenomic response to drugs in AD fall into five major categories:

• • Genes associated with AD pathogenesis and neurodegeneration ( APP , PSEN1 , PSEN2 , MAPT , PRNP , APOE , and others)
• • Genes associated with the mechanism of action of drugs (enzymes, receptors, transmitters, messengers)
• • Genes associated with drug metabolism (phase I ( CYP s) and phase II reactions ( UGT s, NAT s))
• • Genes associated with drug transporters ( ABC s, SLC s)
• • Pleiotropic genes involved in multifaceted cascades and metabolic reactions ( APOs , ILs , MTHFR , ACE , AGT , NOS , etc) [18] ( Figure 27.1 )

27.7.1. Pathogenic Genes

In more than 100 clinical trials for dementia, APOE has been used as the only gene of reference for the pharmacogenomics of AD [7] , [12] , [15] , [16] , [22] , [23] , [24] , [25] , [26] , [27] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] . Several studies indicate that the presence of the APOE-4 allele differentially affects the quality and extent of drug responsiveness in AD patients treated with cholinergic enhancers (tacrine, donepezil, galantamine, rivastigmine), neuroprotective compounds (nootropics), endogenous nucleotides (CDP-choline), immunotrophins (anapsos), neurotrophic factors (cerebrolysin), rosiglitazone, or combination therapies [108] , [109] , [110] ; however, controversial results are frequently found that are due to methodological problems, study design, and patient recruitment in clinical trials.

The major conclusion in most studies is that APOE-4 carriers are the worst responders to conventional treatments [7] , [12] , [15] , [16] , [22] , [23] , [24] , [25] , [26] , [27] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] . When APOE and CYP2D6 genotypes are integrated in bigenic clusters and the APOE + CYP2D6 -related therapeutic response to a combination therapy is analyzed in AD patients, it becomes clear that the presence of the APOE-4/4 genotype is able to convert pure CYP2D6*1/*1 extensive metabolizers (EMs) into full poor responders to conventional treatments, indicating the existence of a powerful influence of the APOE-4 homozygous genotype on the drug-metabolizing capacity of pure CYP2D6 EMs. In addition, a clear accumulation of APOE-4/4 genotypes is observed among CYP2D6 poor (PMs) and ultrarapid metabolizers (UMs) [12] .

27.7.2. Genes Involved in the Mechanism of Action of CNS Drugs

Most genes associated with the mechanism of action of CNS drugs encode receptors, enzymes, and neurotransmitters on which psychotropic drugs act as ligands (agonists, antagonists), enzyme modulators (substrates, inhibitors, inducers), or neurotransmitter regulators (releasers, reuptake inhibitors) [111] . In the case of conventional antidementia drugs, tacrine, donepezil, rivastigmine and galantamine are cholinesterase inhibitors, and memantine is a partial NMDA antagonist ( Table 27.1 ).

Pharmacogenomic Profile of Antidementia Drugs

27.7.3. Genes Involved in Drug Metabolism

Drug metabolism includes phase I reactions (i.e., oxidation, reduction, hydrolysis) and phase II conjugation reactions (i.e., acetylation, glucuronidation, sulphation, methylation) ( Table 27.2 ). The principal enzymes with polymorphic variants involved in phase I reactions are the following: cytochrome P450 monooxygenases (CYP3A4/5/7, CYP2E1, CYP2D6, CYP2C19, CYP2C9, CYP2C8, CYP2B6, CYP2A6, CYP1B1, CYP1A1/2), epoxide hydrolase, esterases, NQO1 (NADPH-quinone oxidoreductase), DPD (dihydropyrimidine dehydrogenase), ADH (alcohol dehydrogenase), and ALDH (aldehyde dehydrogenase). The major enzymes involved in phase II reactions include UGTs (uridine 5′-triphosphate glucuronosyl transferases), TPMT (thiopurine methyltransferase), COMT (catechol-O-methyltransferase), HMT (histamine methyl-transferase), STs (sulfotransferases), GST-A (glutathione S-transferase A), GST-P, GST-T, GST-M, NAT1 (N-acetyl transferase 1), NAT2, and others ( Table 27.2 ).

Drug Metabolism-Related Genes

Note: See Appendix B for long-form names of genes listed.

Among these enzymes, CYP2D6, CYP2C9, CYP2C19, and CYP3A4/5 are the most relevant in the pharmacogenetics of CNS drugs [15] , [111] ( Table 27.1 ). Approximately 18% of neuroleptics are major substrates of CYP1A2 enzymes, 40% of CYP2D6, and 23% of CYP3A4; 24% of antidepressants are major substrates of CYP1A2 enzymes, 5% of CYP2B6, 38% of CYP2C19, 85% of CYP2D6, and 38% of CYP3A4; 7% of benzodiazepines are major substrates of CYP2C19 enzymes, 20% of CYP2D6, and 95% of CYP3A4 [15] , [111] . Most CYP enzymes exhibit ontogenic-, age-, sex-, circadian-, and ethnic-related differences [112] .

In dementia, as in any other CNS disorder, CYP genomics is a very important issue, since in practice more than 90% of patients with dementia are daily consumers of psychotropics. Furthermore, some acetylcholinesterase inhibitors (the most prescribed antidementia drugs worldwide) are metabolized via CYP enzymes ( Table 27.1 ). Most CYP enzymes display highly significant ethnic differences, indicating that the enzymatic capacity of these proteins varies depending upon the polymorphic variants present in their coding CYP genes.

The practical consequence of this genetic variation is that the same drug can be differentially metabolized according to the genetic profile of each subject, and that, if an individual’s pharmacogenomic profile is known, his/her pharmacodynamic response is potentially predictable. This is the cornerstone of pharmacogenetics. In this regard, the CYP2D6 , CYP2C19 , CYP2C9 , and CYP3A4 / 5 genes and their respective protein products deserve special consideration.

27.7.3.1. CYP2D6

CYP2D6 is a 4.38 kb gene with 9 exons mapped on 22q13.2. Four RNA transcripts of 1190–1684 bp are expressed in the brain, liver, spleen, and reproductive system, where 4 major proteins of 48–55 kDa (439–494 aa) are identified. It is a transport enzyme of the cytochrome P450 subfamily IID or multigenic cytochrome P450 superfamily of mixed-function monooxygenases. The cytochrome P450 proteins are monooxygenases which catalyze many reactions involved in drug metabolism and synthesis of cholesterol, steroids, and other lipids. CYP2D6 localizes to the endoplasmic reticulum and is known to metabolize as many as 25% of commonly prescribed drugs, and more than 60% of current psychotropics. Its substrates include debrisoquine, an adrenergic-blocking drug; sparteine and propafenone, both antiarrhythmic drugs; and amitryptiline, an antidepressant. CYP2D6 is highly polymorphic in the population.

There are 141 CYP2D6 allelic variants, of which -100C > T, -1023C > T, -1659G > A, -1707delT, -1846G > A, -2549delA, -2613-2615delAGA, -2850C > T, -2988G > A, and -3183G > A represent the ten most important [113] , [114] , [115] . Different alleles result in the extensive, intermediate, poor, and ultrarapid metabolizer phenotypes, characterized by normal, intermediate, decreased, and multiplied ability to metabolize the enzyme’s substrates, respectively. The hepatic cytochrome P450 system is responsible for the first phase in the metabolism and elimination of numerous endogenous and exogenous molecules and ingested chemicals. P450 enzymes convert these substances into electrophilic intermediates, which are then conjugated by phase II enzymes (e.g., UDP glucuronosyltransferases, N-acetyltransferases) to hydrophilic derivatives that can be excreted. According to the database of the World Guide for Drug Use and Pharmacogenomics [113] , 982 drugs are CYP2D6 -related: 371 are substrates, more than 300 are inhibitors, and 18 are CYP2D6 inducers.

In healthy subjects, extensive metabolizers (EMs) account for 55.71% of the population; intermediate metabolizers (IMs) account for 34.7%; poor metabolizers (PMs), 2.28%; and ultrarapid metabolizers (UMs), 7.31%. Remarkable worldwide interethnic differences exist in the frequency of the PM and UM phenotypes [116] , [117] , [118] . On average, approximately 6.28% of the world’s population belongs to the PM category. Europeans (7.86%), Polynesians (7.27%), and Africans (6.73%) show the highest rate of PMs, whereas Orientals (0.94%) show the lowest [116] . The frequency of PMs among Middle Eastern populations, Asians, and Americans is in the range of 2–3%. CYP2D6 gene duplications are relatively infrequent among Northern Europeans, but in East Africa the frequency of alleles with duplication of CYP2D6 is as high as 29% [119] . In Europe, there is a North–South gradient in the frequency of PMs (6–12% of PMs in Southern European countries, and 2–3% of PMs in Northern latitudes) [111] .

In AD, EMs, IMs, PMs, and UMs are 56.38%, 27.66%, 7.45%, and 8.51%, respectively, and in vascular dementia, they are, respectively, 52.81%, 34.83%, 6.74%, and 5.62% ( Figure 27.3 ). There is an accumulation of AD-related risk genes in PMs and UMs. EMs and IMs are the best responders, and PMs and UMs are the worst responders to a combination therapy of cholinesterase inhibitors, neuroprotectants, and vasoactive substances. The pharmacogenetic response in AD appears to depend on the networking activity of genes involved in drug metabolism and genes involved in AD pathogenesis [7] , [12] , [15] , [16] , [22] , [23] , [24] , [25] , [26] , [27] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] .

Distribution and frequency of CYP2D6 phenotypes in AD and vascular dementia.

EM—extensive metabolizer; IM—intermediate metabolizer; PM—poor metabolizer; UM—ultrarapid metabolizer.

27.7.3.2. CYP2C9

CYP2C9 is a gene (50.71 kb) with 9 exons mapped on 10q24. An RNA transcript of 1860 bp is mainly expressed in hepatocytes, where a protein of 55.63 kDa (490 aa) can be identified. More than 600 drugs are CYP2C9 -related: 311 act as substrates (177 major, 134 minor); 375, as inhibitors (92 weak, 181 moderate, and 102 strong); and 41 as inducers of the CYP2C9 enzyme [113] . There are 481 CYP2C9 SNPs. By phenotype ( Figure 27.4 ), in the control population, PMs represent 7.04%, IMs 32.39%, and EMs 60.56%. In AD, PMs, IMs, and EMs are 6.45%, 37.64%, and 55.91%, respectively, and in vascular dementia they are 3.61%, 28.92%, and 67.47%, respectively [18] ( Figure 27.4 ).

Distribution and frequency of CYP2C9 phenotypes in AD and vascular dementia.

EM—extensive metabolizer; IM—intermediate metabolizer; PM—poor metabolizer.

27.7.3.3. CYP2C19

CYP2C19 is a gene (90.21 kb) with 9 exons mapped on 10q24.1q24.3. RNA transcripts of 1901 bp, 2395 bp, and 1417 bp are expressed in liver cells, where a protein of 55.93 kDa (490 aa) has been identified. Nearly 500 drugs are CYP2C19 -related, with 281 acting as substrates (151 major, 130 minor), 263 as inhibitors (72 weak, 127 moderate, and 64 strong), and 23 as inducers of the CYP2C19 enzyme [113] . About 541 SNPs have been detected in the CYP2C19 gene. The frequencies of the three major CYP2C19 geno-phenotypes in the control population are CYP2C19-*1/*1 -EMs, 68.54%; CYP2C19-*1/*2 -IMs, 30.05%; and CYP2C19-*2/*2 -PMs, 1.41%. EMs, IMs, and PMs account for 69.89%, 30.11%, and 0%, respectively, in AD, and 66.27%, 30.12%, and 3.61%, respectively, in vascular dementia [18] ( Figure 27.5 ).

Distribution and frequency of CYP2C19 pheno- genotypes in AD and vascular dementia.

27.7.3.4. CYP3A4/5

CYP3A4 is a gene (27.2 kb) with 13 exons mapped on 7q21.1. RNA transcripts of 2153 bp, 651 bp, 564 bp, 2318 bp, and 2519 bp are expressed in intestine, liver, prostate, and other tissues, where four protein variants of 57.34 kDa (503 aa), 17.29 kDa (153 aa), 40.39 kDa (353 aa), and 47.99 kDa (420 aa) have been identified. The human CYP3A locus contains the three CYP3A genes ( CYP3A4 , CYP3A5, and CYP3A7 ), three pseudogenes, and a novel CYP3A gene termed CYP3A43 . The gene encodes a putative protein with 71.5–75.8% identity with the other CYP3A proteins. The predominant hepatic form is CYP3A4, but CYP3A5 contributes significantly to total liver CYP3A activity.

CYP3A4 metabolizes more than 1900 drugs: 1033 act as substrates (897 major, 136 minor); 696, as inhibitors (118 weak, 437 moderate, and 141 strong); and 241, as inducers of the CYP3A4 enzyme [113] . About 347 SNPs have been identified in the CYP3A4 gene ( CYP3A4*1A : wild-type), 25 of which are of clinical relevance. Concerning CYP3A4/5 polymorphisms in AD, 82.75% of cases are EMs ( CYP3A5*3/*3 ), 15.88% are IMs ( CYP3A5*1/*3 ), and 1.37% are UMs ( CYP3A5*1/*1 ). Unlike other human P450s ( CYP2D6 , CYP2C19 ), there is no evidence of a “null” allele for CYP3A4 [113] .

27.7.3.5. CYP Clustering

The construction of a genetic map integrating the most prevalent CYP2D6 + CYP2C19 + CYP2C9 polymorphic variants in a trigenic cluster yields 82 different haplotype-like profiles. The most frequent trigenic genotypes in the AD population are *1*1-*1*1-*1*1 (25.70%), *1*1-*1*2-*1*2 (10.66%), *1*1-*1*1-*1*1 (10.45%), *1*4-*1*1-*1*1 (8.09%), *1*4-*1*2-*1*1 (4.91%), *1*4-*1*1-*1*2 (4.65%), and *1*1-*1*3-*1*3 (4.33%). These 82 trigenic genotypes represent 36 different pharmacogenetic phenotypes.

According to these trigenic clusters, only 26.51% of patients show a pure 3EM phenotype, 15.29% are 2EM1IM, 2.04% are pure 3IM, 0% are pure 3PM, and 0% are 1UM2PM (the worst possible phenotype). This implies that only one-quarter of the population normally process the drugs that are metabolized via CYP2D6, CYP2C9, and CYP2C19 (approximately 60% of the drugs in current use) [12] . Taking into consideration the data available, it might be inferred that at least 20–30% of the AD population may exhibit an abnormal metabolism of cholinesterase inhibitors and/or other drugs that undergo oxidation via CYP2D6 -related enzymes.

Approximately 50% of this population cluster shows an ultrarapid metabolism, requiring higher doses of cholinesterase inhibitors in order to reach a therapeutic threshold. The other 50% of the cluster exhibit a poor metabolism, displaying potential adverse events at low doses. If we take into account that approximately 60–70% of therapeutic outcomes depend on pharmacogenomic criteria (e.g., pathogenic mechanisms associated with AD-related genes), it can be postulated that pharmacogenetic and pharmacogenomic factors are responsible for 75–85% of therapeutic response (efficacy) in AD patients treated with conventional drugs [12] , [15] , [16] , [17] , [18] , [22] , [23] , [24] , [25] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] .

27.7.4. Drug Transporters

ABC genes—especially ABCB1 (ATP-binding cassette, subfamily B, member 1P-glycoprotein-1, P-gp1, Multidrug Resistance 1, MDR (17q21.12), ABCC1 (9q31.1), ABCG2 (White121q22.3), and other genes of this family—encode proteins that are essential for drug metabolism and transport. The multidrug efflux transporters P-gp, the multidrug resistance-associated protein 4 (MRP4), and the breast cancer resistance-protein (BCRP), located on endothelial cells lining the brain vasculature, play important roles in limiting the movement of substances into the brain and in enhancing their efflux from the brain.

Transporters also cooperate with phase I/phase II metabolism enzymes by eliminating drug metabolites. Their major features are their capacity to recognize drugs belonging to unrelated pharmacological classes and their redundancy, by which a single molecule can act as a substrate for different transporters. This ensures efficient neuroprotection against xenobiotic invasions. The pharmacological induction of ABC gene expression is a mechanism of drug interaction, which may affect substrates of the upregulated transporter; overexpression of MDR transporters confers resistance to anti-cancer agents and CNS drugs [120] , [121] .

Also of importance for CNS pharmacogenomics are transporters encoded by genes of the solute carrier superfamily (SLC) and solute carrier organic (SLCO) transporter family, which are responsible for the transport of multiple endogenous and exogenous compounds, including folate ( SLC19A1 ), urea ( SLC14A1 , SLC14A2 ), monoamines ( SLC29A4 , SLC22A3 ), aminoacids ( SLC1A5 , SLC3A1 , SLC7A3 , SLC7A9 , SLC38A1 , SLC38A4 , SLC38A5 , SLC38A7 , SLC43A2 , SLC45A1 ), nucleotides ( SLC29A2 , SLC29A3 ], fatty acids ( SLC27A1-6 ), neurotransmitters ( SLC6A2 [noradrenaline transporter]), SLC6A3 [dopamine transporter], SLC6A4 [serotonin transporter, SERT], SLC6A5 , SLC6A6 , SLC6A9 , SLC6A11 , SLC6A12 , SLC6A14 , SLC6A15 , SLC6A16 , SLC6A17 , SLC6A18 , SLC6A19 ), glutamate ( SLC1A6 , SLC1A7 ), and others [122] .

Some organic anion transporters (OAT), which belong to the solute carrier (SLC) 22A family, are also expressed at the BBB, and regulate the excretion of endogenous and exogenous organic anions and cations [123] . The transport of amino acids and di- and tripeptides is mediated by a number of different transporter families, and the bulk of oligopeptide transport is attributable to the activity of members of the SLC15A superfamily (peptide transporters 1 and 2 ( SLC15A1 [PepT1]) and SLC15A2 [PepT2], and peptide/histidine transporters 1 and 2 ( SLC15A4 [PHT1] and SLC15A3 [PHT2]). ABC and SLC transporters expressed at the BBB may cooperate to regulate the passage of different molecules into the brain [124] . Polymorphic variants in ABC and SLC genes may also be associated with pathogenic events in CNS disorders and drug-related safety and efficacy complications [111] , [122] .

27.7.5. Pleiotropic Activity of APOE in Dementia

APOE is the prototypical paradigm of a pleiotropic gene with multifaceted activities in physiological and pathological conditions [16] , [22] . ApoE is consistently associated with the amyloid plaque marker for AD. APOE -4 may influence AD pathology interacting with APP metabolism and Aβ accumulation, enhancing hyperphosphorylation of tau protein and NFT formation, reducing choline acetyltransferase activity, increasing oxidative processes, modifying inflammation-related neuroimmunotrophic activity and glial activation, altering lipid metabolism, lipid transport, and membrane biosynthesis in sprouting and synaptic remodeling, and inducing neuronal apoptosis [16] , [23] , [24] , [25] .

To address the complex misfolding and aggregation that initiates the toxic cascade resulting in AD, Petrlova et al. [26] developed a 2,2,6,6-tetramethylpiperidine-1-oxyl-4-amino-4-carboxylic acid spin-labeled amyloid-β (Aβ) peptide to observe its isoform-dependent interaction with the ApoE protein. Oligomer binding involves the C-terminal domain of ApoE, with ApoE3 reporting a much greater response through this conformational marker. ApoE3 displays a higher affinity and capacity for the toxic Aβ oligomer. ApoE polymorphism and AD risk can largely be attributed to the reduced ability of ApoE4 to function as a clearance vehicle for the toxic form of Aβ. MAPT and APOE are involved in the pathogenic mechanisms of AD, and both the MAPT H1/H1 genotype and the APOE ɛ4 allele lead to a more rapid progression to dementia among MCI subjects, probably mediating an increased rate of amyloid-β and tau brain deposition [27] .

The distribution of APOE genotypes in the Iberian peninsula is as follows: APOE-2/2 0.32%; APOE-2/3 7.3%; APOE-2/4 1.27%; APOE-3/3 71.11%; APOE-3/4 18.41%; and APOE-4/4 1.59% [18] ( Figure 27.2 ). These frequencies are very similar in Europe and in other Western societies. There is a clear accumulation of APOE-4 carriers among patients with AD ( APOE-3/4 30.30%, APOE-4/4 6.06%) and vascular dementia ( APOE-3/4 35.85%, APOE-4/4 6.57%) as compared to controls ( Figure 27.2 ). Different APOE genotypes confer specific phenotypic profiles to AD patients [15] , [16] , [22] . Some of these profiles may add risk or benefit when patients are treated with conventional drugs, and in many instances the clinical phenotype demands the administration of additional drugs that increase the complexity of therapeutic protocols.

From studies designed to define APOE -related AD phenotypes [7] , [12] , [23] , [24] , [25] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] , several conclusions can be drawn, which are shown in Box 27.2 . These 20 major phenotypic features clearly illustrate the biological disadvantage of APOE-4 homozygotes and the potential consequences that these patients may experience when they receive pharmacological treatment for AD and/or concomitant pathologies [7] , [12] , [23] , [24] , [25] , [28] , [53] , [54] , [55] , [56] , [57] , [58] , [59] .

Key Conclusions Regarding APOE-Related AD Phenotypes

• 1. The age at onset is 5–10 years earlier in approximately 80% of AD cases harboring the APOE-4/4 genotype.
• 2. The serum levels of ApoE are lowest in APOE-4/4 , intermediate in APOE-3/3 and APOE-3/4 , and highest in APOE-2/3 and APOE-2/4 .
• 3. Serum cholesterol levels are higher in APOE-4/4 than in other genotypes.
• 4. HDL-cholesterol levels tend to be lower in APOE-3 homozygotes than in APOE-4 allele carriers.
• 5. LDL-cholesterol levels are systematically higher in APOE-4/4 than in any other genotype.
• 6. Triglyceride levels are significantly lower in APOE-4/4 .
• 7. Nitric oxide levels are slightly lower in APOE-4/4 .
• 8. Serum and cerebrospinal fluid Aβ levels tend to differ between APOE-4/4 and the other most frequent genotypes ( APOE-3/3 , APOE-3/4 ).
• 9. Blood histamine levels are dramatically reduced in APOE-4/4 as compared to the other genotypes.
• 10. Brain atrophy is markedly increased in APOE-4/4 > APOE-3/4 > APOE-3/3 .
• 11. Brain mapping activity shows a significant increase in slow wave activity in APOE-4/4 from the early stages of the disease.
• 12. Brain hemodynamics, as reflected by reduced brain blood flow velocity and increased pulsatility and resistance indices, is significantly worse in APOE-4/4 (and in APOE-4 carriers in general, as compared with APOE-3 carriers); brain hypoperfusion and neocortical oxygenation is also more deficient in APOE-4 carriers.
• 13. Lymphocyte apoptosis is markedly enhanced in APOE-4 carriers.
• 14. Cognitive deterioration is faster in APOE-4/4 patients than in carriers of any other APOE genotype.
• 15. In approximately 3–8% of AD cases, some dementia-related metabolic dysfunctions accumulate more in APOE-4 carriers than in APOE-3 carriers.
• 16. Some behavioral disturbances, alterations in circadian rhythm patterns, and mood disorders are slightly more frequent in APOE-4 carriers.
• 17. Aortic and systemic atherosclerosis is more frequent in APOE-4 carriers.
• 18. Liver metabolism and transaminase activity differ in APOE-4/4 with respect to other genotypes.
• 19. Hypertension and other cardiovascular risk factors accumulate in APOE-4 carriers.
• 20. APOE-4/4 carriers are the poorest responders to conventional drugs.

27.7.6. Pharmacogenomics of Antidementia Drugs

The following list describes the pharmacogenomics of the most common antidementia drugs ( Table 27.1 ).

Donepezil: is a centrally active, reversible acetylcholinesterase inhibitor that increases the acetylcholine available for synaptic transmission in the CNS. The therapeutic response of donepezil is influenced by pathogenic gene variants ( APOE , CHAT ), as well as mechanistic gene polymorphic variants ( CHAT , ACHE , and BCHE ). It is a major substrate of CYP2D6, CYP3A4, ACHE, and UGTs; it inhibits ACHE and BCHE; and it is transported by ABCB1 [113] .

Galantamine: is a reversible and competitive acetylcholinesterase inhibitor leading to increased concentration of acetylcholine at cholinergic synapses. It also modulates nicotinic acetylcholine receptors and may increase glutamate and serotonin levels. APOE , APP , ACHE , BCHE , CHRNA4 , CHRNA7 , and CHRNB2 variants may potentially influence galantamine efficacy and safety. Galantamine is a major substrate of CYP2D6, CYP3A4, and UGT1A1, and an inhibitor of ACHE and BCHE [113] .

Rivastigmine: is a cholinesterase inhibitor that increases acetylcholine in the CNS through reversible inhibition of its hydrolysis by cholinesterase. APOE , APP , CHAT , ACHE , BCHE , CHRNA4 , CHRNB2 , and MAPT variants may affect its pharmacokinetics and pharmacodynamics [113] .

Tacrine: is the first FDA-approved antidementia drug. Its use was stopped due to hepatotoxicity. Tacrine is a cholinesterase inhibitor that elevates acetylcholine in the cerebral cortex by slowing degradation of acetylcholine. ACHE , BCHE , CHRNA4 , CHRNB2 , APOE , MTHFR , CES1 , LEPR , GSTM1 , and GSTT1 variants may affect its therapeutic and toxic effects. Tacrine is a major substrate of CYP1A2 and CYP3A4, a minor substrate of CYP2D6, and is transported via SCN1A. It is an inhibitor of ACHE, BCHE, and CYP1A2 [113] .

Memantine: is an N-Methyl-D-Aspartate (NMDA) receptor antagonist that binds preferentially to NMDA receptor-operated cation channels. It may act by blocking the actions of glutamate, mediated in part by NMDA receptors, and it is also an antagonist of GRIN2A, GRIN2B, GRIN3A, HTR3A, and CHRFAM7A. Several pathogenic ( APOE , PSEN1 , MAPT ) and mechanistic gene variants ( GRIN2A , GRIN2B , GRIN3A , HTR3A , CHRFAM7A ) may influence its therapeutic effects. Memantine is a strong inhibitor of CYP2B6 and CYP2D6, and a weak inhibitor of CYP1A2, CYP2A6, CYP2C9, CYP2C19, CYP2E1, and CYP3A4 [113] .

27.7.7. Multifactorial Therapy

Some studies using a multifactorial approach also have shown that diverse pharmacogenomic factors may influence efficacy and safety. In one of these studies [15] , [58] , patients with dementia received the following for three months: a multifactorial therapy integrated by CDP-choline (500 mg/day, p.o.), Nicergoline (5 mg/day, p.o.), Sardilipin (E-SAR-94010) (LipoEsar ® )(250 mg, t.i.d.), and Animon Complex ® (2 capsules/day)—a nutraceutical compound integrated by a purified extract of Chenopodium quinoa (250 mg), ferrous sulphate (38.1 mg equivalent to 14 mg of iron), folic acid (200 μg), and vitamin B 12 (1 μg) per capsule (RGS: 26.06671/C).

Patients with chronic deficiencies of iron (<35 μg/mL), folic acid (<2.5 ng/mL), or vitamin B 12 (<150 pg/mL) received an additional supplement of iron (80 mg/day), folic acid (5 mg/day), and B complex vitamins (B 1 , 15 mg/day; B 2 , 15 mg/day; B 6 , 10 mg/day; B 12 , 10 μg/day; nicotinamide, 50 mg/day), respectively, to maintain stable levels of serum iron (50–150 μg/mL), folic acid (5–20 ng/mL) and vitamin B 12 levels (500–1000 pg/mL) in order to avoid the negative influence of all these metabolic factors on cognition. Patients with hypertension (>150/85 mmHg) received Enalapril (20 mg/day).

The frequency of APOE genotypes was APOE-2/3, 7.97%; APOE-2/4 , 1.18%; APOE-3 , 58.95%; APOE-3/4 , 27.32%; and APOE-4/4 , 4.58%. Cognitive function (as assessed by MMSE); 20.51 ± 6.51 vs. 21.45 ± 6.95, p < 0.0000000001; ADAS-Cog, 22.94 ± 13.87 vs. 21.23 ± 12.84, p < 0.0001; ADAS-Non-Cog, 5.26 ± 4.18 vs. 4.15 ± 3.63, p < 0.0000000001; ADAS-Total, 27.12 ± 16.93 vs. 24.28 ± 15.06, p < 0.00009) improved after treatment. Mood (HAM-A, 11.35 ± 5.44 vs. 9.79 ± 4.33, p < 0.0000000001; HAM-D, 10.14 ± 5.23 vs. 8.59 ± 4.30, p < 0.0000000001) also improved. Glucose levels did not change.

Total cholesterol levels (224.78 ± 45.53 vs. 203.64 ± 39.69 mg/dL, p < 0.0000000001), HDL-cholesterol levels (54.11 ± 14.54 vs. 52.54 ± 14.86 mg/dL, p < 0.0001), and LDL-cholesterol levels (148.15 ± 39.13 vs. 128.89 ± 34.83 mg/dL, p < 0.0000000001) were significantly reduced. Folate (7.07 ± 3.61 vs. 18.14 ± 4.23 ng/mL, p < 0.000000001) and vitamin B 12 levels (459.65 ± 205.80 vs. 689.78 ± 338.82 pg/mL, p < 0.000000001) also increased, and both TSH and T 4 levels remained unchanged after treatment. The response rate in terms of cognitive improvement was as follows: 59.74% responders (RRs), 24.44% nonresponders (NRs), and 15.82% stable responders (SRs) (no change in MMSE score after three months of treatment). The response rate in cholesterol levels was very similar: 57.78% RRs, 28.50% NRs, and 13.72% SRs [15] .

27.7.7.1. APOE-Related Cognitive Function Changes

In this study, the basal MMSE score differed in APOE-2/3 carriers with respect to APOE-2/4 (p < 0.02), APOE-3/4 (p < 0.004), and APOE-4/4 (p < 0.0009), in APOE-3/3 vs. APOE-3/4 (p < 0.0005), and in APOE-3/3 vs. APOE-4/4 (p < 0.002). The best responders were APOE-3/3 (p < 0.0000000001) > APOE-3/4 (p < 0.00001) > APOE-4/4 carriers (p < 0.05). Patients harboring the APOE-2/3 and APOE-2/4 genotypes did not show any significant improvement. The response rate by genotype was the following: APOE-2/3 : 44.26% RRs, 36.07% NRs, 19.67% SRs; APOE-2/4 : 55.56% RRs, 44.44% NRs, 0.0% SRs; APOE-3/3 : 63.42% RRs, 21.06% NRs, 15.52% SRs; APOE-3/4 : 56.94% RRs, 27.75% NRs, 15.31% SRs; and APOE-4/4 : 51.43% RRs, 28.57% NRs, 20.00% SRs [15] ( Figure 27.6 , Figure 27.7 ).

APOE -related cognitive performance in response to multifactorial therapy in patients with dementia.

Tb—basal MMSE score prior to treatment; Tt—MMSE score after 3 months treatment in total sample. E2/3b—basal MMSE score in APOE-2/3 carriers; E2/3t—MMSE score after treatment in APOE-2/3 carriers; E2/4b—basal MMSE score in APOE-2/4 carriers; E2/4t—MMSE score after treatment in APOE-2/4 carriers; E3/3b—basal MMSE score in APOE-3/3 carriers; E3/3t—MMSE score after treatment in APOE-3/3 carriers; E3/4b basal MMSE score in APOE-3/4 carriers; E3/4t—MMSE score after treatment in APOE-3/4 carriers; E4/4b—basal MMSE score in APOE-4/4 carriers; E4/4—MMSE score after treatment in APOE-4/4 carriers.

APOE -related cognitive response rate in patients with dementia treated with multifactorial therapy.

27.7.7.2. APOE-Related Changes in Blood Pressure Values

Systolic blood pressure (SBP) was significantly reduced in patients with the APOE-3/3 (p < 0.00007) and APOE-3/4 genotypes (p < 0.01), and diastolic blood pressure exhibited a similar pattern ( APOE-3/3 , p < 0.005; APOE-3/4 , p < 0.01), with no changes in either SBP or DBP in APOE-2/3 , APOE-2/4 , and APOE-4/4 carriers [15] .

27.7.7.3. APOE-Related Blood Lipid Response to Sardilipin

Basal cholesterol levels were significantly different in patients with the APOE-2/3 genotype vs. APOE-3/3 (p < 0.007), vs. APOE-3/4 (p < 0.001), vs. APOE-4/4 (p < 0.00002); APOE-2/4 vs. APOE-4/4 (p < 0.01); APOE-3/3 vs. APOE-4/4 (p < 0.005); and APOE-3/4 vs. APOE-4/4 (p < 0.01).

The highest cholesterol levels were seen in APOE-4/4 > APOE-3/4 > APOE-3/3. All patients showed a clear reduction in cholesterol levels after treatment with Sardilipin. This was particularly significant in APOE-3/3 (p < 0.0000000001) > APOE-3/4 (p < 0.00000008) > APOE-4/4 (p < 0.002) > APOE-2/3 (p < 0.02) > APOE-2/4 carriers (p: 0.26). The response rate by genotype was as follows: APOE-2/3 : 63.93% RRs, 29.51% NRs, 6.56% SRs; APOE-2/4 : 44.44% RRs, 22.22% NRs, 33.34% SRs; APOE-3/3 : 54.32% RRs, 28.16% NRs, 17.52% SRs; APOE-3/4 : 53.59% RRs, 31.58% NRs, 14.83% SRs; APOE-4/4 : 65.71% RRs, 20.00% NRs, 14.29% SRs [15] .

HDL-cholesterol levels significantly decreased in APOE-3/3 (p < 0.001) > APOE-3/4 (p < 0.05), with no significant changes in patients with other genotypes. In contrast, LDL-cholesterol levels showed changes identical to those observed in total cholesterol, with similar differences among genotypes at baseline and almost identical decreased levels after treatment ( APOE-3/3 , p > 0.0000000001 > APOE-3/4 , p < 0.00001 > APOE-2/3 , p < 0.0004 > APOE-4/4 , p < 0.001 > APOE-2/4 , p:0.31) [15] .

Sardilipin (E-SAR-94010, LipoEsar ® , LipoSea ® ) is a natural product extracted from the marine species Sardina pilchardus by means of nondenaturing biotechnological procedures. The main chemical compounds of LipoEsar ® are lipoproteins (60–80%), whose micelle structure probably mimics that of physiological lipoproteins involved in lipid metabolism. In preclinical studies, Sardilipinhas been shown to be effective in:

• 1. Reducing blood cholesterol (CHO), triglyceride (TG), uric acid (UA), and glucose (Glu) levels, as well as liver alanine aminotransferase (ALT) and aspartate aminotransferase (AST) activity.
• 2. Enhancing immunological function by regulating both lymphocyte and microglia activity.
• 3. Inducing antioxidant effects mediated by superoxide dismutase activity.
• 4. Improving cognitive function [15] .

According to these results, it appears that the therapeutic response of patients with dyslipidemia to Sardilipin is APOE -related. The best responders were patients with APOE-3/3 > APOE-3/4 > APOE-4/4 . Patients with the other APOE genotypes ( 2/2 , 2/3 , 2/4 ) did not show any hypolipidemic response to this novel compound. In patients with dementia, the effects of Sardilipin were very similar to those observed in patients with chronic dyslipidemia, suggesting that the lipid-lowering properties of Sardilipin are APOE -dependent [15] ( Figure 27.8 ).

APOE -related total cholesterol levels in response to multifactorial therapy in patients with dementia.

Tb—basal cholesterol levels prior to treatment; Tt—cholesterol levels after 3 months treatment in total sample. E2/3b—basal cholesterol levels in APOE-2/3 carriers; E2/3t—total cholesterol levels after treatment in APOE-2/3 carriers; E2/4b—basal cholesterol levels in APOE-2/4 carriers; E2/4t—total cholesterol levels after treatment in APOE-2/4 carriers; E3/3b—basal cholesterol levels in APOE-3/3 carriers; E3/3t—total cholesterol levels after treatment in APOE-3/3 carriers; E3/4b—basal cholesterol levels in APOE-3/4 carriers; E3/4t—total cholesterol levels after treatment in APOE-3/4 carriers; E4/4b—basal cholesterol levels in APOE-4/4 carriers; E4/4—total cholesterol levels after treatment in APOE-4/4 carriers.

27.8. Future Perspective

To make AD a global health priority in the coming years, conceptual and procedural changes are needed on several grounds, such as (1) political, administrative, economic, legal, ethical, industrial, regulatory and educational issues; (2) novel biomarkers (genomics, proteomics, molecular neuroimaging) as diagnostic aids; (3) innovative therapeutics; (4) pharmacogenomics in clinical practice to optimize therapeutics; and (5) selective preventive plans for the population at risk.

There is disharmony concerning the public and governmental interest in dementia and its social, medical, and economic implications. The diagnosis and management of dementia is dissimilar in Europe, North America, Latin America, Asia, Africa, and Oceania. The economic/cultural status of each country (developed versus developing), the particular epidemiology of aging and dementia in each latitude, national standards of education, health priorities (infectious diseases versus degenerative diseases), and the quality and efficiency of medical services are conditioning factors for investing (or not investing) national resources in dementia as a health priority.

Educational programs, international guidelines, and consensus protocols for the management of dementia are necessary for global harmonization, for professionals from different countries to speak the same conceptual language, and to improve cost-effectiveness ratios [125] , [126] , [127] , [128] . There are many legal issues (e.g., informed consent, lawsuit, testament, tutorship) and ethical issues (e.g., clinical trials, use of genetic information, institutionalization) that deserve more attention in order to humanize the end of life in the very frail conditions under which demented patients survive.

The updating of regulatory issues is also a matter of deep concern. Regulatory aspects of drug development are not universal, with notable peculiarities in the European Union (EMA), the United States (FDA), and Japan (Koseisho). Because the costs of dementia cannot be fully assumed by more than 60% of the European population, European authorities must take into account this circumstance when health reform is implemented in the coming years [8] , [18] .

Genomics, transcriptomics, proteomics, and metabolomics will revolutionize medicine in the next decades. Genetic testing is gaining acceptance among physicians and patients in different countries [128] , [129] , [130] , [131] , although Americans, Europeans, and Japanese differ notably in their knowledge, beliefs, and attitudes regarding genetic testing for AD [128] , [131] , [132] . The validation of protocols for genomic screening will contribute to the implementation of structural genomics, functional genomics, and proteomics as diagnostic aids and therapeutic targets [133] .

An accurate diagnosis of AD demands the use of reliable biomarkers in routine protocols at a reasonable price [68] . Levels of specific secreted cellular signaling proteins in cerebrospinal fluid or plasma correlate with pathological changes in the AD brain; therefore, proteomic analysis of these levels can be used to discover said biomarkers [134] . It is likely that the best biomarkers result from a combination of genomic, transcriptomic, and proteomic analyses of body fluids. The measurement of these biomarkers correlates with brain imaging markers and cognitive performance [73] , [74] , [75] .

New initiatives for the prevention of dementia (global versus selective prevention) will also emerge [135] , together with new insights into the role of nutrition and nutrigenomics in brain function and neurodegeneration [59] , [136] . In terms of prevention, it must be taken into consideration that neuronal death and Aβ accumulation starts many years before the onset of the disease, and that preventive strategies should be selective to protect the population at risk. For this purpose, accurate biomarkers are essential, and surrogate markers are needed to facilitate primary prevention.

Without doubt, the highest priority for the coming decade will be an intense search for novel therapeutic options in the form of both symptomatic treatments and preventive strategies. Past failures must be studied by researchers and the pharmaceutical industry in order to avoid unnecessary expenses in redundant trials that lead nowhere. Combination treatments require further evaluation and more sophisticated strategies than dual combinations [137] , [138] . The administration of psychotropic drugs to demented patients should be reduced and predicted with pharmacogenetic markers to minimize side effects, cerebrovascular risk, and cognitive deterioration.

Priority areas for pharmacogenetic research are the prediction of serious adverse drug reactions (ADRs) and the determination of efficacy variation [139] . Both are necessary in CNS disorders and dementia to cope with efficacy and safety issues associated with current psychotropics and antidementia drugs, as well as new CNS drugs. With regard to the future of pharmacogenomics as a practical discipline, several issues should be addressed:

• • The education of physicians in medical genomics and pharmacogenomics is fundamental (less than 2% of clinicians are familiar with genomic science)
• • Genomic screening of gene clusters involved in pharmacogenomic outcomes must become a clinical routine (without genetic testing, there is no pharmacogenetics)
• • Each patient must be a carrier of a pharmacogenetic card [140] indicating what kind of drugs he/she can take and which medications he/she should avoid
• • Regulatory agencies should request pharmacogenetic data from the pharmaceutical industry when applying for drug approval
• • Pharmacogenetic data must be incorporated into patient information leaflets and the pharmaceutical vade mecum
• • New guidelines for daily praxis, such as those given in the World Guide for Drug Use and Pharmacogenomics [113] , will promote understanding of the relationship between drugs and genes to make drug prescription truly personalized

27.9. Conclusion

AD is a major health problem that comes with a high cost to society. As a clinical entity, AD is a polygenic/complex disorder in which many different gene clusters may be involved. Most genes screened to date belong to different proteomic and metabolomic pathways that potentially affect AD pathogenesis, represented by accumulation of Aβ deposits in senile plaques, intracellular NFTs with hyperphosphorylated tau, and neuronal loss.

The presence of the APOE-4 allele of the apolipoprotein E gene seems to be a major risk factor for both degenerative and vascular dementia, and APOE variants are directly involved in AD pathogenesis at multiple levels. Specific biomarkers (structural and functional genomic markers, proteomic markers in body fluids, neuroimaging markers) are needed for an accurate AD diagnosis. Current pharmacological treatment of AD with cholinesterase inhibitors (donepezil, rivastigmine, galantamine) and memantine is not cost-effective; moreover, the overuse of psychotropic drugs in patients with dementia contributes to deteriorating cognitive and psychomotor functions.

Old treatments addressed memory impairment. New treatments are oriented to halting disease progression by interfering with Aβ accumulation, NFT formation, oxidative stress, neuroinflammation, and cerebrovascular damage. Over the past few years diverse candidate drugs have been investigated in AD models, but not one has reached the market. Since only 25–30% of the population is an extensive metabolizer for drugs metabolized via CYP2D6, CYP2C9, and CYP2C19 enzymes, it seems reasonable to use pharmacogenomic procedures as a way to optimize AD therapeutics, thus reducing ADRs and unnecessary costs. The therapeutic response to conventional drugs in patients with AD is genotype-specific, with CYP2D6 -PMs, CYP2D6 -UMs, and APOE-4/4 carriers shown to be the worst responders. APOE and CYP2D6 may cooperate, as pleiotropic genes, in the metabolism of drugs and hepatic function.

If we know the pharmacogenomic profiles of patients who require treatment with antidementia drugs and/or psychotropic drugs currently in use, we may be able to achieve the following benefits:

• • Identifying candidate patients with the ideal genomic profile to receive a particular drug
• • Adapting the dose in more than 90% of cases according to the condition of EM, IM, PM, or UM, which will limit the occurrence of direct side effects in 30–50% of cases
• • Reducing drug interactions by 30–50% (avoiding the administration of inhibitors or inducers able to modify the normal enzymatic activity on a particular substrate)
• • Enhancing efficacy
• • Eliminating unnecessary costs (>30% of pharmaceutical direct costs) deriving from the consequences of inappropriate drug selection and overmedication to mitigate ADRs [18]

Selected Genes Potentially Associated with Alzheimer’s Disease

Long-Form Names for Genes Listed in Table 27.2

• ADH1A : Alcohol dehydrogenase 1A (class I), alpha polypeptide
• AADAC : Arylacetamide deacetylase
• AANAT : Aralkylamine N-acetyltransferase
• ACSL1 : Acyl-CoA synthetase long-chain family member 1
• ACSL3 : Acyl-CoA synthetase long-chain family member 3
• ACSL4 : Acyl-CoA synthetase long-chain family member 4
• ACSM1 : Acyl-CoA synthetase medium-chain family member 1
• ACSM2B : Acyl-CoA synthetase medium-chain family member 2B
• ACSM3 : Acyl-CoA synthetase medium-chain family, member 3
• ADH1B : Alcohol dehydrogenase 1B (class I), beta polypeptide
• ADH1C : Alcohol dehydrogenase 1C (class I), gamma polypeptide
• ADH4 : Alcohol dehydrogenase 4 (class II), pi polypeptide
• ADH5 : Alcohol dehydrogenase 5 (class III), chi polypeptide
• ADH6 : Alcohol dehydrogenase 6 (class V)
• ADH7 : Alcohol dehydrogenase 7 (class IV), mu or sigma polypeptide
• ADHFE1 : Alcohol dehydrogenase, iron containing, 1
• AGXT : Alanine-glyoxylate aminotransferase
• AKR1A1 : Aldo-keto reductase family 1, member A1 (aldehyde reductase)
• AKR1B1 : Aldo-keto reductase family 1, member B1 (aldose reductase)
• AKR1C1 : Aldo-keto reductase family 1, member C1
• AKR1D1 : Aldo-keto reductase family 1, member D1
• ALDH1A1 : Aldehyde dehydrogenase 1 family, member A1
• ALDH1A2 : Aldehyde dehydrogenase family 1, subfamily A2
• ALDH1A3 : Aldehyde dehydrogenase family 1, subfamily A3
• ALDH1B1 : Aldehyde dehydrogenase 1 family, member B1
• ALDH2 : Aldehyde dehydrogenase 2 family (mitochondrial)
• ALDH3A1 : Aldehyde dehydrogenase 3 family, member A1
• ALDH3A2 : Aldehyde dehydrogenase 3 family, member A2
• ALDH3B1 : Aldehyde dehydrogenase 3 family, member B1
• ALDH3B2 : Aldehyde dehydrogenase 3 family, member B2
• ALDH4A1 : Aldehyde dehydrogenase 4 family, member A1
• ALDH5A1 : Aldehyde dehydrogenase 5 family, member A1
• ALDH6A1 : Aldehyde dehydrogenase 6 family, member A1
• ALDH7A1 : Aldehyde dehydrogenase 7 family, member A1
• ALDH8A1 : Aldehyde dehydrogenase 8 family, member A1
• ALDH9A1 : Aldehyde dehydrogenase 9 family, member A1
• AOX1 : Aldehyde oxidase 1
• AS3MT : Arsenic (+3 oxidation state) methyltransferase
• ASMT : Acetylserotonin O-methyltransferase
• BAAT : Bile acid CoA: amino acid N-acyltransferase (glycine N-choloyltransferase)
• CBR1 : Carbonyl reductase 1
• CBR3 : Carbonyl reductase 3
• CBR4 : Carbonyl reductase 4
• CCBL1 : Cysteine conjugate-beta lyase, cytoplasmic
• CDA : Cytidine deaminase
• CEL : Carboxyl ester lipase
• CES1 : Carboxylesterase 1
• CES1P1 : Carboxylesterase 1 pseudogene 1
• CES2 : Carboxylesterase 2
• CES3 : Carboxylesterase 3
• CES5A : Carboxylesterase 5A
• CHST1 : Carbohydrate (keratan sulfate Gal-6) sulfotransferase 1
• CHST2 : Carbohydrate (N-acetylglucosamine-6-O) sulfotransferase 2
• CHST3 : Carbohydrate (chondroitin 6) sulfotransferase 3
• CHST4 : Carbohydrate (N-acetylglucosamine 6-O) sulfotransferase 4
• CHST5 : Carbohydrate (N-acetylglucosamine 6-O) sulfotransferase 5
• CHST6 : Carbohydrate (N-acetylglucosamine 6-O) sulfotransferase 6
• CHST7 : Carbohydrate (N-acetylglucosamine 6-O) sulfotransferase 7
• CHST8 : Carbohydrate (N-acetylgalactosamine 4-0) sulfotransferase 8
• CHST9 : Carbohydrate (N-acetylgalactosamine 4-0) sulfotransferase 9
• CHST10 : Carbohydrate sulfotransferase 10
• CHST11 : Carbohydrate (chondroitin 4) sulfotransferase 11
• CHST12 : Carbohydrate (chondroitin 4) sulfotransferase 12
• CHST13 : Carbohydrate (chondroitin 4) sulfotransferase 13
• COMT : Catechol-O-methyltransferase
• CYB5R3 : Cytochrome b5 reductase 3
• CYP1A1 : Cytochrome P450, family 1, subfamily A, polypeptide 1
• CYP1A2 : Cytochrome P450, family 1, subfamily A, polypeptide 2
• CYP1B1 : Cytochrome P450, family 1, subfamily B, polypeptide 1
• CYP2A6 : Cytochrome P450, family 2, subfamily A, polypeptide 6
• CYP2A7 : Cytochrome P450, family 2, subfamily A, polypeptide 7
• CYP2A13 : Cytochrome P450, family 2, subfamily A, polypeptide 13
• CYP2B6 : Cytochrome P450, family 2, subfamily B, polypeptide 6
• CYP2C8 : Cytochrome P450, family 2, subfamily C, polypeptide 8
• CYP2C9 : Cytochrome P450, family 2, subfamily C, polypeptide 9
• CYP2C18 : Cytochrome P450, family 2, subfamily C, polypeptide 18
• CYP2C19 : Cytochrome P450, family 2, subfamily C, polypeptide 19
• CYP2D6 : Cytochrome P450, family 2, subfamily D, polypeptide 6
• CYP2D7P1 : Cytochrome P450, family 2, subfamily D, polypeptide 7 pseudogene 1
• CYP2E1 : Cytochrome P450, family 2, subfamily E, polypeptide 1
• CYP2F1 : Cytochrome P450, family 2, subfamily F, polypeptide 1
• CYP2J2 : Cytochrome P450, family 2, subfamily J, polypeptide 2
• CYP2R1 : Cytochrome P450, family 2, subfamily R, polypeptide 1
• CYP2S1 : Cytochrome P450, family 2, subfamily S, polypeptide 1
• CYP2W1 : Cytochrome P450, family 2, subfamily W, polypeptide 1
• CYP3A4 : Cytochrome P450, family 3, subfamily A, polypeptide 4
• CYP3A5 : Cytochrome P450, family 3, subfamily A, polypeptide 5
• CYP3A7 : Cytochrome P450, family 3, subfamily A, polypeptide 7
• CYP3A43 : Cytochrome P450, family 3, subfamily A, polypeptide 43
• CYP4A11 : Cytochrome P450, family 4, subfamily A, polypeptide 11
• CYP4A22 : Cytochrome P450, family 4, subfamily A, polypeptide 22
• CYP4B1 : Cytochrome P450, family 4, subfamily B, polypeptide 1
• CYP4F2 : Cytochrome P450, family 4, subfamily F, polypeptide 2
• CYP4F3 : Cytochrome P450, family 4, subfamily F, polypeptide 3
• CYP4F8 : Cytochrome P450, family 4, subfamily F, polypeptide 8
• CYP4F11 : Cytochrome P450, family 4, subfamily F, polypeptide 11
• CYP4F12 : Cytochrome P450, family 4, subfamily F, polypeptide 12
• CYP4Z1 : Cytochrome P450, family 4, subfamily Z, polypeptide 1
• CYP7A1 : Cytochrome P450, family 7, subfamily A, polypeptide 1
• CYP7B1 : Cytochrome P450, family 7, subfamily B, polypeptide 1
• CYP8B1 : Cytochrome P450, family 8, subfamily B, polypeptide 1
• CYP11A1 : Cytochrome P450, family 11, subfamily A, polypeptide 1
• CYP11B1 : Cytochrome P450, family 11, subfamily B, polypeptide 1
• CYP11B2 : Cytochrome P450, family 11, subfamily B, polypeptide 2
• CYP17A1 : Cytochrome P450, family 17, subfamily A, polypeptide 1
• CYP19A1 : Cytochrome P450, family 19, subfamily A, polypeptide 1
• CYP20A1 : Cytochrome P450, family 20, subfamily A, polypeptide 1
• CYP21A2 : Cytochrome P450, family 21, subfamily A, polypeptide 2
• CYP24A1 : Cytochrome P450, family 24, subfamily A, polypeptide 1
• CYP26A1 : Cytochrome P450, family 26, subfamily A, polypeptide 1
• CYP26B1 : Cytochrome P450, family 26, subfamily B, polypeptide 1
• CYP26C1 : Cytochrome P450, family 26, subfamily C, polypeptide 1
• CYP27A1 : Cytochrome P450, family 27, subfamily A, polypeptide 1
• CYP27B1 : Cytochrome P450, family 27, subfamily B, polypeptide 1
• CYP39A1 : Cytochrome P450, family 39, subfamily A, polypeptide 1
• CYP46A1 : Cytochrome P450, family 46, subfamily A, polypeptide 1
• CYP51A1 : Cytochrome P450, family 51, subfamily A, polypeptide 1
• DDOST : Dolichyl-diphosphooligosaccharide-protein glycosyltransferase subunit (non-catalytic)
• DHRS1 : Dehydrogenase/reductase (SDR family) member 1
• DHRS2 : Dehydrogenase/reductase (SDR family) member 2
• DHRS3 : Dehydrogenase/reductase (SDR family) member 3
• DHRS4 : Dehydrogenase/reductase (SDR family) member 4
• DHRS7 : Dehydrogenase/reductase (SDR family) member 7
• DHRS9 : Dehydrogenase/reductase (SDR family) member 9
• DHRS12 : Dehydrogenase/reductase (SDR family) member 12
• DHRS13 : Dehydrogenase/reductase (SDR family) member 13
• DHRSX : Dehydrogenase/reductase (SDR family) X-linked
• DPEP1 : Dipeptidase 1 (renal)
• DPYD : Dihydropyrimidine dehydrogenase
• EPHX1 : Epoxide hydrolase 1, microsomal (xenobiotic)
• EPHX2 : Epoxide hydrolase 2, microsomal (xenobiotic)
• ESD : Esterase D
• FMO1 : Flavin containing monooxygenase 1
• FMO2 : Flavin containing monooxygenase 2
• FMO3 : Flavin containing monooxygenase 3
• FMO4 : Flavin containing monooxygenase 4
• FMO5 : Flavin containing monooxygenase 5
• FMO6P : Flavin containing monooxygenase 6 pseudogene
• GAL3ST1 : Galactose-3-O-sulfotransferase 1
• GAMT : Guanidinoacetate N-methyltransferase
• GLRX : Glutaredoxin (thioltransferase)
• GLYAT : Glycine-N-acyltransferase
• GNMT : Glycine N-methyltransferase
• GPX1 : Glutathione peroxidase 1
• GPX2 : Glutathione peroxidase 2 (gastrointestinal)
• GPX3 : Glutathione peroxidase 3 (plasma)
• GPX4 : Glutathione peroxidase 4
• GPX5 : Glutathione peroxidase 5
• GPX6 : Glutathione peroxidase 6 (olfactory)
• GPX7 : Glutathione peroxidase 7
• GSR : Glutathione reductase
• GSTA1 : Glutathione S-transferase alpha 1
• GSTA2 : Glutathione S-transferase alpha 2
• GSTA3 : Glutathione S-transferase alpha 3
• GSTA4 : Glutathione S-transferase alpha 4
• GSTA5 : Glutathione S-transferase alpha 5
• GSTCD : Glutathione S-transferase, C-terminaldomain containing
• GSTK1 : Glutathione S-transferase kappa 1
• GSTM1 : Glutathione S-transferase mu 1
• GSTM2 : Glutathione S-transferase mu 2 (muscle)
• GSTM3 : Glutathione S-transferase mu 3 (brain)
• GSTM4 : Glutathione S-transferase mu 4
• GSTM5 : Glutathione S-transferase mu 5
• GSTO1 : Glutathione S-transferase omega 1
• GSTO2 : Glutathione S-transferase omega 2
• GSTP1 : Glutathione S-transferase pi 1
• GSTT1 : Glutathione S-transferase theta 1
• GSTT2 : Glutathione S-transferase theta 2
• GSTZ1 : Glutathione S-transferase zeta 1
• GZMA : Granzyme A (granzyme 1, cytotoxic T-lymphocyte-associated serine esterase 3)
• GZMB : Granzyme B (granzyme 2, cytotoxic T-lymphocyte-associated serine esterase 1)
• HNMT : Histamine N-methyltransferase
• HSD11B1 : Hydroxysteroid (11-beta) dehydrogenase 1
• HSD17B10 : Hydroxysteroid (17-beta) dehydrogenase 10
• HSD17B11 : Hydroxysteroid (17-beta) dehydrogenase 11
• HSD17B14 : Hydroxysteroid (17-beta) dehydrogenase 14
• INMT : Indolethylamine N-methyltransferase
• MAOA : Monoamine oxidase A
• MAOB : monoamine oxidase B
• METAP1 : Methionyl aminopeptidase 1
• MGST1 : Microsomal glutathione S-transferase 1
• MGST2 : Microsomal glutathione S-transferase 1
• MGST3 : Microsomal glutathione S-transferase 3
• NAA20 : N(alpha)-acetyltransferase 20, NatB catalytic subunit
• NAT1 : N-acetyltransferase 1 (arylamine N-acetyltransferase)
• NAT2 : N-acetyltransferase 2 (arylamine N-acetyltransferase)
• NNMT : Nicotinamide N-methyltransferase
• NQO1 : NAD(P)H dehydrogenase, quinone 1
• NQO2 : NAD(P)H dehydrogenase, quinone 2
• PNMT : Phenylethanolamine N-methyltransferase
• PON1 : Paraoxonase 1
• PON2 : Paraoxonase 2
• PON3 : Paraoxonase 3
• POR : P450 (cytochrome) oxidoreductase
• PTGES : Prostaglandin E synthase
• PTGS1 : Prostaglandin-endoperoxide synthase 1 (prostaglandin G/H synthase and cyclooxygenase)
• PTGS2 : Prostaglandin-endoperoxide synthase 2 (prostaglandin G/H synthase and cyclooxygenase)
• SAT1 : Spermidine/spermine N1-acetyltransferase 1
• SMOX : Spermine oxidase
• SOD1 : Superoxide dismutase 1, soluble
• SOD2 : Superoxide dismutase 2, mitochondrial
• SULT1A1 : Sulfotransferase family, cytosolic, 1A, phenol-preferring, member 1
• SULT1A2 : Sulfotransferase family, cytosolic, 1A, phenol-preferring, member 2
• SULT1A3 : Sulfotransferase family, cytosolic, 1A, phenol-preferring, member 3
• SULT1B1 : Sulfotransferase family, cytosolic, 1B, member 1
• SULT1C1 : Sulfotransferase family, cytosolic, 1C, member 1
• SULT1C2 : Sulfotransferase family, cytosolic, 1C, member 2
• SULT1C3 : Sulfotransferase family, cytosolic, 1C, member 3
• SULT1C4 : Sulfotransferase family, cytosolic, 1C, member 4
• SULT1E1 : Sulfotransferase family 1E, estrogen-preferring, member 1
• SULT2A1 : Sulfotransferase family, cytosolic, 2A,dehydroepiandrosterone (DHEA)-preferring, member 1
• SULT2B1 : Sulfotransferase family, cytosolic, 2B, member 1
• SULT4A1 : Sulfotransferase family 4A, member 1
• SULT6B1 : sulfotransferase family, cytosolic, 6B, member 1
• TBXAS1 : Thromboxane A synthase 1 (platelet)
• TPMT : Thiopurine S-methyltransferase
• TST : Thiopurine S-methyltransferase
• UCHL1 : Ubiquitin carboxyl-terminal esterase L1 (ubiquitin thiolesterase)
• UCHL3 : Ubiquitin carboxyl-terminal esterase L3 (ubiquitin thiolesterase)
• UGT1A1 : UDP glucuronosyltransferase 1 family, polypeptide A1
• UGT1A3 : UDP glucuronosyltransferase 1 family, polypeptide A3
• UGT1A4 : UDP glucuronosyltransferase 1 family, polypeptide A4
• UGT1A5 : UDP glucuronosyltransferase 1 family, polypeptide A5
• UGT1A6 : UDP glucuronosyltransferase 1 family, polypeptide A6
• UGT1A7 : UDP glucuronosyltransferase 1 family, polypeptide A7
• UGT1A8 : UDP glucuronosyltransferase 1 family, polypeptide A8
• UGT1A9 : UDP glucuronosyltransferase 1 family, polypeptide A9
• UGT1A10 : UDP glucuronosyltransferase 1 family, polypeptide A10
• UGT2A1 : UDP glucuronosyltransferase 2 family, polypeptide A1, complex locus
• UGT2A3 : UDP glucuronosyltransferase 2 family, polypeptide A3
• UGT2B10 : UDP glucuronosyltransferase 2 family, polypeptide B10
• UGT2B11 : UDP glucuronosyltransferase 2 family, polypeptide B11
• UGT2B15 : UDP glucuronosyltransferase 2 family, polypeptide B15
• UGT2B17 : UDP glucuronosyltransferase 2 family, polypeptide B17
• UGT2B28 : UDP glucuronosyltransferase 2 family, polypeptide B28
• UGT2B4 : UDP glucuronosyltransferase 2 family, polypeptide B4
• UGT2B7 : UDP glucuronosyltransferase 2 family, polypeptide B7
• UGT3A1 : UDP glycosyltransferase 3 family, polypeptide A1
• UGT8 : UDP glycosyltransferase 8
• XDH : Xanthine dehydrogenase