critical care case study example

• FCUS Courses
• Register for FCUS Course

• What is the ICN?
• The ICN Story
• ICN Hot cases
• Semantic sMatter
• ECG Proving Ground
• ICU Radiology Cases
• Game Changing Evidence
• Pharmacology
• Past Papers
• Approaches to Questions
• Clinical Exam
• Clinical Governance
• Past Papers with Answers
• The ICN meta feed
• Conferences
• Websites & Blogs
• Books & Journals
• Clinical Calculators

Clinical Cases

This section is a collection of critical care clinical cases to test yourself and hopefully get some new ideas.

Please leave feedback and comments, and if you want to put your own hot cases up, please get in touch and we can make it happen.

Can you work out the cause of this DVT?

Pulmonary Artery Catheters: PAC traps for young snake charmers

End-o-bed-o-gram – Case 3

Spot Diagnosis #2

End-o-bed-o-gram – case 2.

The NEJM Critical Care Challenge: Case 11 (Answer)

Spot diagnosis #1, the nejm critical care challenge: case 11 (end of life care), nejm critical care challenge case 10 (sdh) answer, the nejm critical care challenge: case 10 (acute sdh), icu-acquired weakness: nejm critical care challenge case 9 (question and answer).

Nutrition: NEJM Critical Care Challenge Case 8 Answer

End-o-bed-o-gram 1, the nejm critical care challenge: case 8, coagulopathy: nejm critical care challenge case 7 answer, coagulopathy: nejm critical care challenge case 7, icu sedation: nejm critical care challenge case 6 answer, icu sedation: nejm critical care challenge case 6, acute liver failure: nejm critical care challenge case 5 answer, acute liver failure: nejm critical care challenge case 5, ventilation: nejm critical care challenge case 4 answer.

Haemodynamic Monitoring: NEJM Critical Care Challenge Case 3 ANSWER

Haemodynamic monitoring: the nejm critical care challenge case 3, resuscitation fluid: nejm critical care challenge case 2 answer.

The NEJM Critical Care Challenge: Case 2

Septic shock: nejm critical care challenge case 1 answer, septic shock: nejm critical care challenge case 1.

The NEJM Critical Care Challenge: Case 1

Hot Case #14 – I’m pregnant … HELLP!

Hot Case #13 – To bust or not to bust?

ICN Hot Case #12 – Pull my finger!

Hot Case #11 – Testing testing

ICN Hot Case #10

ICN Hot Case #9

ICN Hot Case #8

ICN Hot Case #7

ICN Hot Case # 6

The nejm critical care challenge: case 4.

ICN Hot Case #5

September Case of the month

ICN Hot Case #4

ICN Hot Case #3

•   What is ICN
•   The Team
•   ICN NSW
•   ICN QLD
•   ICN VIC
•   ICN WA
•   ICN NZ
•   ICN UK
•   Lung US
•   Exam Help
•   Clinical Cases
•   Echo Cases
•   Echo Guide
•   ICU Radiology
•   Game Changing Evidence
•   ICN Metafeed
•   Simulation Resouces
•   SMACC Posters
•   Audio
•   Video
•   Pecha Kuchas
•   End-o-Bed-o-Gram
•   ICU Primary Exam
•   CICM Fellowship
•   ANZCA Fellowship
•   Paeds Fellowship
•   Emergency Primary
•   Anaesthetics Primary

® 2024  The Intensive Care Network  || All rights reserved ||  Disclaimer ||  Site Map  ||  Contact ICN Support

Log in with your credentials

or      Create an account

Forgot your details?

Create account.

CritCases is a collaboration between Dr. Mike Betzner , director of STARS air ambulance service and EM Cases.

There are many approaches and therapeutic options in critical care that do not have a strong evidence base to guide us. We use our clinical judgement, gestalt and basic principles to make rapid decisions. This blog explores therapeutic options to various challenging critical care scenarios through an interactive case-based, Q&A framework. The case unfolds in a step-wise manner, as it would in real practice, and integrates opinions from critical care providers, not only from STARS, but from around the world. Each blog includes an open peer review from an expert on the given topic. Welcome to CritCases!

Crit Cases 14 Bronchopleural Fistula Management

Dr. Mike Misch guides us through this part 2 of a blunt chest trauma case with a presumed bronchopulmonary fistula requiring 3 chest tubes and describes options to optimize one-lung ventilation for safe transport...

CritCases 13 Shock and Hypoxia in Blunt Chest Trauma

In this CritCases blog, Shock and Hypoxia in Blunt Chest Trauma, a collaboration between STARS Air Ambulance Service, Mike Betzner and EM Cases, Mike Misch guides us through a hairy thoracic trauma case, reviewing principles of trauma resuscitation, airway considerations, tension pneumothorax management and a rare and challenging trauma diagnosis...

CritCases 12 Accidental Hypothermia and Cardiac Arrest

In this CritCases blog Michael Misch takes us through a case of accidental hypothermia and cardiac arrest, reviewing the controversies in management as well as the guidelines for rewarming, the role of ECMO and the alterations to ACLS cardiac arrest medications, CPR and defibrillations...

CritCases 11 LVAD Management in the GI Bleed Patient

In this CritCases blog Mike Misch presents a case of a patient with a left ventricular assist device (LVAD) with a postoperative acute lower GI bleed and answers practical questions such as:  How do you measure blood pressure in a patient with an LVAD? What are the common complications of LVADs that we must be aware of? What information can the LVAD controller provide? Why are LVAD patients at high risk for bleeding? and many more...

CritCases 10 Hyponatremia Associated Seizures

In this EM Cases CritCases blog - a collaboration between STARS Air Ambulance Service, Mike Betzner and EM Cases, a middle aged woman presents to a rural ED with headache and vomiting, normal vital signs with subsequent status epilepticus and serum sodium of 110 mmol/L. What management recommendations would you make to the rural ED physician, the transport team and in your ED with regards to treatment of seizures, safe correction of hyponatremia, airway management, search for underlying cause and prevention of Osmotic Demyelenation Syndrome?

CritCases 9 Pre-Eclampsia and Preterm Labor – Time Sensitive Management

In this CritCases blog we present a case of a 30 week gestational age pregnant woman with high blood pressure, headache, blurry vision and pelvic cramping. We discuss the management challenges of transporting a patient with severe pre-eclampsia and preterm labor, with special attention to dosing of magnesium, antihypertensive agents choice, and indications for steroids, tocolytics and antibiotics.

Support EM Cases

Subscribe to, our newsletter.

Unlock exclusive free content and resources by subscribing to our newsletter now!

• Case report
• Open access
• Published: 03 May 2021

Severe COVID-19 in the intensive care unit: a case series

• Hori Hariyanto   ORCID: orcid.org/0000-0001-6746-4406 1 , 3 ,
• Corry Quando Yahya 2 &
• Ronald Christian Agustinus Aritonang 1 , 3  

Journal of Medical Case Reports volume  15 , Article number:  259 ( 2021 ) Cite this article

3944 Accesses

1 Citations

7 Altmetric

Metrics details

Coronavirus disease 2019 (COVID-19) was first identified in Indonesia in March 2020, and the number of infections has grown exponentially. The situation is at its worst, overwhelming intensive care unit (ICU) resources and capacity.

Case presentation

This is a single-center observational case study of 21 confirmed COVID-19 patients admitted to the ICU from March 20, 2020, to April 31, 2020. Demographics, baseline comorbidities, clinical symptoms, laboratory tests, electrocardiogram (ECG) and chest imaging were obtained consecutively during patient care. We identified 21 patients with confirmed COVID-19 severe infection in our ICU. The mean (± standard deviation) age of the patients was 54 ± 10 years; 95% were men, with shortness of breath (90.6%) the most common symptom. Hypertension was identified as a comorbidity in 28.6% of patients. The most common reason for admission to the ICU was hypoxemic respiratory failure, with 80% (17 patients) requiring mechanical ventilation. Half of the patients (10) died between day 1 and day 18, with septic shock as the primary cause of death. Of the 11 surviving patients, five were discharged home, while six were discharged from the ICU but remained in the hospital ward. Even then, the median length of ICU stay amongst survivors was 18 days.

Conclusions

To date, there are no known effective antiviral agents or specific therapy to treat COVID-19. As severe systemic inflammatory response and multiple organ failure seems to be the primary cause of death, supportive care in maintaining oxygenation and hemodynamic stability remain the mainstay goals in treating critically ill COVID-19 patients.

Peer Review reports

Coronavirus disease 2019 (COVID-19) has spread from a single city to the entire globe with alarming speed. Arising from China, this virus has expanded rapidly to all parts of the world, knowing no geopolitical boundaries in infecting the human population. The first case of COVID-19 in Indonesia was identified in March 2020. Since then, the number of cases in Indonesia has grown exponentially; as of October 7, 2020, there had been 315,714 confirmed COVID-19 cases and 11,472 deaths [ 1 ]. While most patients with COVID-19 are asymptomatic or experience only mild symptoms, some individuals develop acute respiratory distress syndrome (ARDS) requiring mechanical ventilation, while some succumb to septic shock. Reports describing patients admitted to the intensive care unit (ICU) in Indonesia are sparse; therefore, it is our aim to share our early experience of COVID-19 pandemic care amongst ICU patients.

Study design and participants

This is a single-center observational case series study. All patients completed an informed consent form that was approved by the Ethical Committee at Siloam Hospital Kelapa Dua (Study protocol: 19-03-0317). Data were collected consecutively during admission. Enrollment included all patients admitted to the ICU starting with the first patient in March 20, 2020 up to April 31, 2020. All 21 cases enrolled in this study were confirmed COVID-19 from double-gene polymerase chain reaction (PCR) detection of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) using a nasopharyngeal swab in line with the diagnostic criteria guideline established by the Indonesian Ministry of Health.

Data collection

Demographics, baseline comorbidities, clinical symptoms, laboratory tests, chest imaging and electrocardiogram (ECG) changes were obtained consecutively during patient visits to the ICU. Diagnoses during the hospital course, inpatient medications, treatments including invasive mechanical ventilation and kidney replacement therapy, and outcomes including length of stay, discharge and mortality were also recorded. To quantify the extent of infection, a severity score was calculated using the CURB-65 [confusion, urea, respiratory rate, blood pressure, and 65 years of age or older] pneumonia risk score and Acute Physiology Assessment and Chronic Health Evaluation II (APACHE II) score.

Statistical analysis

Variables are reported as frequency, percentage (%), mean (SD) if they were normally distributed, and median with range (min–max) for non-normal distribution. Laboratory results are presented as actual data, and all data analysis was carried out using STATA version 12 software (StataCorp LLC, College Station, TX, USA).

Patient characteristics

During the period from March 20, 2020, through April 31, 2020, we identified 21 critically ill patients with confirmed COVID-19 infection admitted to the ICU. The demographic and clinical characteristics of the patients are shown in Table 1 . The mean (± SD) age of the patients was 54 ± 10 years (range 31–79); 20 (95%) were male and one (4.8%) was female. The mean duration of symptoms before hospital admission was 8 ± 3 days. All patients were Indonesian citizens of Malay ethnicity, and none had recently traveled to a country with known transmission such as China, South Korea, Iran or Italy. However, the majority of patients confirmed recent contact exposure from various cluster sites including family and religious gatherings. Comorbidities of patients in this critically ill population included diabetes 1 (4.8%), hypertension 6 (28.6%) and cerebrovascular disease 1 (4.8%). One (4.8%) patient was documented to be a former smoker, and another patient (4.8%) had chronic obstructive pulmonary disease.

Symptoms presented upon admission included fever [18 (85.7%) of 21 patients], cough [18 (85.7%)] and shortness of breath [19 (90.4%)]. Other symptoms reported were fatigue [3 (14.2%)], sore throat [2 (9.5%)] and myalgia [2 (9.4%)]. Upon admission, the mean APACHE score was 10–14 in seven patients (33.3%), 15–19 in 10 (47.6%), 20–24 in two (9.5%) and greater than 25 in two (9.5%). The mean CURB-65 score was 0 was nine patients (42.9%); 1 in nine patients (42.9%) and 2 in three patients (14.3%).

In this study, all patients received hydroxychloroquine, azithromycin, meropenem and antifungal prophylaxis; eight patients (38%) received compassionate-use tocilizumab, and no patients received systemic steroids. Thromboprophylaxis was given with heparin 250 U/hour, intravenously.

Laboratory findings

Table 2 shows the laboratory and radiologic findings of patients upon admission to the ICU. On admission, lymphocytopenia was common (in 86% of the patients), with a mean leukocyte count of 11.056 ± 6.604 × 10 3 /μL and low median lymphocyte count of 13.5% (interquartile range 1–19%). Inflammation markers including erythrocyte sedimentation rate (ESR), C-reactive protein (CRP) and lactate dehydrogenase were also measured, and all values were dramatically elevated. Mean lactate dehydrogenase was uniformly elevated at 951 ± 140, along with mean CRP level of 217 ± 122. Hepatic alanine aspartate enzyme was 40 U/L or higher in all patients.

Chest radiographs were obtained in all 21 patients, all of which showed bilateral pulmonary opacities, while pleural effusion was seen in 12 (57.1%) of the patients (Fig. 1 ). A computed tomography (CT) scan of the chest was obtained in six patients (29%); five of which showed bilateral ground glass opacities and one consolidation. Overall, 17 patients progressed to respiratory distress and required mechanical ventilation, while the other four were discharged to the ward after a mean of 13 days in the ICU.

Chest films of severe COVID-19 patients upon admission to the intensive care unit

• Respiratory failure

Seventeen patients (80.9%) received invasive mechanical ventilation, as their ratios of arterial oxygen partial pressure to fraction of inspired oxygen [PaO 2 :FiO 2 (p/f ratio)] were consistent with severe acute respiratory distress syndrome (ARDS): mean p/f ratio 100 ± 36. The time to initiation of mechanical ventilation was 4 ± 3 days, and all patients were placed in the prone position starting day 2 of mechanical ventilation.

The median FiO 2 on day 1 of mechanical ventilation was 0.9 (interquartile range 0.7–1.0); on day 3, median FiO 2 was 0.6 (interquartile range 0.5–0.7), and on day 5 median FiO 2 was 0.4 (interquartile range 0.35–0.55). The median driving pressure [the difference between plateau pressure and positive end-expiratory pressure (PEEP)] on day 1 of mechanical ventilation was 23 ± 5 cmH 2 O, with median pulmonary compliance of 20 mL/cmH 2 O (interquartile range, 13–27). Initial PEEP was set at 11 ± 2 cmH 2 O. Throughout 5 days of mechanical ventilation, the median driving pressure was gradually lowered to 15 ± 3 cmH 2 O, pulmonary compliance improved to 42 mL/cmH 2 O (interquartile range, 28–52), and PEEP was maintained at 9 ± 1 cmH 2 O. The mean p/f ratio was 150 ± 62 on day 1, 193 ± 112 on day 3, and 235 ± 109 on day 5. Out of 17 patients, two (13%) developed progressive ARDS and died. Seven (41%) patients survived, with a mean duration of mechanical ventilation of 10 ± 4.8 days. Amongst these, one underwent bronchoscopy due to atelectasis; three encountered pneumothorax, and two underwent tracheostomy due to difficulty in weaning and prolonged mechanical ventilation support (greater than 20 days of mechanical ventilation).

Twelve patients (75%) presented with concurrent hypotension requiring vasopressors without clear evidence of secondary infection. Of these patients, three (18%) had transient hypotension after intubation; nine (56%) had hypotension that was unrelated to intubation or that persisted for more than 12 hours after intubation. Six patients (38%) developed septic shock and died; one (6%) experienced cardiac arrest upon prone positioning, and another patient (6%) experienced cardiac arrest due to intractable hyperkalemia and persistent acidosis, despite undergoing hemodialysis.

As of May 31, out of the 21 patients cared for in the ICU, 10 (47%) had died and 11 survived, with six (23%) patients who had been discharged from the ICU but remained in the hospital and five (23%) who had been discharged from the hospital (Fig. 2 ). The median length of ICU stay among survivors was 18 days (interquartile range, 7–36), while the median length of ward stay after ICU discharge was 11 days (interquartile range, 7–25). Fitness for discharge was based on the absence of fever for at least 7 days, improvement in chest radiograph and negative nasopharyngeal PCR test.

Duration of therapy amongst 11 intensive care unit survivors of severe COVID-19. LOS length of stay

Discussion and conclusion

The majority of patients admitted to our ICU were men, with a mean age of 54 ± 10 years, and had hypertension as a comorbidity. Clinical manifestations were fever, cough and shortness of breath. No gastrointestinal, renal or cerebrovascular manifestations were documented in our study. All 21 patients had abnormal blood test results with elevated CRP and liver enzymes, decreased lymphocytes, increased D-dimer and coagulation abnormalities, all of which were similar to reports from China [ 2 , 3 ]. Six chest CT scans were performed showing ground-glass opacities and/or consolidation similar to other reports [ 4 ].

Recent studies have highlighted two phenotypes in COVID-19 pneumonia. The L-type lung is characterized by normal compliance, low ventilation-to-perfusion ratio and low lung weight. Over time, the lungs may either improve or evolve into an H-type pneumonia characterized by low compliance, high right-to-left shunt and increasing pulmonary edema, which contribute to the deadly cycle of hypoxemia and strain on body organs [ 5 ]. In this report, the majority [18 (85.8%)] of the 21 patients had an admission CURB score of 0–1. Nevertheless, more than half progressed to severe ARDS and respiratory failure as evidenced by hypoxemia, progressive bilateral infiltrates and decreased respiratory system compliance (H-type COVID-19 pneumonia). Out of 17 patients receiving mechanical ventilation, two rapidly progressed to severe ARDS and died.

High-flow nasal cannula was initially used to improve oxygenation, but promptly escalated to mechanical ventilation once increased work of breathing was observed. Notably, high oxygen requirements and poor lung compliance were observed soon after initiation of mechanical ventilation. In severe ARDS, damage to type II alveolar cells not only renders surfactant inactive, but these edematous alveoli also compress alveoli in dependent regions, thereby contributing to alveolar collapse [ 6 ]. Prone positioning has the benefit of reopening collapsed alveoli, as the heart rests on the sternum and exerts less pressure on the pleura and lung [ 7 ]. This together with the lung recruitment maneuver opens the dorsal parts of the lung and allows more homogeneous ventilation and perfusion [ 8 ]. Therefore, a high initial PEEP (10-12 cmH 2 O) was given and patients were placed in the prone position for 6 hours per day. Prone positioning started on day 2 of mechanical ventilation, and an increased p/f ratio was observed from day 3 onwards.

Early in the clinical course, sputum production was minimal and sterile. As mechanical ventilation continued, coexisting lower respiratory bacterial infections were identified, further complicating the course of disease and resulting in longer ICU stays. Seven patients survived, with two encountering pneumothorax and placed on tracheostomy due to prolonged ventilator support, while the other five were successfully liberated from mechanical ventilation without any long-term sequelae. Even then, the median ICU stay among the survivors was a lengthy 18 days (interquartile range, 7–36 days).

In this study, all 21 patients received hydroxychloroquine, azithromycin, meropenem and antifungal prophylaxis, with eight patients (38%) receiving compassionate-use tocilizumab. Unfortunately, one of our patients experienced Torsades de pointes and died. Such fatal arrhythmia may have been caused by the direct effect of hydroxychloroquine and azithromycin on ventricular repolarization, thus prolonging the QT interval [ 9 ]. Hence, hydroxychloroquine and azithromycin use was terminated halfway through the course of ICU care. None of our patients received steroids, as studies during that time were inconclusive for the use of systemic glucocorticoids.

Upon admission, the majority of our patients had an APACHE score of 10–19 (mortality score of 12–22%); nevertheless, six (35%) of 17 patients who received mechanical ventilation died due to septic shock. Symptoms were similar to septic shock caused by bacterial infections, but one distinctive difference was the deterioration that occurred within a very short time (< 24 hours). This might be attributable to the massive explosive release of viral antigens, thus creating a violent inflammatory response and sudden hemodynamic collapse, as others have speculated [ 10 , 11 ]. Taken together, this suggests that no severity scores seem to aid in predicting the future course and prognosis of COVID-19 infection.

To date, there are still no solid markers for predicting disease progression, and various treatments with immunomodulators, antivirals and interleukin inhibitors are given with hopes of halting the progression of the disease, but no consensus guidelines have yet been developed. To make matters worse, this virus possesses remarkable mimicry capability, as it displays atypical presentation ranging from gastrointestinal symptoms, neurologic complications, antiphospholipid syndrome and acute myocardial injury to fatal ventricular arrhythmia [ 12 , 13 , 14 , 15 ], all of which may lead to a false diagnosis, delay treatment and postpone isolation measures within a community.

COVID-19 has emerged as a complex disease that appears to have many “faces.” Despite evidence of extensive damage both in radiologic and laboratory findings, the clinical presentation does not always seem to conform. In the midst of this pandemic, we would like to share our experience of caring for those with the greatest severity of illness: the ICU population. We understand the limitations of our study relating to its small sample size and limited laboratory investigations. However, our experience in caring for these patients has reminded us that supportive therapy remains the hallmark in fighting this self-limiting disease. Until new evidence becomes available, physicians can expect mechanical ventilation to be a lengthy journey, with bacterial co-infections, sepsis and pneumothorax encountered along the course of ICU stay.

Availability of data and materials

Please contact the author for data requests.

Abbreviations

Coronavirus disease 2019

Intensive care unit

Standard deviation

World Health Organization

Acute respiratory distress syndrome

Polymerase chain reaction

Confusion, urea, respiratory rate, blood pressure, and 65 years of age or older pneumonia risk score

Acute Physiology and Chronic Health Evaluation II

Erythrocyte sedimentation rate

C-Reactive protein

Computed tomography

PaO 2 :FiO 2

Fraction of inspired oxygen

Positive end-expiratory pressure

World Health Organization (WHO). Coronavirus Disease (COVID-19) Situation Reports. 2020.

Yang X, Yu Y, Xu J, Shu H, Ja X, Liu H, et al. Clinical course and outcomes of critically ill patients with SARS-CoV-2 pneumonia in Wuhan, China: a single-centered, retrospective, observational study. Lancet Respir Med. 2020;8(5):475–81.

Article   CAS   Google Scholar  

Wang D, Hu B, Hu C, Zhu F, Liu X, Zhang J, et al. Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan China. JAMA. 2020;323(11):1061–9.

Shi H, Han X, Jiang N, Cao Y, Alwalid O, Gu J, et al. Radiological findings from 81 patients with COVID-19 pneumonia in Wuhan, China: a descriptive study. Lancet Infect Dis. 2020;20(4):425–34.

Gattinoni L, Chiumello D, Caironi P, Busana M, Romitti F, Brazzi L, et al. COVID-19 pneumonia: different respiratory treatments for different phenotypes? Intensive Care Med. 2020;46(6):1099–102.

Gibson PG, Qin L, Puah SH. COVID-19 acute respiratory distress syndrome (ARDS): clinical features and differences from typical pre-COVID-19 ARDS. Med J Aust. 2020;213(2):54-6.e1.

Article   Google Scholar  

Gattinoni L, Pesenti A, Carlesso E. Body position changes redistribute lung computed-tomographic density in patients with acute respiratory failure: impact and clinical fallout through the following 20 years. Intensive Care Med. 2013;39(11):1909–15.

Scholten EL, Beitler JR, Prisk GK, Malhotra A. Treatment of ARDS with prone positioning. Chest. 2017;151(1):215–24.

Jankelson L, Karam G, Becker ML, Chinitz LA, Tsai MC. QT prolongation, torsades de pointes, and sudden death with short courses of chloroquine or hydroxychloroquine as used in COVID-19: a systematic review. Heart Rhythm. 2020;17(9):1472–9.

Coz Yataco AO, Simpson SQ. Coronavirus Disease 2019 Sepsis: a nudge toward antibiotic stewardship. Chest.

Li H, Liu L, Zhang D, Xu J, Dai H, Tang N, et al. SARS-CoV-2 and viral sepsis: observations and hypotheses. Lancet. 2020;395(10235):1517–20.

The LR. High-stakes heterogeneity in COVID-19. Lancet Rheumatol. 2020;2(10):e577.

Zhang Y, Xiao M, Zhang S, Xia P, Cao W, Jiang W, et al. Coagulopathy and antiphospholipid antibodies in patients with COVID-19. N Engl J Med. 2020;382(17):e38.

Villapol S. Gastrointestinal symptoms associated with COVID-19: impact on the gut microbiome. Transl Res. 2020.

Fifi JT, Mocco J. COVID-19 related stroke in young individuals. Lancet Neurol. 2020;19(9):713–5.

Download references

Acknowledgements

Not applicable.

None received.

Author information

Authors and affiliations.

Faculty of Medicine, Department of Anesthesiology and Intensive Care, Universitas Pelita Harapan, Jl. M. H. Thamrin Boulevard 1100, Lippo Village Tangerang, Tangerang, Banten, 15811, Indonesia

Hori Hariyanto & Ronald Christian Agustinus Aritonang

Faculty of Medicine, Universitas Indonesia, Jalan Diponegoro No 77, Jakarta Pusat, 10430, Indonesia

Corry Quando Yahya

Siloam Hospitals Kelapa Dua, Jl. Kelapa Dua Raya No.1001, Kelapa Dua, Tangerang, Banten, 15810, Indonesia

You can also search for this author in PubMed   Google Scholar

Contributions

HH and CY contributed to the conceptualization, data curation, formal analysis and investigation of the patients. RA contributed to data curation, formal analysis, investigation, project administration and resources. CY and HH contributed in writing this report and coordinated to draft the manuscript. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Hori Hariyanto .

Ethics declarations

Ethics approval.

All patients completed an informed consent form that was approved by the Ethical Committee at Siloam Hospitals, Kelapa Dua (Study protocol: 19-03-0317).

Consent for publication

Written informed consent was obtained from the patients’ or from the next of kin (in case the patient deceased) for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's note.

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

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.

Hariyanto, H., Yahya, C.Q. & Aritonang, R.C.A. Severe COVID-19 in the intensive care unit: a case series . J Med Case Reports 15 , 259 (2021). https://doi.org/10.1186/s13256-021-02799-1

Download citation

Received : 06 November 2020

Accepted : 18 March 2021

Published : 03 May 2021

DOI : https://doi.org/10.1186/s13256-021-02799-1

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

• Septic shock
• Mechanical ventilation
• Severe infection

Journal of Medical Case Reports

ISSN: 1752-1947

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

LOGIN 

Annual Report

• Board of Directors
• Nomination Process
• Organizational Structure
• ATS Policies
• ATS Website
• MyATS Tutorial
• ATS Experts
• Press Releases

Member Newsletters

• ATS in the News
• ATS Conference News
• Embargo Policy

ATS Social Media

Breathe easy podcasts, ethics & coi, health equity, industry resources.

• Value of Collaboration
• Corporate Members
• Advertising Opportunities
• Clinical Trials
• Financial Disclosure

In Memoriam

Global health.

• International Trainee Scholarships (ITS)
• MECOR Program
• Forum of International Respiratory Societies (FIRS)
• 2019 Latin American Critical Care Conference

Peer Organizations

Careers at ats, affordable care act, ats comments and testimony, forum of international respiratory societies, tobacco control, tuberculosis, washington letter.

• Clinical Resources
• ATS Quick Hits
• Asthma Center

Best of ATS Video Lecture Series

• Coronavirus
• Critical Care
• Disaster Related Resources
• Disease Related Resources
• Resources for Patients
• Resources for Practices
• Vaccine Resource Center
• Career Development
• Resident & Medical Students
• Junior Faculty
• Training Program Directors
• ATS Reading List
• ATS Scholarships
• ATS Virtual Network

ATS Podcasts

• ATS Webinars
• Professional Accreditation

Pulmonary Function Testing (PFT)

• Calendar of Events

Patient Resources

• Asthma Today
• Breathing in America
• Fact Sheets: A-Z
• Fact Sheets: Topic Specific
• Patient Videos
• Other Patient Resources

Lung Disease Week

Public advisory roundtable.

• PAR Publications
• PAR at the ATS Conference

Assemblies & Sections

• Abstract Scholarships
• ATS Mentoring Programs
• ATS Official Documents
• ATS Interest Groups
• Genetics and Genomics
• Medical Education
• Terrorism and Inhalation Disasters
• Allergy, Immunology & Inflammation
• Behavioral Science and Health Services Research
• Clinical Problems
• Environmental, Occupational & Population Health
• Pulmonary Circulation
• Pulmonary Infections and Tuberculosis
• Pulmonary Rehabilitation
• Respiratory Cell & Molecular Biology
• Respiratory Structure & Function
• Sleep & Respiratory Neurobiology
• Thoracic Oncology
• Joint ATS/CHEST Clinical Practice Committee
• Clinicians Advisory
• Council of Chapter Representatives
• Documents Development and Implementation
• Drug/Device Discovery and Development
• Environmental Health Policy
• Ethics and Conflict of Interest
• Health Equity and Diversity Committee
• Health Policy
• International Conference Committee
• International Health
• Members In Transition and Training
• View more...
• Membership Benefits
• Categories & Fees
• Special Membership Programs
• Renew Your Membership
• Update Your Profile
• ATS DocMatter Community
• Respiratory Medicine Book Series
• Elizabeth A. Rich, MD Award
• Member Directory
• ATS Career Center
• Welcome Trainees
• ATS Wellness
• Thoracic Society Chapters
• Chapter Publications
• CME Sponsorship

Corporate Membership

Clinical cases, professionals.

• Respiratory Health Awards
• Clinicians Chat
• Ethics and COI
• Pulmonary Function Testing
• ATS Resources
• Live from the CCD
• Pediatric Division Directors

Reviewed By Behavioral Science Assembly

Submitted by

Lokesh Venkateshaiah, MD

Division of Pulmonary, Critical Care and Sleep Medicine

The MetroHealth System, Case Western Reserve University

Cleveland, Ohio

Bruce Arthur, MD

J. Daryl Thornton, MD, MPH

Assistant Professor

Division of Pulmonary, Critical Care and Sleep Medicine, Center for Reducing Health Disparities

Submit your comments to the author(s).

A 60-year-old man presented to the emergency department complaining of persistent right-sided chest pain and cough. The chest pain was pleuritic in nature and had been present for the last month. The associated cough was productive of yellow sputum without hemoptysis. He had unintentionally lost approximately 30 pounds over the last 6 months and had nightly sweats. He had denied fevers, chills, myalgias or vomiting. He also denied sick contacts or a recent travel history. He recalled childhood exposures to persons afflicted with tuberculosis. 

The patient smoked one pack of cigarettes daily for the past 50 years and denied recreational drug use. He reported ingesting twelve beers daily and had had delirium tremens, remote right-sided rib fractures and a wrist fracture as a result of alcohol consumption. He had worked in the steel mills but had discontinued a few years previously. He collected coins and cleaned them with mercury. 

The patient’s past medical history was remarkable for chronic “shakes” of the upper extremities for which he had not sought medical attention. Other than daily multivitamin tablets, he took no regular medications. 

Hospital course  He was initially admitted to the general medical floor for treatment of community-acquired pneumonia (see Figure 1) and for the prevention of delirium tremens. He was initiated on ceftriaxone, azithromycin, thiamine and folic acid. Diazepam was initiated and titrated using the Clinical Institute Withdrawal Assessment for Alcohol Scale (CIWAS-Ar), a measure of withdrawal severity (1).  By hospital day 5, his respiratory status continued to worsen, requiring transfer to the intensive care unit (ICU) for hypoxemic respiratory failure. His neurologic status had also significantly deteriorated with worsening confusion, memory loss, drowsiness, visual hallucinations (patient started seeing worms) and worsening upper extremity tremors without generalized tremulousness despite receiving increased doses of benzodiazepines.

Physical Exam

White blood cell count was 11,000/mm 3 with 38% neutrophils, 8% lymphocytes, 18 % monocytes and 35% bands

Hematocrit 33%

Platelet count was 187,000/mm 3

Serum sodium was 125 mmol/L, potassium 3 mmol/L, chloride 91 mmol/L, bicarbonate 21 mmol/L, blood urea nitrogen 14 mg /dl, serum creatinine  0.6 mg/dl and anion gap of 14.

Urine sodium <10 mmol/L, urine osmolality 630 mosm/kg

Liver function tests revealed albumin 2.1 with total protein 4.6, normal total bilirubin, aspartate transaminase (AST) 49, Alanine transaminase (ALT) 19 and alkaline phosphatase 47.

Three sputum samples were negative for acid-fast bacilli (AFB).

Bronchoalveolar lavage (BAL) white blood cell count 28 cells/µl, red blood cell count 51 cells/µl, negative for AFB and negative Legionella culture.  BAL gram stain was without organisms or polymorphonuclear leukocytes.

Blood cultures were negative for growth.

Sputum cultures showed moderate growth of Pasteurella multocida.

2D transthoracic ECHO of the heart showed normal valves and an ejection fraction of 65% with a normal left ventricular end-diastolic pressure and normal left atrial size.  No vegetations were noted.

Purified protein derivative (PPD) administered via Mantoux testing was 8 mm in size at 72 hr after placement.

Human immunodeficiency virus (HIV) serology was negative. 

Arterial blood gas (ABG) analysis performed on room air on presentation to the ICU: pH 7.49, PaCO 2 29 mm Hg, PaO 2 49 mm Hg.

After admission to the ICU, the patient was noted to be in acute lung injury (ALI), a subset of acute respiratory distress syndrome (ARDS). The diagnosis of ALI requires all three of the following:  (a) bilateral pulmonary infiltrates, (b) a PaO 2 :FiO 2 ratio of ≤ 300 and (c) echocardiographic evidence of normal left atrial pressure or pulmonary-artery wedge pressure of ≤ 18 mm Hg (2). 

While patients with ALI and ARDS can be maintained with pressure-limited or volume-limited modes of ventilation, only volume assist-control ventilation was utilized in the ARDS Network multicenter randomized controlled trial that demonstrated a mortality benefit.

Noninvasive ventilation has not been demonstrated to be superior to endotracheal intubation in the treatment of ARDS or ALI and is not currently recommended (4).

This is a case of heavy metal poisoning with mercury.  The patient used mercury to clean coins.  Family members who had visited his house while he was hospitalized found several jars of mercury throughout his home.  The Environmental Protection Agency (EPA) was notified and visited the home.  They found aerosolized mercury levels of > 50,000 PPM and had the home immediately demolished. 

Alcoholic hallucinosis is a rare disorder occurring in 0.4 - 0.7% of alcohol-dependent inpatients (5).  Affected persons experience predominantly auditory but occasionally visual hallucinations.  Delusions of persecution may also occur.  However, in contrast to alcohol delirium, other alcohol withdrawal symptoms are not present and the sensorium is generally unaffected.

Delerium tremens (DT) occurs in approximately 5% of patients who withdraw from alcohol and is associated with a 5% mortality rate. DT typically occurs between 48 and 96 hr following the last drink and lasts 1-5 days.  DT is manifested by generalized alteration of the sensorium with vital sign abnormalities.  Death often results from arrhythmias, pneumonia, pancreatitis or failure to identify another underlying problem (6).  While DT certainly could have coexisted in this patient, an important initial step in the management of DT is to identify and treat alternative diagnoses.

Delirium is frequent among older patients in the ICU (7), and may be complicated by pneumonia and sepsis.  However, pneumonia and sepsis as causes for delirium are diagnoses of exclusion and should only be attributed after other possibilities have been ruled out. 

Frontal lobe stroke is unlikely, given the absence of other findings in the history or physical examination present to suggest an acute cerebrovascular event. 

In 1818, Dr. John Pearson coined the term erethism for the characteristic personality changes attributed to mercury poisoning (8).  Erethism is classically the first symptom in chronic mercury poisoning (9).  It is a peculiar form of timidity most evident in the presence of strangers and closely resembles an induced paranoid state.  In the past, when mercury was used in making top hats, the term “mad as a hatter” was used to describe the psychiatric manifestations of mercury intoxication.  Other neurologic manifestations include tremors, especially in patients with a history of alcoholism, memory loss, drowsiness and lethargy.  All of these were present in this patient. 

Acute respiratory failure (ALI/ARDS) can occur following exposure to inhalation of mercury fumes (10). Mercury poisoning has also been associated with acute kidney injury (11). 

Although all of the options mentioned above could possibly contribute to the development of delirium, only mercury poisoning would explain the constellation of findings of confusion, upper extremity tremors, visual hallucinations, somnolence and acute respiratory failure (ALI/ARDS).

Knowledge of the form of mercury absorbed is helpful in the management of such patients, as each has its own distinct characteristics and toxicity. There are three types of mercury: elemental, organic and inorganic. This patient had exposure to elemental mercury from broken thermometers. 

Elemental mercury is one of only two known metals that are liquid at room temperature and has been referred to as quicksilver (12). It is commonly found in thermometers, sphygmomanometers, barometers, electronics, latex paint, light bulbs and batteries (13).  Although exposure can occur transcutaneously or by ingestion, inhalation is the major route of toxicity.  Ingested elemental mercury is poorly absorbed and typically leaves the body unchanged without consequence (bioavailability 0.01% [13]). However, inhaled fumes are rapidly absorbed through the pulmonary circulation allowing distribution throughout the major organ systems.  Clinical manifestations vary based on the chronicity of the exposure (14).  Mercury readily crosses the blood-brain barrier and concentrates in the neuronal lysosomal dense bodies. This interferes with major cell processes such as protein and nucleic acid synthesis, calcium homeostasis and protein phosphorylation.  Acute exposure symptoms manifest within hours as gastrointestinal upset, chills, weakness, cough and dyspnea.

Inorganic mercury salts are earthly-appearing, red ore found historically in cosmetics and skin treatments.  Currently, most exposures in the United States occur from exposure through germicides or pesticides (15).  In contrast to elemental mercury, inorganic mercury is readily absorbed through multiple routes including the gastrointestinal tract.  It is severely corrosive to gastrointestinal mucosa (16).  Signs and symptoms include profuse vomiting and often-bloody diarrhea, followed by hypovolemic shock, oliguric renal failure and possibly death (12).

Organic mercury, of which methylmercury is an example, has garnered significant attention recently following several large outbreaks as a result of environmental contamination in Japan in 1956 (17) and grain contamination in Iraq in 1972 (18).  Organic mercury is well absorbed in the GI tract and collects in the brain, reaching three to six times the blood concentration (19).  Symptoms may manifest up to a month after exposure as bilateral visual field constriction, paresthesias of the extremities and mouth, ataxia, tremor and auditory impairments (12).  Organic mercury is also present in a teratogenic agent leading to development of a syndrome similar to cerebral palsy termed "congenital Minamata disease" (20).

The appropriate test depends upon the type of mercury to which a patient has been exposed.  After exposure to elemental or inorganic mercury, the gold standard test is a 24-hr urine specimen for mercury.  Spot urine samples are unreliable.  Urine concentrations of greater than 50 μg in a 24-hr period are abnormal (21).  This patient’s 24-hr urine level was noted to be 90 μg.  Elemental and inorganic mercury have a very short half-life in the blood.

Exposure to organic mercury requires testing hair or whole blood.  In the blood, 90% of methyl mercury is bound to hemoglobin within the RBCs.  Normal values of whole blood organic mercury are typically < 6 μg/L. This patient’s whole blood level was noted to be 26 μg/L.  This likely reflects the large concentration of elemental mercury the patient inhaled and the substantial amount that subsequently entered the blood.

Mercury levels can be reduced with chelating agents such as succimer, dimercaprol (also known as British anti-Lewisite (BAL)) and D-penicillamine, but their effect on long-term outcomes is unclear (22-25).

• Sullivan JT, Sykora K, Schneiderman J, et al. Assessment of alcohol withdrawal: the revised clinical institute withdrawal assessment for alcohol scale (CIWA-Ar). Br J Addict 1989;84:1353-1357.
• Bernard GR, Artigas A, Brigham KL, et al. The American-European Consensus Conference on ARDS. Definitions, mechanisms, relevant outcomes, and clinical trial coordination. Am J Respir Crit Care Med 1994;149:818-824.
• The Acute Respiratory Distress Syndrome Network. Ventilation with lower tidal volumes as compared with traditional tidal volumes for acute lung injury and the acute respiratory distress syndrome. N Engl J Med 2000;342:1301-1308.
• Agarwal R, Reddy C, Aggarwal AN, et al. Is there a role for noninvasive ventilation in acute respiratory distress syndrome? A meta-analysis. Respir Med 2006;100:2235-2238.
• Soyka M. Prevalence of alcohol-induced psychotic disorders. Eur Arch Psychiatry Clin Neurosci 2008;258:317-318.
• Tavel ME, Davidson W, Batterton TD. A critical analysis of mortality associated with delirium tremens. Review of 39 fatalities in a 9-year period. Am J Med Sci 1961;242:18-29.
• McNicoll L, Pisani MA, Zhang Y, et al. Delirium in the intensive care unit: occurrence and clinical course in older patients. J Am Geriatr Soc 2003;51:591-598.
• Bateman T. Notes of a case of mercurial erethism. Medico-Chirurgical Transactions 1818;9:220-233.
• Buckell M, Hunter D, Milton R, et al. Chronic mercury poisoning. 1946. Br J Ind Med 1993;50:97-106.
• Rowens B, Guerrero-Betancourt D, et al. Respiratory failure and death following acute inhalation of mercury vapor. A clinical and histologic perspective. Chest 1991;99:185-190.
• Aguado S, de Quiros IF, Marin R, et al. Acute mercury vapour intoxication: report of six cases. Nephrol Dial Transplant 1989;4:133-136.
• Ibrahim D, Froberg B, Wolf A, et al. Heavy metal poisoning: clinical presentations and pathophysiology. Clin Lab Med 2006;26:67-97, viii.
• A fact sheet for health professionals - elemental mercury. Available from: http://www.idph.state.il.us/envhealth/factsheets/mercuryhlthprof.htm
• Clarkson TW, Magos L, Myers GJ. The toxicology of mercury - current exposures and clinical manifestations. N Engl J Med 2003;349:1731-1737.
• Boyd AS, Seger D, Vannucci S, et al. Mercury exposure and cutaneous disease. J Am Acad Dermatol 2000;43:81-90.
• Dargan PI, Giles LJ, Wallace CI, et al. Case report: severe mercuric sulphate poisoning treated with 2,3-dimercaptopropane-1-sulphonate and haemodiafiltration. Crit Care 2003;7:R1-6.
• Eto K. Minamata disease. Neuropathology 2000;20:S14-9.
• Bakir F, Damluji SF, Amin-Zaki L, et al. Methylmercury poisoning in Iraq. Science 1973;181:230-241.
• Berlin M, Carlson J, Norseth T. Dose-dependence of methylmercury metabolism. A study of distribution: biotransformation and excretion in the squirrel monkey. Arch Environ Health 1975;30:307-313.
• Harada M. Congenital Minamata disease: intrauterine methylmercury poisoning. Teratology 1978;18:285-288.
• Graeme KA, Pollack CVJ. Heavy metal toxicity Part I: Arsenic and mercury. J Emerg Med 1998;16:45-56.
• Aaseth J, Frieheim EA. Treatment of methylmercury poisoning in mice with 2,3-dimercaptosuccinic acid and other complexing thiols. Acta Pharmacol Toxicol (Copenh) 1978;42:248-252.
• Archbold GP, McGuckin RM, Campbell NA. Dimercaptosuccinic acid loading test for assessing mercury burden in healthy individuals. Ann Clin Biochem 2004;41:233-236.
• Kosnett MJ. Unanswered questions in metal chelation. J Toxicol Clin Toxicol 1992;30:529-547.
• Zimmer LJ, Carter DE. The efficacy of 2,3-dimercaptopropanol and D-penicillamine on methyl mercury induced neurological signs and weight loss. Life Sci 1978;23:1025-1034.

The American Thoracic Society improves global health by advancing research, patient care, and public health in pulmonary disease, critical illness, and sleep disorders. Founded in 1905 to combat TB, the ATS has grown to tackle asthma, COPD, lung cancer, sepsis, acute respiratory distress, and sleep apnea, among other diseases.

AMERICAN THORACIC SOCIETY 25 Broadway New York, NY 10004 United States of America Phone: +1 (212) 315-8600 Fax: +1 (212) 315-6498 Email: [email protected]

Privacy Statement | Term of Use | COI Conference Code of Conduct

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 09 September 2019

Critical Transitions in Intensive Care Units: A Sepsis Case Study

• Pejman F. Ghalati 1   na1 ,
• Satya S. Samal 1   na1   nAff3 ,
• Jayesh S. Bhat 1 ,
• Robert Deisz 2 ,
• Gernot Marx 2 &
• Andreas Schuppert 1  

Scientific Reports volume  9 , Article number:  12888 ( 2019 ) Cite this article

5089 Accesses

12 Citations

3 Altmetric

Metrics details

• Computational science
• Health care

The progression of complex human diseases is associated with critical transitions across dynamical regimes. These transitions often spawn early-warning signals and provide insights into the underlying disease-driving mechanisms. In this paper, we propose a computational method based on surprise loss (SL) to discover data-driven indicators of such transitions in a multivariate time series dataset of septic shock and non-sepsis patient cohorts (MIMIC-III database). The core idea of SL is to train a mathematical model on time series in an unsupervised fashion and to quantify the deterioration of the model’s forecast (out-of-sample) performance relative to its past (in-sample) performance. Considering the highest value of the moving average of SL as a critical transition, our retrospective analysis revealed that critical transitions occurred at a median of over 35 hours before the onset of septic shock, which suggests the applicability of our method as an early-warning indicator. Furthermore, we show that clinical variables at critical-transition regions are significantly different between septic shock and non-sepsis cohorts. Therefore, our paper contributes a critical-transition-based data-sampling strategy that can be utilized for further analysis, such as patient classification. Moreover, our method outperformed other indicators of critical transition in complex systems, such as temporal autocorrelation and variance.

Similar content being viewed by others

High-throughput prediction of protein conformational distributions with subsampled AlphaFold2

Gabriel Monteiro da Silva, Jennifer Y. Cui, … Brenda M. Rubenstein

An overview of clinical decision support systems: benefits, risks, and strategies for success

Reed T. Sutton, David Pincock, … Karen I. Kroeker

Adapted large language models can outperform medical experts in clinical text summarization

Dave Van Veen, Cara Van Uden, … Akshay S. Chaudhari

Introduction

Certain biological systems exhibit nonlinear dynamics that undergo sudden regime transitions at tipping points 1 , 2 . In a medical context, these transitions often indicate changes in clinical phenotypes, e.g., disease-onset 3 . Such phenomena have been studied mathematically with techniques from the application of singularity theory to dynamical systems 4 , 5 , 6 . In addition, data-driven methods use statistical indicators known as early-warning signals to model the dynamics of systems approaching transitions 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 . Modeling such transitions is beneficial for several applications in systems medicine, such as monitoring health 15 , 16 , predicting disease-onset and gaining an improved understanding of the underlying disease progression 17 .

Our focus is on sepsis, a common complication in the intensive care unit (ICU), and we introduce a notion of regime transition in septic dynamics. As stated in the Third International Consensus Definitions of Sepsis and Septic Shock (Sepsis-3), “sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection”, and “septic shock is a subset of sepsis in which underlying circulatory and cellular/metabolic abnormalities are profound enough to substantially increase mortality 18 ”. Sepsis causes a high rate of in-hospital mortality and costs the healthcare sector billions due to rising incidence rates and prolonged hospital stays 19 , 20 . Accurate diagnosis, however, remains a challenging task for physicians due to the heterogeneity of infectious agents and the frequent presence of multiple comorbidities. Early, aggressive administration of antibiotics is crucial, and delays in this treatment significantly increase mortality 21 , 22 .

To detect signs of sepsis early, numerous illness severity scores or early-warning signals exist: the Acute Physiology and Chronic Health Evaluation (APACHE II), the Simplified Acute Physiology Score (SAPS II), the Sepsis-related Organ Failure Assessment Score (SOFA), the Modified Early Warning Score (MEWS), and the Simple Clinical Score 23 . These scores are good predictors of general disease severity and mortality but cannot estimate the risk of developing sepsis with reasonable sensitivity and specificity 23 .

Numerous machine learning (ML) methods were therefore developed to predict sepsis onset 24 , 25 , 26 . Rothman et al . 27 used structured information from electronic health records (EHRs) to identify sepsis on admission or to predict its onset during hospitalization. For septic shock prediction, Ghosh et al . 28 proposed an integrative model combining sequential contrast patterns with coupled hidden Markov models. Henry et al . 23 developed a targeted real-time early-warning score (TREWScore) by training a Cox regression model to identify patients at high risk of developing septic shock. Additionally, Horng et al . 29 argued that combining free-text patient data with other predictor features significantly improved the performance of ML models. Although these ML approaches have the potential to increase diagnostic accuracy, they involve time-consuming and domain-specific variable/feature selection 30 , 31 . Our proposed method can be considered in the preprocessing stages to select appropriate data for further downstream analysis.

Our computational method aims to identify and characterize signals indicative of critical transitions based on the concept of surprise loss (SL) 32 . SL was originally developed in econometrics to assess forecast breakdown, i.e., instability in the model’s forecasting ability. Such instability was attributed to instability in the underlying data-generating process, whose effects have been studied from a mathematical perspective 33 , 34 . We assume that similar instability occurs in patient data because of changes in the underlying biological mechanism due to medical intervention or disease progression.

We utilize SL to identify regions in the time series where the data-generating process changes and quantify them with a numerical score. The score captures the extent of deviation between the past performance of a model and its future performance. We consider the highest value of such a score to be a putative tipping point in the disease dynamics, and we consider it as a surrogate for critical transition. In addition, we present a critical-transition-based data-sampling strategy is also presented where data are sampled at regions around critical transition; this strategy outperforms random sampling in differentiation between septic shock and non-sepsis patients. We also compare our approach to methods based on autocorrelation and variance 7 , 15 , 16 , 35 , which have been used to identify early-warning signals of critical transitions.

Materials and Methods

Data source.

We sourced patients’ multivariate time series data from the publicly available EHR database, Medical Information Mart for Intensive Care MIMIC-III v1.4 36 , which contained longitudinal data of 46,520 deidentified patients from 58,976 distinct ICU admissions. For ease of interpretation, we treated each admission as a distinct patient.

In the ICU, clinical staff make swift decisions or take prompt actions during patient management. These employees prioritize timely and correct treatment over consistent documentation of their processes, thereby limiting the reliability of clinical annotation for retrospective analysis. Furthermore, the execution of guidelines for identifying imminent disease varies across hospitals. Hence, we restricted our data analysis to predominantly machine-recorded quantitative variables.

Decision rules for retrospective annotation of the sepsis syndrome have evolved over the decades as knowledge of its pathophysiology and epidemiological impact have increased 37 . Whereas earlier definitions (1991 38 , 2001 39 ) focused on uncontrolled systemic inflammation as the major indicator, the latest 2016 18 definition, commonly known as Sepsis-3, emphasizes organ dysfunction as the leading effect of the sepsis syndrome and proposes to update the International Classification of Diseases (ICD) coding system 40 , 41 (ICD-9: 995.92, 785.52; ICD-10: R65.20, R65.21). SOFA scoring system grades the extent of organ dysfunction and is calculated every 24 hours during a patient’s ICU stay 42 , 43 .

Because the ICD-9 codes in our data were not compatible with Sepsis-3, we annotated the patient data in accordance with Table  2 from the 2016 consensus definition 18 . Fig.  1 illustrates a general schematic of our annotation framework.

Over the length of a patient’s ICU stay, all timestamps of body fluid (blood, urine, cerebrospinal fluid) sampling and antibiotic administration were retrieved. For each of the timestamps, an infection was suspected if antibiotics were administered within 72 hours of any prior body fluid sampling (irrespective of culture findings) or if any body fluids were sampled within 24 hours of prior antibiotic administration. Sepsis-3 criteria were independently evaluated over time windows around the infection-suspected timestamps (IST). Each time window began 48 hours prior to IST until 24 hours post IST. If the criteria were satisfied during a given time window, then the beginning of the window was annotated as the onset time. In the schematic, the 2 nd antibiotics administration falls within 72 hours of previous body fluid sampling; thus, an infection is suspected.

The annotation framework was applied to all 58,976 patients, identifying 22,547 (38.2%) sepsis patients and 3208 (5.4%) septic shock patients. Among the 3208 septic shock patients, we analyzed only adults (18+ years old at admission) with at least a 36-hours stay and at most 144 hours spent in the unit before onset, which generated a cohort of 630 patients. Our non-sepsis cohort comprised 6,236 patients who lacked Sepsis-3 annotation or sepsis-specific ICD-9 codes and who stayed between 36 and 144 hours in the ICU. Demographic information on the two cohorts can be found in Supplementary Table  S1 .

We cannot exhaustively evaluate and validate the accuracy of our annotation framework owing to the absence of a manually curated “ground truth” dataset of Sepsis-3 patients. Software implementations with different data cleaning processes and patient exclusion criteria (PEC) from the same annotation framework could result in divergent cohorts. For example, for the same database, another implementation 44 annotated almost half (49.1%) of their analysis cohort (n = 11,791; reasonable PEC) as Sepsis-3, whereas our implementation annotated approximately 38% of the entire population (n = 58,976; no PEC). There may be a high degree of overlap in the annotated cohorts; thus, a comparison of the two implementations is currently under way.

Based on availability and relevance to sepsis, we preselected groups of variables: the laboratory variables included bicarbonate, creatinine, blood urea nitrogen (BUN), hematocrit, hemoglobin, platelet count, white blood cell count (WBC), potassium, and sodium; the vital signs and physiological variables comprised body temperature, heart rate, respiratory rate, oxygen saturation (SpO2), arterial blood pressure (systolic, mean, and diastolic), and urine output; the two septic markers comprised the shock index (ratio of heart rate over systolic blood pressure), and the ratio of BUN to creatinine 23 . Table  1 shows the mean sampling rates of the variables in the respective patient cohorts, and their distribution can be seen in Supplementary Fig.  S1 .

Missing value imputation and time binning

Data representation is a crucial step in analyzing time series. Continuous EHRs suffer from missing values due to insufficient data collection and lack of documentation. Additionally, high heterogeneity in variable type and irregular sampling intervals make such data difficult to handle. To address the problems of missing data and data sparsity, we transformed our time series into 30-minute time bins by imputing values in the bins and averaging measurement values over the bins. We experimented with different imputation methods, such as linear, polynomial and Stineman interpolation 45 . The Stineman method was chosen due to its superior performance in reducing overshoots and handling sharp changes in the imputed values.

Data normalization

Our variables (Table  1 ) had different scales and measurement units. Data normalization was therefore needed for our method. For this purpose, we transformed the observables by Z-score normalization to address the use of different units of measurement.

State space model

To define SL, we require a dynamical mathematical model for our multivariate clinical time series. Here, we consider a state space model (SSM) approach 46 , 47 , which models the data in a hierarchical manner with hidden states that give rise to observables. In our context, the hidden states can be assumed to represent the biological processes, and the observables represent the clinically measured variables. The observables in our SSM are expressed as linear combinations of hidden random states. Such a model incorporates the variations in the biological processes and a measurement noise term. The variations due to biology are modeled by adding a stochastic term to the hidden states, whereas the measurement noise term is added to the observables. Both terms are assumed to follow a multivariate normal (MVN) distribution.

The computation of SL is agnostic to the underlying dynamical model. The SL literature 32 uses a linear dynamical model, whereas we use an SSM for our application. The primary reason to use this type of model is to separate the biological processes from the observables, i.e., to model two sources of variability. Below, we represent such an SSM model.

where the indices of the time series are from t  = 1, …, T ; e is the number of hidden trends; x is an e  ×  T matrix of hidden states; y is an n  ×  T matrix of n observables; and w is an e  ×  T matrix of process error. In general, \(e\ll n\) . The process error at time t follows an MVN distribution with mean 0 and e  ×  e covariance matrix Q ; v is an n  ×  T matrix of observation error. The observation error at time t follows an MVN distribution with mean 0 and n  ×  n covariance matrix R ; Z is an n  ×  e parameter matrix; a is a vector of offsets; π is a matrix of e  × 1 means; ∧ is an e  ×  e covariance matrix. The set of parameters can be represented in compact form as θ  = ( Q , R , Z , x 1 , …, T , π , ∧ ), and their estimate is \(\hat{\theta }\) . \({\hat{y}}_{t}\) and \({\tilde{y}}_{t+\lambda }\) are the estimate and λ -step-ahead forecast, respectively, of the given observables y t .

Our implementation incorporated MARSS 48 , 49 , which is an R package for fitting constrained and unconstrained linear multivariate autoregressive SSMs by maximum likelihood parameter estimation. We utilized MARSS to fit an SSM to our multivariate time series data, using its recommended initial conditions that ensure parameter identifiability. We assumed the presence of multiple hidden states and fixed e  = 3. Furthermore, we evaluated the robustness of our results with respect to the changes in the model parameters (see ‘Robustness of the SSM model’).

Perturbations in the dynamics

Early-warning indicator.

Our proposed computational method based on surprise loss (SL) 32 computes the difference between the forecast error, i.e., out-of-sample error, and the in-sample performance. The out-of-sample error measures the quality of model forecasts, i.e., the prediction of the model for the data that were not used for fitting, whereas the in-sample error quantifies the deviation between the model estimates and the data that were used for model fitting. A high out-of-sample error compared to the in-sample error is suggestive of instability in the patient data. In such a scheme, our model may be a poor fit for the data, but we are interested in evaluating whether the past performance of the model is consistent with future forecasts. The performance is measured for a fixed loss function using a moving time window. Furthermore, the SL computation is unsupervised, i.e., the clinical conditions of patients, such as septic shock or non-sepsis, are not required. Originally, the idea of SL was used to perform a statistical test to determine forecast breakdown in time series, i.e., to determine whether the average of SL is close to zero 32 . However, in our application, the aim is not to test whether a given time series underwent a statistically significant forecast breakdown; rather, it is to identify high SL values in the given time series and later use this information in postprocessing steps (see ‘Data-sampling strategy with SLMean’).

In spirit, this approach is close to the identification of structural breaks or change-points analysis 50 , 51 . However, the SL-based approach has the additional advantage of being robust to model misspecification. Specifically, in practice, the SSM model (i.e., the functional form and variables) is likely to be misspecified and may not be a good approximation of the underlying disease processes. By formalizing SL as the difference between in-sample and out-of-sample performance and not relying on model parameters or error variances, the SL-based approach provides a natural way to handle such scenarios (see ‘Relationship with the literature’ in Giacomini et al . 32 ).

With a moving time window of width m , the SSM model (see equation ( 1 )) was fitted for time indices \(t-m+\mathrm{1,}\ldots ,t\) . \({y}_{t}^{{i}_{c}}\) denotes the observables of a given patient i with clinical condition c at time index t , and \({T}^{{i}_{c}}\) is the length of the corresponding time series. The in-sample error is a quadratic loss function that averages the squared differences between the estimated and the given observables, and it is denoted as \({L}_{j}({\hat{\theta }}_{t}^{{i}_{c}})=\frac{1}{n}{\sum }_{k=1}^{n}\,{(y{(k)}_{j}^{{i}_{c}}-\hat{y}{(k)}_{j}^{{i}_{c}})}^{2}\) where \(y{(k)}_{j}^{{i}_{c}}\) is the k th element of column vector \({y}_{j}^{{i}_{c}}\) . Similarly, the out-of-sample error is a quadratic loss function that averages the squared differences between the λ -step-ahead forecast and the given observables, and it is denoted as \({L}_{t+\lambda }({\hat{\theta }}_{t}^{{i}_{c}})=\frac{1}{n}{\sum }_{k=1}^{n}\,{(y{(k)}_{t+\lambda }^{{i}_{c}}-\tilde{y}{(k)}_{t+\lambda }^{{i}_{c}})}^{2}\) . The SL is the difference between the out-of-sample and the in-sample error:

To remove short-term fluctuations, a moving-average filter (with size δ ) smooths the SL:

For a given patient i , prior to the clinically annotated onset of disease c , a relatively high \(SLMea{n}_{t}^{{i}_{c}}\) suggests putative transitions across dynamical regimes and serves as an early-warning indicator. We consider the maximum of \(SLMea{n}^{{i}_{c}}\) at time index \({t}_{max}^{{i}_{c}}\) to denote a critical transition. Fig.  2a illustrates the calculation of \(S{L}^{{i}_{c}}\) , \(SLMea{n}^{{i}_{c}}\) and \({t}_{max}^{{i}_{c}}\) . A simulated example using synthetic data is shown in Fig.  3 .

( a ) A schematic for the calculation of \(S{L}^{{i}_{c}}\) , \(SLMea{n}^{{i}_{c}}\) , and \({t}_{max}^{{i}_{c}}\) for a given patient i and clinical condition c . The SSM was fitted with a moving time window of length m (as shown in blue) and the \(S{L}^{{i}_{c}}\) was calculated. A second sliding window of length δ was used to compute the \(SLMea{n}^{{i}_{c}}\) (as illustrated in green). The \({T}^{{i}_{c}}\) denotes disease onset in septic shock patients and it represents the time of discharge or death in non-sepsis patients. The \({t}_{max}^{{i}_{c}}\) denotes the time index of the highest \(SLMea{n}^{{i}_{c}}\) and it was used in our data-sampling approach. ( b ) A schematic diagram illustrating our data-sampling strategy using our method. Observables at the time of highest SLMean magnitudes, i.e., critical transition points, were selected from septic shock and non-sepsis patients. The Wilcoxon rank-sum test was used to determine the statistical significance of the changes in the observables.

Artificial example showing the calculation of SLMean from a synthetic dataset that was generated by concatenating 50 points, drawn independently from three univariate normal distributions with different means (5, 10, 15) and a standard deviation of 0.5. Computed with a moving time window of length 30 and the number of hidden states set to 1, the magnitude of SL intensified at the 50 th and 100 th time-points, where the parameters of the data-generating process changed, i.e., a proxy for transitions across different dynamical regimes.

Uncertainty in SLMean

Uncertainty in out-of-sample forecasting and in-sample performance adds noise to the precise location of \({t}_{max}^{{i}_{c}}\) . Let \({t}_{max(up)}^{{i}_{c}}\) and \({t}_{max(low)}^{{i}_{c}}\) , respectively, be the time indices corresponding to the modes of the upper and lower bounds of the 95% prediction interval of SLMean . Our approach is robust if the deviations of \({t}_{max}^{{i}_{c}}\) from \({t}_{max(up)}^{{i}_{c}}\) and \({t}_{max(low)}^{{i}_{c}}\) are close to zero.

Data-sampling strategy with SLMean

Here, we demonstrate a method for sampling data from the critical transition points (derived from SLMean ) to differentiate the septic shock cohort from the non-sepsis cohort (see Fig.  2b ). We also propose a bootstrap test (based on a random sampling of data) to evaluate whether it outperforms the SL-based approach. Such a data selection step can be seen as a preprocessing step for the machine learning-based techniques being developed to study sepsis (as described in ‘Introduction’). The data sampling step is agnostic to the clinical condition of the patient, i.e., data for each patient are based on SL (see ‘Perturbations in the dynamics’), and in a subsequent step, we used the clinical condition to perform statistical tests.

Specifically, we selected the data at \({t}_{max}^{{i}_{c}}\) , i.e., the critical transition points (in the case of multiple \({t}_{max}^{{i}_{c}}\) values, the one closer to the disease-onset was selected), sampled the corresponding data and represented them as an n  ×  v variable matrix \({S}^{c}=[{y}_{{t}_{{\max }}}^{{1}_{c}},\ldots ,{y}_{{t}_{{\max }}}^{{v}_{c}}]\) where c   ∈  {0, 1} i.e., non-sepsis and septic shock conditions, and v is the total number of patients. Thereafter, for each variable, a p-value based on Wilcoxon rank-sum test 52 was calculated, quantifying the significance of differences between the two patient cohorts (as shown in the equation ( 4 )).

where pval (.) returns the p-value based on the Wilcoxon rank-sum test. \({S}_{j}^{0}\) and \({S}_{j}^{1}\) denote the j th row vectors of matrices S 0 and S 1 matrices, respectively. Furthermore, we performed the Benjamini and Hochberg correction method to adjust the p-values 53 accounting for multiple comparisons.

Bootstrapping

Furthermore, a bootstrap test was used to compare the p-values calculated at critical transition points from the p-values that were obtained from random points in our time series. For a randomly selected time index t with its corresponding observation \({y}_{t}^{{i}_{c}}\) , where \(t\in \mathrm{(1,}\,{T}^{{i}_{c}})\) , the t random p-values were calculated by replacing t max with t . The test was repeated 1000 times. Bootstrap frequency (BF) denotes the fraction of replications wherein t max p-values were less than t random p-values. A high BF value indicates that the SL based approach has an advantage over the random approach. In addition to computing the BF on data randomly sampled from all times, we computed BF on randomly sampled data of septic patients from two arbitrary time intervals, 36 hours and 18 hours before the onset of septic shock. This step allows us to test whether merely randomly sampling data close to the onset time can outperform the SL approach.

Autocorrelation and variance as early-warning signals

In the dynamics of a system, increased temporal autocorrelation and increased variance are hypothesized to be two indicators that the system is approaching a state transition 7 . To evaluate the SL concept, we calculated these two presumed early warning signals and compared the results with those obtained from the SL approach. As these measures are both univariate, to apply them to our multivariate time series data, we formulated them as follows:

where AC and AC1 are autocorrelation and variance functions applied on variable y ( k ) for time indices \(t-m+\mathrm{1,}\ldots ,t\) . t is the time index, and m is the width of a moving time window. The first coefficient of auto-correlation \(AC{1}_{t}^{{i}_{c}}\) and variance \(VA{R}_{t}^{{i}_{c}}\) were computed by averaging over N variables. i is the index of a given patient with clinical condition c , and \({T}^{{i}_{c}}\) is the length of the corresponding time series.

Similar to the SL concept, t max is defined as the time index where the highest value of the early-warning signal occurs (here, the largest value of \(AC{1}^{{i}_{c}}\) or \(VA{R}^{{i}_{c}}\) ). P-values and bootstrap frequencies were computed as described in ‘SLMean-based data-sampling strategy’ and ‘Data-sampling strategy with SLMean’.

To support reproducible research, our computational method is available at https://github.com/JRC-COMBINE/SL-MTS .

SLMean as an early-warning indicator

Over a moving time window ( m  = 36, i.e., 18 hours; e  = 3; λ -step-ahead = 1, i.e., 30 minutes; δ  = 6, i.e., 3 hours), the \(SLMea{n}^{{i}_{c}}\) values (‘Perturbations in the dynamics’), as shown in Fig.  4 , were computed. A positive \(SLMea{n}_{t}^{{i}_{c}}\) indicates higher out-of-sample error than in-sample error, signaling putative transitions in the underlying dynamics. The componentwise mean vector and associated standard deviation of all septic shock patients, i.e., \(SLMea{n}^{{1}_{c}},\ldots ,SLMea{n}^{{N}_{c}}\) (where N is the total number of septic shock patients and c is the septic shock clinical condition), intensified as the moving time window approached the disease onset. For the same cohort of septic shock patients, a slight increase in the componentwise mean vector and associated standard deviation of \(VA{R}^{{i}_{c}},\ldots ,VA{R}^{{N}_{c}}\) could be seen, while those of \(AC{1}^{{1}_{c}},\ldots ,AC{1}^{{N}_{c}}\) did not show any changes over time. The findings are summarized in Fig.  5 .

The changes over time in a group of clinical variables used in this study and the corresponding computed \(SLMea{n}^{{i}_{c}}\) of a sample septic patient before the onset of septic shock (violet line). The \(SLMea{n}^{{i}_{c}}\) is calculated over a moving time window ( m  = 36, i.e., 18 hours; e  = 3; λ -step-ahead = 1, i.e., 30 minutes; δ  = 6, i.e., 3 hours). The red line shows the time location ( \({t}_{max}^{{i}_{c}}\) ) of the largest \(SLMea{n}^{{i}_{c}}\) (i.e., the critical transition point).

Componentwise mean (red dots) and ±standard deviation (blue lines) of ( a ) SLMean ( b ) AC1 and ( c ) VAR for all septic shock patients prior to disease-onset (see ‘SLMean as an early-warning indicator’); T is the length of the time series (i.e. max( \({T}^{{1}_{c}},\ldots ,{T}^{{N}_{c}}\) ), where c  = 1 represents septic shock condition and N is the total number of septic shock patients), and t  −  T is the time before the onset of septic shock. The number of samples per time point could be different due to the heterogeneous length of hospitalization (see ‘Data source’). As the maximum length of hospitalization was 144 hours, with a moving time-window length of 18 hours and an average window of 3 hours, the minimum value of t  −  T was −123 hours.

It should be taken into account that the largest \(SLMea{n}_{t}^{{i}_{c}}\) need not necessarily occur exactly at the time of disease onset. For septic shock patients, the location of the time index t max from the onset time ( T ) is shown in Fig.  6b . In the majority of our patients’ data, the highest SLMean occurred near septic shock onset (60% of the patients, the signal occurred less than 48 hours prior to onset, as shown in Fig.  6b ). However, in some patients, the signal was observed beyond onset time. Possible explanations include a lack of records or a low sampling rate of variables a few days before the onset of septic shock, resulting in a nonsignificant SLMean . The highest SLMean , on average, occurred 46 hours (median of 35.6 hours) prior to the appearance of septic shock symptoms. In comparison, TREWScore 23 identified septic patients at a median of 28.2 hours before onset.

( a ) Componentwise mean of SLMean for all septic shock patients prior to disease-onset (see ‘SLMean as an early-warning indicator’); T is the length of the time series, ( b ) Distribution of the times of critical transitions from the onset times of septicshock, i.e. \({t}_{max}^{{1}_{c}}-{T}^{{1}_{c}},\ldots ,{t}_{max}^{{N}_{c}}-{T}^{{N}_{c}}\) , where c  = 1 represents septic shock condition and N is the total number of septic shock patients. \(SLMea{n}^{{i}_{c}}\) reaches a maximum at \({t}_{max}^{{i}_{c}}\) .

While the median time of the peak SLMean occurred at 35.6 hours before the onset of septic shock, visual inspection of the mean and standard deviation of SLMean indicates an upward trend starting from approximately 24 hours (Figs  5a and 6a ). The explanation for the apparent deviation from the baseline is that the highest \(SLMea{n}_{t}^{{i}_{c}}\) values that occurred closer to onset were greater in magnitude.

Furthermore, we determined the uncertainty in SL calculation using prediction intervals (as described in ‘Uncertainty in SLMean’). Our results show negligible deviation in t max i.e., the median deviation is 0, and the interquartile range (IQR) is 5.4 hours.

SLMean -based data-sampling strategy

We compared the p-values for data sampled at t max (i.e., critical transition point) to those obtained via random sampling (see equation ( 4 ) and ‘Data-sampling strategy with SLMean’). The same procedure was implemented for AC1 and VAR , and the bootstrap test was performed for all time indices. The bootstrap frequencies were denoted as BF ( SL ), BF ( AC1 ) and BF ( VAR ), respectively (see Table  2 ). The different BF computations test the association of the bootstrap frequency values of some variables with high SLMean , AC1 and VAR . In 14 out of 19 variables, BF ( SL ) demonstrates superior results. In the next step, in addition to all the time indices, the bootstrap test was performed for time-windows of 18 and 36 hours before the onset of septic shock; the bootstrap frequencies are represented as BF ( Full ), BF (18  hours ), and BF (36  hours ). Fig.  7a plots BF ( Full ) against p-values computed at t random and at high SLMean (i.e., t max ). Most of the variables show a good BF with high log-transformed p-values when sampled at large SLMean , particularly in the case of variables such as blood pressures, temperature and SpO2, where random sampling leads to poor p-values. As the random sampling strategy changed to either to 36 or 18 hours in Fig.  7b , BF reduced for six variables (WBC, diastolic blood pressure, Hemoglobin, SpO2, creatinine, and BUN), but it was preserved for nine variables (respiratory rate, heart rate, potassium, mean blood pressure, hematocrit, shock index, temperature, BUN-creatinine, and systolic blood pressure), i.e., the differences among BF ( Full ), BF (36  hours ), and BF (18  hours ) were small. Four variables, bicarbonate, urine output, platelets and sodium, had low BF ( Full ), BF (36  hours ), and BF (18  hours ).

( a ) A statistical significance test (see ‘Data-sampling strategy with SLMean’) was performed to test whether the values of the clinical variables at largest SLMean were able to differentiate septic shock patients from non-sepsis patients. The − log 10 ( P - value ) of each variable at t max was compared with the median − log 10 ( P - value ) of randomly selected points from the whole sequence. A bootstrap test (denoted as BF Full ) was performed to quantify the number of times the p-values at t max are lower than those at t random . ( b ) Another bootstrap test, performed by randomly sampling points from 18 hours and 36 hours windows prior to the onset of septic shock, tested whether low p-values at t max are an effect of time or a characteristic of regions with high SLMean values.

Robustness of the SSM model

We assessed the robustness of our method to perturbations in the model parameters. We changed the length of the moving time window, m   ∈  (24, 30), and the number of trends in the SSM model ( e   ∈  (4, 5)) and compared the changes in the t max with respect to the reference setting, i.e., m  = 36 and e  = 3. The chosen values of e are based on the assumptions described in ‘State space model’ ( e  = 3 and \(e\ll n\) ). The length of the moving time window was selected with regard to the average variables sampling rate (see Table  1 , as well as the length of hospitalization in the ICU (see ‘Data source’). The differences in t max due to the perturbations are summarized in online Supplementary Fig.  S2 . The zero median of such differences confirmed the robustness of our approach. Due to multiple similar high values SLMean in some patients, alteration of model parameters led to different t max values in these patients, which caused the outliers in online Supplementary Fig.  S2 .

Discussions

Healthcare can benefit from the analysis of continuously monitored health data, which are rapidly growing in quantity due to the increasing availability of long time series collected either by wearables or by monitoring systems such as those established in the ICU. However, significant challenges remain unresolved. A major drawback is the restriction of data availability to variables that are easy to collect by noninvasive sensors. These variables provide only correlated surrogates of the primary disease-driving processes. Hence, sensor signals are rarely specific on their own; advanced computational processing is typically necessary to identify relevant signals to improve therapy.

Focusing data analysis on the prediction and identification of critical transitions, i.e., instabilities in patient data, may complement established scoring methods in the classification of stable states. Although critical transitions differ qualitatively from scores in classifying stable states, the former method provides an independent assessment of health status. Because critical transitions are simply identified through the evolution of individual longitudinal time series, in contrast to established scores based on absolute variable values, markers for the detection of critical transitions are relatively robust to normalization and data standardization issues.

To identify such critical transitions in ICU patients, we applied the concept of surprise loss (SL), which was originally developed for determining instability in a model’s forecasting ability in econometrics. We changed the model in the original SL approach to a multivariate SSM model to model two sources of variability, namely, the hidden underlying biological processes and the observables. Despite a multitude of interventions in the ICU, our moving average SL, SLMean , showed, on average, an increasing signal approximately 24 hours before the expert-annotated onset of septic shock (see Fig.  6a ), thereby indicating its applicability as an early-warning indicator. We utilized such an indicator to devise a critical-transition-based data-sampling strategy for discriminating septic shock from non-sepsis patients. Additionally, through a bootstrap test (quantified through BF ( Full )), the benefit of our method is shown with respect to a random data selection strategy (as summarized in Table  2 and Fig.  7a ). Except for bicarbonate, urine output, platelets and sodium, the SL-based approach results in better p-values and BF ( Full ) than the random strategy. In addition, we selectively sampled random data from 36 hours and 18 hours before the septic onset to compute BF (36  h ) and BF (18  h ), respectively (see Fig.  7b and Supplementary Table  S2 ). Such selective sampling evaluates whether merely sampling data close to the onset time of septic shock outperforms our method in distinguishing sepsis from non-sepsis. These new BF values seem to be well-preserved for most variables that have correspondingly high BF ( Full ). Therefore, an SL-informed sampling strategy may improve the quality of patient classification and eventually enable the reduction of sample sizes.

Moreover, from a systems theory point of view, mechanisms that control the system in homeostasis begin to collapse around a critical transition or tipping point. Consequently, variables that are under tight control within stable states may be more sensitive to systemic variability around an unstable point. Our data analysis supports this hypothesis (see Fig.  7a ): some variables under tight control, e.g., blood pressure and body temperature, showed significant improvement in p-values compared to random sampling. We compared our method with two other univariate early-warning measures for critical transitions in complex systems: temporal autocorrelation and variance 7 , 15 , 16 , 35 . As shown in Fig.  5 , our method outperformed these estimators as an early-warning indicator for septic shock patients. Similarly, the p-values and BF of our method were also more favorable than those of the other methods (Table  2 ).

Conceptually, SL computation is based on the premise that the underlying system has a stable stationary state and that all observed deviations can be explained as responses to stochastic perturbations. The permissible amount of deviation is controlled by the system’s robustness at the time of computation. As a result, SL-based analysis can forewarn of a “loss of stability” even before the underlying system has changed its state. In that sense, SL provides indicators similar to those from the analysis of critical slowing down 35 . One drawback is that local loss of robustness may neither result in a transition to another state nor indicate a new state. SL-based warning systems, in isolation, may thus lead to false alarms and could be improved by combining them with ML classifiers. Additionally, moving-window length restricts the capability of the SL-based warning system, and analysis can only be performed only when sufficient data have been collected. Hence, further evaluations must be performed towards utilization of SL-based analysis in practice. As a high SL is not specific and can be generated by any sudden event affecting the data, either errors in the monitoring system or health-related covariates, a robust characterization of the standard SL patterns characterizing control states is crucial. As sudden, high SL peaks can arise from sudden monitoring aberrations, we expect that a threshold-based alarm system might result in an unacceptable false positive rate. Therefore, emphasis should be placed on the characterization of SL patterns that are representative of the control state, eventually for each individual patient, followed by an AI-based pattern classifier. Effectively, this method will result in significant calibration times to setup the alarm system for each patient, such that effective training procedures for the learning of the control state patterns might be essential for transfer to clinical applications.

Mojtahedi, M. et al . Cell fate decision as high-dimensional critical state transition. PLOS Biol. 14 , 1–28, https://doi.org/10.1371/journal.pbio.2000640 (2016).

Article   CAS   Google Scholar  

Samal, S. S. et al . Metastable regimes and tipping points of biochemical networks with potential applications in precision medicine. bioRxiv , https://doi.org/10.1101/466714 (2018).

Trefois, C., Antony, P. M., Goncalves, J., Skupin, A. & Balling, R. Critical transitions in chronic disease: transferring concepts from ecology to systems medicine. Curr. Opin. Biotechnol. 34 , 48–55, https://doi.org/10.1016/j.copbio.2014.11.020 (2015).

Article   CAS   PubMed   Google Scholar  

Thomas, R. & d’Ari, R. Biological feedback (CRC press, 1990).

Waddington, C. H. The strategy of the genes. London : Allen 86 (1957).

Arnold, V. I. Singularities, Bifurcations, and Catastrophes. In Catastrophe Theory , 1–2, https://doi.org/10.1007/978-3-642-58124-3 (Springer Berlin Heidelberg, Berlin, Heidelberg, 1992).

Book   Google Scholar  

Scheffer, M. et al . Early-warning signals for critical transitions. Nature 461 , 53–59, https://doi.org/10.1038/nature08227 (2009).

Article   ADS   CAS   PubMed   Google Scholar  

Guttal, V. & Jayaprakash, C. Changing skewness: an early warning signal of regime shifts in ecosystems. Ecol. Lett. 11 , 450–460, https://doi.org/10.1111/j.1461-0248.2008.01160.x (2008).

Article   PubMed   Google Scholar  

Dakos, V. et al . Slowing down as an early warning signal for abrupt climate change. Proc. Natl. Acad. Sci. 105 , 14308–14312, https://doi.org/10.1073/pnas.0802430105 (2008).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Dakos, V., van Nes, E. H., Donangelo, R., Fort, H. & Scheffer, M. Spatial correlation as leading indicator of catastrophic shifts. Theor. Ecol. 3 , 163–174, https://doi.org/10.1007/s12080-009-0060-6 (2010).

Article   Google Scholar  

Carpenter, S. R. & Brock, W. A. Rising variance: a leading indicator of ecological transition: Variance and ecological transition. Ecol. Lett. 9 , 311–318, https://doi.org/10.1111/j.1461-0248.2005.00877.x (2006).

Chen, L., Liu, R., Liu, Z.-P., Li, M. & Aihara, K. Detecting early-warning signals for sudden deterioration of complex diseases by dynamical network biomarkers. Sci. reports 2 , 342 (2012).

Lenton, T. M., Livina, V. N., Dakos, V., Van Nes, E. H. & Scheffer, M. Early warning of climate tipping points from critical slowing down: Comparing methods to improve robustness. Philos. Transactions Royal Soc. A: Math. Phys. Eng. Sci. 370 , 1185–1204, https://doi.org/10.1098/rsta.2011.0304 (2012).

Article   ADS   CAS   Google Scholar  

Chen, P., Liu, R., Li, Y. & Chen, L. Detecting critical state before phase transition of complex biological systems by hidden Markov model. Bioinformatics 32 , 2143–2150, https://doi.org/10.1093/bioinformatics/btw154 (2016).

Gijzel, S. M. W. et al . Dynamical resilience indicators in time series of self-rated health correspond to frailty levels in older adults. 72 , 991–996, https://doi.org/10.1093/gerona/glx065 (2017).

Gijzel, S. M. W. et al . Dynamical indicators of resilience in postural balance time series are related to successful aging in high-functioning older adults, https://doi.org/10.1093/gerona/gly170 (2018).

Liu, R., Chen, P., Aihara, K. & Chen, L. Identifying early-warning signals of critical transitions with strong noise by dynamical network markers. Sci . Reports 5 , https://doi.org/10.1038/srep17501 (2015).

Singer, M. et al . The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA 315 , 801, https://doi.org/10.1001/jama.2016.0287 (2016).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Fleischmann, C. et al . Epidemiology of sepsis in germany: Incidence, mortality and associated costs of care 2007-2013. Intensive Care Medicine Exp. 3 , A50, https://doi.org/10.1186/2197-425X-3-S1-A50 (2015).

Bhattacharjee, P., Edelson, D. P. & Churpek, M. M. Identifying Patients With Sepsis on the Hospital Wards. Chest 151 , 898–907, https://doi.org/10.1016/j.chest.2016.06.020 (2017).

Gaieski, D. F. et al . Impact of time to antibiotics on survival in patients with severe sepsis or septic shock in whom early goal-directed therapy was initiated in the emergency department*. Critical Care Medicine 38 , 1045–1053, https://doi.org/10.1097/CCM.0b013e3181cc4824 (2010).

Kumar, A. et al . Duration of hypotension before initiation of effective antimicrobial therapy is the critical determinant of survival in human septic shock*. Critical Care Medicine 34 , 1589–1596, https://doi.org/10.1097/01.CCM.0000217961.75225.E9 (2006).

Henry, K. E., Hager, D. N., Pronovost, P. J. & Saria, S. A targeted real-time early warning score (TREWScore) for septic shock. Sci . Transl . Medicine 7 , 299ra122–299ra122, https://doi.org/10.1126/scitranslmed.aab3719 (2015).

Mitchell, S. et al . Optimization of sepsis risk assessment for ward patients. 107–112, https://doi.org/10.1109/SIEDS.2016.7489280 (IEEE, 2016).

Carrara, M., Baselli, G. & Ferrario, M. Mortality prediction in septic shock patients: Towards new personalized models in critical care. 2792–2795, https://doi.org/10.1109/EMBC.2015.7318971 (IEEE, 2015).

Tang, C. H. H. et al . Non-invasive classification of severe sepsis and systemic inflammatory response syndrome using a nonlinear support vector machine: a preliminary study. Physiol. Meas. 31 , 775–793, https://doi.org/10.1088/0967-3334/31/6/004 (2010).

Rothman, M. et al . Sepsis as 2 problems: Identifying sepsis at admission and predicting onset in the hospital using an electronic medical record–based acuity score. J. Critical Care 38 , 237–244, https://doi.org/10.1016/j.jcrc.2016.11.037 (2017).

Ghosh, S., Li, J., Cao, L. & Ramamohanarao, K. Septic shock prediction for ICU patients via coupled HMM walking on sequential contrast patterns. J. Biomed. Informatics 66 , 19–31, https://doi.org/10.1016/j.jbi.2016.12.010 (2017).

Horng, S. et al . Creating an automated trigger for sepsis clinical decision support at emergency department triage using machine learning. Plos One 12 , e0174708, https://doi.org/10.1371/journal.pone.0174708 (2017).

Cook, R. D. Detection of influential observation in linear regression. Technometrics 19 , 15–18, https://doi.org/10.1080/00401706.1977.10489493 (1977).

Article   MathSciNet   MATH   Google Scholar  

Guyon, I. & Elisseeff, A. An introduction to variable and feature selection. J. machine learning research 3 , 1157–1182 (2003).

MATH   Google Scholar  

Giacomini, R. & Rossi, B. Detecting and predicting forecast breakdowns. Rev. Econ. Stud. 76 , 669–705, https://doi.org/10.1111/j.1467-937X.2009.00545.x (2009).

Clements, M. & Hendry, D. Forecasting economic time series (Cambridge University Press, 1998).

Hendry, D. F. & Doornik, J. A. The implications for econometric modelling of forecast failure. Scott. J. Polit. Econ. 44 , 437–461, https://doi.org/10.1111/1467-9485.t01-1-00065 (1997).

Scheffer, M. et al . Quantifying resilience of humans and other animals. 115 , 11883–11890, https://doi.org/10.1073/pnas.1810630115 (2018).

Johnson, A. E. W. et al . MIMIC-III, a freely accessible critical care database. Sci . Data 3 , sdata201635, https://doi.org/10.1038/sdata.2016.35 (2016).

Abraham, E. New definitions for sepsis and septic shock: continuing evolution but with much still to be done. Jama 315 , 757–759 (2016).

Bone, R. C. et al . Definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Chest 101 , 1644–1655 (1992).

Levy, M. M. et al . 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference. Critical Care Medicine 31 , 1250–1256, https://doi.org/10.1097/01.CCM.0000050454.01978.3B (2003).

Organization, W. H. ICD-10: The ICD-10 Classification of Mental and Behavioural Disorders: diagnostic criteria for research. In ICD-10 : the ICD-10 classification of mental and behavioural disorders : diagnostic criteria for research (1993).

Statistics (US), N. C. f. H. The International Classification of Diseases : 9th Revision , Clinical Modification : ICD-9-CM (1991).

Vincent, J. L. et al . Use of the SOFA score to assess the incidence of organ dysfunction/failure in intensive care units: results of a multicenter, prospective study. Critical care medicine 26 , 1793–1800 (1998).

Vincent, J. L. et al . The SOFA ( Sepsis-related Organ Failure Assessment ) score to describe organ dysfunction/failure (Springer, 1996).

Johnson, A. E. W. et al . A Comparative Analysis of Sepsis Identification Methods in an Electronic Database. Critical care medicine (2018).

Stineman, R. W. A Consistently Well Behaved Method of Interpolation. Creat. Comput. 6 , 54–57 (1980).

Google Scholar  

Harvey, A. C. Forecasting , structural time series models and the Kalman filter (Cambridge university press, 1990).

Auger-Méthé, M. et al . State-space models’ dirty little secrets: even simple linear Gaussian models can have estimation problems. Sci. Reports 6 , 26677, https://doi.org/10.1038/srep26677 (2016).

Holmes, E., Ward, E. & Wills, K. MARSS : Multivariate Autoregressive State-Space Modeling (2018).

Holmes, E. E., Ward, E. J. & Wills, K. MARSS: Multivariate autoregressive state-space models for analyzing time-series data. The R J. 4 , 30 (2012).

Andrews, D. W. K. Tests for parameter instability and structural change with unknown change point. Econometrica 61 , 821–856 (1993).

Article   MathSciNet   Google Scholar  

Chow, G. C. Tests of equality between sets of coefficients in two linear regressions. Econometrica 28 , 591–605 (1960).

Hollander, M., A. Wolfe, D. & Chicken, E. Nonparametric Statistical Methods : Hollander/Nonparametric Statistical Methods (John Wiley & Sons, Inc., Hoboken, NJ, USA, 2015).

Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J. Royal Stat. Soc. Ser. B (Methodological) 57 , 289–300 (1995).

MathSciNet   MATH   Google Scholar  

Download references

Acknowledgements

The computing resources were granted by RWTH Aachen University under project rwth0260. S.S.S. was supported by funding from CompSE profile area, RWTH Aachen University. We wish to thank the anonymous reviewers whose constructive comments helped to improve the manuscript.

Author information

Satya S. Samal

Present address: BASF SE, Carl-Bosch-Strasse 38, 67056, Ludwigshafen am Rhein, Germany

Pejman F. Ghalati and Satya S. Samal contributed equally.

Authors and Affiliations

Joint Research Center for Computational Biomedicine, RWTH Aachen University, 52074, Aachen, Germany

Pejman F. Ghalati, Satya S. Samal, Jayesh S. Bhat & Andreas Schuppert

Klinik für Operative Intensivmedizin und Intermediate Care, Universitätsklinikum Aachen, 52074, Aachen, Germany

Robert Deisz & Gernot Marx

You can also search for this author in PubMed   Google Scholar

Contributions

P.F.G. and S.S.S. developed the idea, conducted the research, and implemented the algorithms. J.S.B. helped in the preparation of the data and in proofreading of the article. R.D. and G.M. provided the clinical insights and interpreted the findings. A.S. supervised and supported the research project. All authors have reviewed the manuscript.

Corresponding author

Correspondence to Andreas Schuppert .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information, 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 license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license 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 license, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Ghalati, P.F., Samal, S.S., Bhat, J.S. et al. Critical Transitions in Intensive Care Units: A Sepsis Case Study. Sci Rep 9 , 12888 (2019). https://doi.org/10.1038/s41598-019-49006-2

Download citation

Received : 21 February 2019

Accepted : 14 August 2019

Published : 09 September 2019

DOI : https://doi.org/10.1038/s41598-019-49006-2

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

This article is cited by

Ground truth labels challenge the validity of sepsis consensus definitions in critical illness.

• Holger A. Lindner
• Shigehiko Schamoni
• Verena Schneider-Lindner

Journal of Translational Medicine (2022)

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: Translational Research newsletter — top stories in biotechnology, drug discovery and pharma.

Advertisement

• Previous Issue
• Previous Article
• Next Article

Case Studies in Pediatric Critical Care

• Split-Screen
• Article contents
• Figures & tables
• Supplementary Data
• Peer Review
• Open the PDF for in another window
• Cite Icon Cite
• Get Permissions
• Search Site

Vanessa G. Henke , Edward A. Bittner , Michael J. Avram; Case Studies in Pediatric Critical Care. Anesthesiology 2010; 113:504 doi: https://doi.org/10.1097/ALN.0b013e3181e4f99e

Download citation file:

• Ris (Zotero)
• Reference Manager

Case Studies in Pediatric Critical Care.   Edited by Peter J. Murray, M.B., Ch.B., D.A., F.R.C.A., Stephen C. Marriage, F.R.C.P., and Peter J. Davis, M.D., F.A.A.P. Cambridge, United Kingdom, Cambridge University Press, 2009. Pages: 335. Price: $70.00.

As the old saying goes, “there is no substitute for experience.” This is certainly true in the specialized field of pediatric critical care, where the clinical reasoning and management skills characteristic of an expert intensivist are the product of extensive experience overlaid on a strong grounding in physiology, pharmacology, and evidence-based medicine. Case Studies in Pediatric Critical Care  , with its concise, practically oriented chapters, provides the intensivist-in-training (or practicing intensivist) with expert insight into the diagnosis and management of a variety of important problems in pediatric critical care.

Each of the book's 27 chapters, written by an international group of experts, leads with a brief introduction, followed by a case history detailing the presentation, symptomatology, results of investigations, decision-making approaches, and management of the patient in the intensive care unit and beyond. The case mix strikes a good balance between pathologies that are common ( e.g.  , respiratory syncytial virus bronchiolitis, diabetic ketoacidosis, and sepsis in the bone marrow transplant recipient) and unusual ( e.g.  , dengue hemorrhagic fever, management of the patient with a failing Fontan repair, and refractory narrow complex tachycardia in infancy). Each case presentation is followed by a discussion of the approach to workup and management (which includes perioperative considerations for surgical cases), a brief conclusion, a list of essential learning points, and key references.

What distinguishes this book is its reader-friendly format and a consistent emphasis on imparting practical knowledge in the context of examples and evidence. Because the authors clearly explain the rationale for their management decisions ( e.g.  , for transfusion, choice of ventilator settings, or the decision to institute and later withdraw extracorporeal membrane oxygenation), each case history creates a memorable, vivid example to help guide the reader's own decision-making. The discussion sections bolster this framework for decision-making with epidemiologic data, information about prognostic factors, and succinct summaries of recent evidence. Another unique feature of this book is the longitudinal perspective imparted by following the course of each critically ill child from intensive care unit admission through discharge, and sometimes for several months thereafter. This perspective provides a more realistic view of the “highs and lows” that often characterize the course of recovery from critical illness.

One suggestion for the next edition would be to present the cases in a logical progression, perhaps organized by primary organ system dysfunction ( e.g.  , the five cases involving management of patients with congenital heart disease might be grouped together), with greater coordination of content across chapters. Another suggestion would be to minimize the overlap between cases, such as that in Chapter 11, “Critical Care for a Child with 80% Burns,” and that in Chapter 23, “The Child with Thermal Injury and Smoke Inhalation,” by consolidating information relevant to both cases in an appendix or in a shared introductory section. Similarly, some repetition of topics and tables presented in both Chapter 5, “Child with a Head Injury,” and Chapter 24, “A Child with Multiple Trauma,” could be eliminated. Overall, the book makes excellent use of figures and tables, but several chapters could benefit from greater use of graphics to highlight salient aspects of the case, pathophysiology, workup, and management. Finally, the scope of each chapter could be more uniform because the discussion sections (most of which run approximately 5 pages) range from 3 to 14 pages. These relatively minor suggestions do not significantly diminish the overall quality of the book.

Overall, this text offers the reader a strong framework for approaching the complex diagnostic, therapeutic, and even ethical challenges in pediatric critical care. It is refreshing to read a text that is so accessible and so deftly addresses the technical and humanistic challenges of caring for critically ill children. Case Studies in Pediatric Critical Care   is an excellent resource for the intensivist-in-training, as well as for the fully trained practitioner interested in viewing the management of important intensive care unit problems from the vantage point of well-constructed case studies.

*Massachusetts General Hospital, Boston, Massachusetts. [email protected]

Citing articles via

Most viewed, email alerts, related articles, social media, affiliations.

• ASA Practice Parameters
• Online First
• Author Resource Center
• About the Journal
• Editorial Board
• Rights & Permissions
• Online ISSN 1528-1175
• Print ISSN 0003-3022
• Anesthesiology
• ASA Monitor

• Terms & Conditions Privacy Policy
• Manage Cookie Preferences
• © Copyright 2024 American Society of Anesthesiologists

This Feature Is Available To Subscribers Only

Sign In or Create an Account

• Remote Access
• Save figures into PowerPoint
• Download tables as PDFs

• Front Matter
• Contributors
• Acknowledgments
• Introduction
• How to Approach Clinical Problems

Case Files: Critical Care 2e

Author(s): Eugene C. Toy; Terrence H. Liu; Manuel Suarez

Introduction & Fundamentals

• 1 Early Awareness of Critical Illness Early Awareness of Critical Illness
• 7 Ethics in Critical Care Ethics in Critical Care
• 4 Hemodynamic Monitoring of the Unstable Patient Hemodynamic Monitoring of the Unstable Patient
• 6 Imaging in Critical Care Imaging in Critical Care
• 3 Scoring Systems and Patient Prognosis Scoring Systems and Patient Prognosis
• 2 Transfer of Critically Ill Patients Transfer of Critically Ill Patients
• 5 Vasopressor Drugs and Pharmacology Vasopressor Drugs and Pharmacology

Pulmonary-CC

• 8 Airway Management/Respiratory Failure Airway Management/Respiratory Failure
• 11 Asthmatic Exacerbation Asthmatic Exacerbation
• 13 DVT/Pulmonary Embolism DVT/Pulmonary Embolism
• 12 Noninvasive Ventilator (NIV) Support for Hypoxic Respiratory Failure Noninvasive Ventilator (NIV) Support for Hypoxic Respiratory Failure
• 10 Respiratory Weaning Respiratory Weaning
• 9 Ventilator Management Ventilator Management

Cardiovascular-CC

• 16 Acute Cardiac Failure/Cardiogenic Shock Acute Cardiac Failure/Cardiogenic Shock
• 14 Acute Coronary Syndrome Acute Coronary Syndrome
• 17 Meningitis/Encephalitis Meningitis/Encephalitis
• 15 Supraventricular Tachycardia (SVT) Supraventricular Tachycardia (SVT)
• 18 Antimicrobial Use in ICU Antimicrobial Use in ICU
• 19 Sepsis Sepsis
• 20 Sepsis in the Immune-Compromised Patient Sepsis in the Immune-Compromised Patient

Gastrointestinal & Renal

• 24 Acid–Base Disorders I Acid–Base Disorders I
• 25 Acid–Base Disorders II Acid–Base Disorders II
• 23 Acute Kidney Injury Acute Kidney Injury
• 22 Acute Liver Failure Acute Liver Failure
• 26 Fluid/Electrolyte Abnormalities Fluid/Electrolyte Abnormalities
• 21 Gastrointestinal Bleeding Gastrointestinal Bleeding
• 28 Blunt Trauma Blunt Trauma
• 29 Trauma and Burns Trauma and Burns
• 27 Traumatic Brain Injury Traumatic Brain Injury

Central Nervous System

• 30 Altered Mental Status Altered Mental Status
• 33 Multiorgan Dysfunction Multiorgan Dysfunction
• 31 Status Epilepticus Status Epilepticus
• 32 Stroke Stroke

Endocrine-CC

• 34 Endocrinopathies in the ICU Patient Endocrinopathies in the ICU Patient

Obstetrical

• 36 Hyperemesis Gravidarum and Obstetric Emergencies Less than 26 Weeks’ Gestation Hyperemesis Gravidarum and Obstetric Emergencies Less than 26 Weeks’ Gestation
• 35 Preeclampsia With Severe Features Preeclampsia With Severe Features

Miscellaneous-CC

• 41 Hemorrhage and Coagulopathy Hemorrhage and Coagulopathy
• 42 Nutritional Issues in ICU Nutritional Issues in ICU
• 38 Pain Control Pain Control
• 37 Poisoning Poisoning
• 40 Postoperative Care in ICU Postoperative Care in ICU
• 39 Post-Resuscitation Management in the ICU Post-Resuscitation Management in the ICU
• Acid–Base Disorders I
• Acid–Base Disorders II
• Acute Cardiac Failure/Cardiogenic Shock
• Acute Coronary Syndrome
• Acute Kidney Injury
• Acute Liver Failure
• Airway Management/Respiratory Failure
• Altered Mental Status
• Antimicrobial Use in ICU
• Asthmatic Exacerbation
• Blunt Trauma
• DVT/Pulmonary Embolism
• Early Awareness of Critical Illness
• Endocrinopathies in the ICU Patient
• Ethics in Critical Care
• Fluid/Electrolyte Abnormalities
• Gastrointestinal Bleeding
• Hemodynamic Monitoring of the Unstable Patient
• Hemorrhage and Coagulopathy
• Hyperemesis Gravidarum and Obstetric Emergencies Less than 26 Weeks’ Gestation
• Imaging in Critical Care
• Meningitis/Encephalitis
• Multiorgan Dysfunction
• Noninvasive Ventilator (NIV) Support for Hypoxic Respiratory Failure
• Nutritional Issues in ICU
• Pain Control
• Postoperative Care in ICU
• Post-Resuscitation Management in the ICU
• Preeclampsia With Severe Features
• Respiratory Weaning
• Scoring Systems and Patient Prognosis
• Sepsis in the Immune-Compromised Patient
• Status Epilepticus
• Supraventricular Tachycardia (SVT)
• Transfer of Critically Ill Patients
• Trauma and Burns
• Traumatic Brain Injury
• Vasopressor Drugs and Pharmacology
• Ventilator Management