Search

Objective: Reinforcement learning (RL) is a machine learning technique uniquely effective at sequential decision-making, which makes it potentially relevant to ICU treatment challenges. We set out to systematically review, assess level-of-readiness and meta-analyze the effect of RL on outcomes for critically ill patients., Data Sources: A systematic search was performed in PubMed, Embase.com, Clarivate Analytics/Web of Science Core Collection, Elsevier/SCOPUS and the Institute of Electrical and Electronics Engineers Xplore Digital Library from inception to March 25, 2022, with subsequent citation tracking., Data Extraction: Journal articles that used an RL technique in an ICU population and reported on patient health-related outcomes were included for full analysis. Conference papers were included for level-of-readiness assessment only. Descriptive statistics, characteristics of the models, outcome compared with clinician's policy and level-of-readiness were collected. RL-health risk of bias and applicability assessment was performed., Data Synthesis: A total of 1,033 articles were screened, of which 18 journal articles and 18 conference papers, were included. Thirty of those were prototyping or modeling articles and six were validation articles. All articles reported RL algorithms to outperform clinical decision-making by ICU professionals, but only in retrospective data. The modeling techniques for the state-space, action-space, reward function, RL model training, and evaluation varied widely. The risk of bias was high in all articles, mainly due to the evaluation procedure., Conclusion: In this first systematic review on the application of RL in intensive care medicine we found no studies that demonstrated improved patient outcomes from RL-based technologies. All studies reported that RL-agent policies outperformed clinician policies, but such assessments were all based on retrospective off-policy evaluation., Competing Interests: Dr. Dam’s institution received funding from ZonMW/Netherlands Organization for Health Research and Development (10430012010003); he received funding from Pacmed BV. Dr. Hengst received funding from ING Bank N.V. Dr. Hoogendoorn disclosed co-ownership of PersonalAIze B.V. The remaining authors have disclosed that they do not have any potential conflicts of interest., (Copyright © 2023 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.)

Objectives: To determine the diagnostic accuracy of lung ultrasound signs for both the diagnosis of interstitial syndrome and for the discrimination of noncardiogenic interstitial syndrome (NCIS) from cardiogenic pulmonary edema (CPE) in a mixed ICU population., Design: A prospective diagnostic accuracy study with derivation and validation cohorts., Setting: Three academic mixed ICUs in the Netherlands., Patients: Consecutive adult ICU patients that received a lung ultrasound examination., Interventions: None., Measurements and Main Result: The reference standard was the diagnosis of interstitial syndrome (NCIS or CPE) or noninterstitial syndromes (other pulmonary diagnoses and no pulmonary diagnoses) based on full post-hoc clinical chart review except lung ultrasound. The index test was a lung ultrasound examination performed and scored by a researcher blinded to clinical information. A total of 101 patients were included in the derivation and 122 in validation cohort. In the derivation cohort, patients with interstitial syndrome ( n = 56) were reliably discriminated from other patients based on the presence of a B-pattern (defined as greater than or equal to 3 B-lines in one frame) with an accuracy of 94.7% (sensitivity, 90.9%; specificity, 91.1%). For discrimination of NCIS ( n = 29) from CPE ( n = 27), the presence of bilateral pleural line abnormalities (at least two: fragmented, thickened or irregular) had the highest diagnostic accuracy (94.6%; sensitivity, 89.3%; specificity, 100%). A diagnostic algorithm (Bedside Lung Ultrasound for Interstitial Syndrome Hierarchy protocol) using B-pattern and bilateral pleural abnormalities had an accuracy of 0.86 (95% CI, 0.77-0.95) for diagnosis and discrimination of interstitial syndromes. In the validation cohort, which included 122 patients with interstitial syndrome, bilateral pleural line abnormalities discriminated NCIS ( n = 98) from CPE ( n = 24) with a sensitivity of 31% (95% CI, 21-40%) and a specificity of 100% (95% CI, 86-100%)., Conclusions: Lung ultrasound can diagnose and discriminate interstitial syndromes in ICU patients with moderate-to-good accuracy. Pleural line abnormalities are highly specific for NCIS, but sensitivity is limited., Competing Interests: Dr. Tuinman received support for article research from departmental funds. The remaining authors have disclosed that they do not have any potential conflicts of interest., (Copyright © 2022 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.)

Objectives: To determine the diagnostic accuracy of extended lung ultrasonographic assessment, including evaluation of dynamic air bronchograms and color Doppler imaging to differentiate pneumonia and atelectasis in patients with consolidation on chest radiograph. Compare this approach to the Simplified Clinical Pulmonary Infection Score, Lung Ultrasound Clinical Pulmonary Infection Score, and the Bedside Lung Ultrasound in Emergency protocol., Design: Prospective diagnostic accuracy study., Setting: Adult ICU applying selective digestive decontamination., Patients: Adult patients that underwent a chest radiograph for any indication at any time during admission. Patients with acute respiratory distress syndrome, coronavirus disease 2019, severe thoracic trauma, and infectious isolation measures were excluded., Interventions: None., Measurements and Main Results: Lung ultrasound was performed within 24 hours of chest radiograph. Consolidated tissue was assessed for presence of dynamic air bronchograms and with color Doppler imaging for presence of flow. Clinical data were recorded after ultrasonographic assessment. The primary outcome was diagnostic accuracy of dynamic air bronchogram and color Doppler imaging alone and within a decision tree to differentiate pneumonia from atelectasis. Of 120 patients included, 51 (42.5%) were diagnosed with pneumonia. The dynamic air bronchogram had a 45% (95% CI, 31-60%) sensitivity and 99% (95% CI, 92-100%) specificity. Color Doppler imaging had a 90% (95% CI, 79-97%) sensitivity and 68% (95% CI, 56-79%) specificity. The combined decision tree had an 86% (95% CI, 74-94%) sensitivity and an 86% (95% CI, 75-93%) specificity. The Bedside Lung Ultrasound in Emergency protocol had a 100% (95% CI, 93-100%) sensitivity and 0% (95% CI, 0-5%) specificity, while the Simplified Clinical Pulmonary Infection Score and Lung Ultrasound Clinical Pulmonary Infection Score had a 41% (95% CI, 28-56%) sensitivity, 84% (95% CI, 73-92%) specificity and 68% (95% CI, 54-81%) sensitivity, 81% (95% CI, 70-90%) specificity, respectively., Conclusions: In critically ill patients with pulmonary consolidation on chest radiograph, an extended lung ultrasound protocol is an accurate and directly bedside available tool to differentiate pneumonia from atelectasis. It outperforms standard lung ultrasound and clinical scores., Competing Interests: Drs. Jonkman and Heunks received funding from Liberate Medical. Dr. Heunks also received funding from Getinge Critical Care and Fisher and Paykel. Dr. Tuinman disclosed departmental work. The remaining authors have disclosed that they do not have any potential conflicts of interest., (Copyright © 2022 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.)

Objectives: Lung- and diaphragm-protective ventilation is a novel concept that aims to limit the detrimental effects of mechanical ventilation on the diaphragm while remaining within limits of lung-protective ventilation. The premise is that low breathing effort under mechanical ventilation causes diaphragm atrophy, whereas excessive breathing effort induces diaphragm and lung injury. In a proof-of-concept study, we aimed to assess whether titration of inspiratory support based on diaphragm effort increases the time that patients have effort in a predefined "diaphragm-protective" range, without compromising lung-protective ventilation., Design: Randomized clinical trial., Setting: Mixed medical-surgical ICU in a tertiary academic hospital in the Netherlands., Patients: Patients (n = 40) with respiratory failure ventilated in a partially-supported mode., Interventions: In the intervention group, inspiratory support was titrated hourly to obtain transdiaphragmatic pressure swings in the predefined "diaphragm-protective" range (3-12 cm H2O). The control group received standard-of-care., Measurements and Main Results: Transdiaphragmatic pressure, transpulmonary pressure, and tidal volume were monitored continuously for 24 hours in both groups. In the intervention group, more breaths were within "diaphragm-protective" range compared with the control group (median 81%; interquartile range [64-86%] vs 35% [16-60%], respectively; p < 0.001). Dynamic transpulmonary pressures (20.5 ± 7.1 vs 18.5 ± 7.0 cm H2O; p = 0.321) and tidal volumes (7.56 ± 1.47 vs 7.54 ± 1.22 mL/kg; p = 0.961) were not different in the intervention and control group, respectively., Conclusions: Titration of inspiratory support based on patient breathing effort greatly increased the time that patients had diaphragm effort in the predefined "diaphragm-protective" range without compromising tidal volumes and transpulmonary pressures. This study provides a strong rationale for further studies powered on patient-centered outcomes., Competing Interests: Drs. de Vries’ and Heunks’ institutions received funding from Amsterdam Cardiovascular Sciences. Dr. de Vries has received speaker fees from the Dutch Ultrasound Center (the Netherlands) and travel and speaker fees from the Chinese Organization of Rehabilitation Medicine (China). Dr. Jonkman has received personal fees from Liberate Medical (United States). Dr. Heunks received research support from Liberate Medical (United States), Fisher and Paykel, and Orion Pharma (Finland), and speakers fee from Getinge (Sweden). Dr. de Man disclosed the off-label product use of oxidation-reduction potential measurement with the RedoxSYS System from Aytu Biosciences. The remaining authors have disclosed that they do not have any potential conflicts of interest., (Copyright © 2022 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the Society of Critical Care Medicine and Wolters Kluwer Health, Inc.)

Objectives: The optimal targeted temperature in patients with shockable rhythm is unclear, and current guidelines recommend targeted temperature management with a correspondingly wide range between 32°C and 36°C. Our aim was to study survival and neurologic outcome associated with targeted temperature management strategy in postarrest patients with initial shockable rhythm., Design: Observational substudy of the Coronary Angiography after Cardiac Arrest without ST-segment Elevation trial., Setting: Nineteen hospitals in The Netherlands., Patients: The Coronary Angiography after Cardiac Arrest trial randomized successfully resuscitated patients with shockable rhythm and absence of ST-segment elevation to a strategy of immediate or delayed coronary angiography. In this substudy, 459 patients treated with mild therapeutic hypothermia (32.0-34.0°C) or targeted normothermia (36.0-37.0°C) were included. Allocation to targeted temperature management strategy was at the discretion of the physician., Interventions: None., Measurements and Main Results: After 90 days, 171 patients (63.6%) in the mild therapeutic hypothermia group and 129 (67.9%) in the targeted normothermia group were alive (hazard ratio, 0.86 [95% CI, 0.62-1.18]; log-rank p = 0.35; adjusted odds ratio, 0.89; 95% CI, 0.45-1.72). Patients in the mild therapeutic hypothermia group had longer ICU stay (4 d [3-7 d] vs 3 d [2-5 d]; ratio of geometric means, 1.32; 95% CI, 1.15-1.51), lower blood pressures, higher lactate levels, and increased need for inotropic support. Cerebral Performance Category scores at ICU discharge and 90-day follow-up and patient-reported Mental and Physical Health Scores at 1 year were similar in the two groups., Conclusions: In the context of out-of-hospital cardiac arrest with shockable rhythm and no ST-elevation, treatment with mild therapeutic hypothermia was not associated with improved 90-day survival compared with targeted normothermia. Neurologic outcomes at 90 days as well as patient-reported Mental and Physical Health Scores at 1 year did not differ between the groups., Competing Interests: Dr. Lemkes received funding from The Netherlands Heart Institute (NHLI) and Biotronik. Drs. Lemkes and Vlachojannis received funding from AstraZeneca. Dr. Rijpstra’s institution received funding from Principle Investigator. Dr. Vlachojannis’ institution received funding from MicroPort and Daiichi Sankyo; he received funding from Abbott. Dr. Vlachojannis reports receiving grant support from MicroPort Orthopedics and Daiichi Sankyo. Dr. van Royen’s institution received funding from Biotronik, AstraZeneca, the NHLI, Abbott, and Medtronic; he received funding from Novartis, MicroPort, Castor, Rainmed, Biotronik, Abbott, Medtronic, and Philips; he received support for article research from the NLHI. The remaining authors have disclosed that they do not have any potential conflicts of interest., (Copyright © 2021 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.)

Objectives: Critical care medicine is a natural environment for machine learning approaches to improve outcomes for critically ill patients as admissions to ICUs generate vast amounts of data. However, technical, legal, ethical, and privacy concerns have so far limited the critical care medicine community from making these data readily available. The Society of Critical Care Medicine and the European Society of Intensive Care Medicine have identified ICU patient data sharing as one of the priorities under their Joint Data Science Collaboration. To encourage ICUs worldwide to share their patient data responsibly, we now describe the development and release of Amsterdam University Medical Centers Database (AmsterdamUMCdb), the first freely available critical care database in full compliance with privacy laws from both the United States and Europe, as an example of the feasibility of sharing complex critical care data., Setting: University hospital ICU., Subjects: Data from ICU patients admitted between 2003 and 2016., Interventions: We used a risk-based deidentification strategy to maintain data utility while preserving privacy. In addition, we implemented contractual and governance processes, and a communication strategy. Patient organizations, supporting hospitals, and experts on ethics and privacy audited these processes and the database., Measurements and Main Results: AmsterdamUMCdb contains approximately 1 billion clinical data points from 23,106 admissions of 20,109 patients. The privacy audit concluded that reidentification is not reasonably likely, and AmsterdamUMCdb can therefore be considered as anonymous information, both in the context of the U.S. Health Insurance Portability and Accountability Act and the European General Data Protection Regulation. The ethics audit concluded that responsible data sharing imposes minimal burden, whereas the potential benefit is tremendous., Conclusions: Technical, legal, ethical, and privacy challenges related to responsible data sharing can be addressed using a multidisciplinary approach. A risk-based deidentification strategy, that complies with both U.S. and European privacy regulations, should be the preferred approach to releasing ICU patient data. This supports the shared Society of Critical Care Medicine and European Society of Intensive Care Medicine vision to improve critical care outcomes through scientific inquiry of vast and combined ICU datasets., Competing Interests: Dr. Sijbrands’ institution received funding from European Institute of Innovation and Technology (EIT) Health and Amgen. Drs. Kaplan and Bailey received funding from Society of Critical Care Medicine. Dr. Cecconi received funding from Directed Systems, Edwards Lifesciences, and Cheetah Medical. Dr. Churpek’s institution received funding from an EarlySense research grant; he is supported by National Institutes of Health (NIH) R01 (GM123193), and he has a patent pending for risk stratification algorithm for hospitalized patients (money from royalties from the University of Chicago). Dr. Clermont received funding from the NIH, Department of Defense, National Science Foundation, and NOMA AI. The remaining authors have disclosed that they do not have any potential conflicts of interest., (Copyright © 2021 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the Society of Critical Care Medicine and Wolters Kluwer Health, Inc.)

Objectives: In critically ill patients, treatment dose or intensity is often related to severity of illness and mortality risk, whereas overtreatment or undertreatment (relative to the individual need) may further increase the odds of death. We aimed to investigate how these relationships affect the results of common statistical methods used in observational studies., Design: Using Monte Carlo simulation, we generated data for 5,000 patients with a treatment dose related to the pretreatment mortality risk but with randomly distributed overtreatment or undertreatment. Significant overtreatment or undertreatment (relative to the optimal dose) further increased the mortality risk. A prognostic score that reflects the mortality risk and an outcome of death or survival was then generated. The study was analyzed: 1) using logistic regression to estimate the effect of treatment dose on outcome while controlling for prognostic score and 2) using propensity score matching and inverse probability weighting of the effect of high treatment dose on outcome. The data generation and analyses were repeated 1,500 times over sample sizes between 200 and 30,000 patients, with an increasing accuracy of the prognostic score and with different underlying assumptions., Setting: Computer-simulated studies., Measurements and Main Results: In the simulated 5,000-patient observational study, higher treatment dose was found to be associated with increased odds of death (p = 0.00001) while controlling for the prognostic score with logistic regression. Propensity-matched analysis led to similar results. Larger sample sizes led to equally biased estimates with narrower CIs. A perfect risk predictor negated the bias only under artificially perfect assumptions., Conclusions: When a treatment dose is associated with severity of illness and should be dosed "enough," logistic regression, propensity score matching, and inverse probability weighting to adjust for confounding by severity of illness lead to biased results. Larger sample sizes lead to more precisely wrong estimates.