Personalized mechanical ventilation guided by ultrasound in patients with acute respiratory distress syndrome (PEGASUS): study protocol for an international randomized clinical trial.

BACKGROUND: Acute respiratory distress syndrome (ARDS) is a frequent cause of hypoxemic respiratory failure with a mortality rate of approximately 30%. Identifying ARDS subphenotypes based on "focal" or "non-focal" lung morphology has the potential to better target mechanical ventilation strategies of individual patients. However, classifying morphology through chest radiography or computed tomography is either inaccurate or impractical. Lung ultrasound (LUS) is a non-invasive bedside tool that can accurately distinguish "focal" from "non-focal" lung morphology. We hypothesize that LUS-guided personalized mechanical ventilation in ARDS patients leads to a reduction in 90-day mortality compared to conventional mechanical ventilation. METHODS: The Personalized Mechanical Ventilation Guided by UltraSound in Patients with Acute Respiratory Distress Syndrome (PEGASUS) study is an investigator-initiated, international, randomized clinical trial (RCT) that plans to enroll 538 invasively ventilated adult intensive care unit (ICU) patients with moderate to severe ARDS. Eligible patients will receive a LUS exam to classify lung morphology as "focal" or "non-focal". Thereafter, patients will be randomized within 12 h after ARDS diagnosis to receive standard care or personalized ventilation where the ventilation strategy is adjusted to the morphology subphenotype, i.e., higher positive end-expiratory pressure (PEEP) and recruitment maneuvers for "non-focal" ARDS and lower PEEP and prone positioning for "focal" ARDS. The primary endpoint is all-cause mortality at day 90. Secondary outcomes are mortality at day 28, ventilator-free days at day 28, ICU length of stay, ICU mortality, hospital length of stay, hospital mortality, and number of complications (ventilator-associated pneumonia, pneumothorax, and need for rescue therapy). After a pilot phase of 80 patients, the correct interpretation of LUS images and correct application of the intervention within the safe limits of mechanical ventilation will be evaluated. DISCUSSION: PEGASUS is the first RCT that compares LUS-guided personalized mechanical ventilation with conventional ventilation in invasively ventilated patients with moderate and severe ARDS. If this study demonstrates that personalized ventilation guided by LUS can improve the outcomes of ARDS patients, it has the potential to shift the existing one-size-fits-all ventilation strategy towards a more individualized approach. TRIAL REGISTRATION: The PEGASUS trial was registered before the inclusion of the first patient, https://clinicaltrials.gov/ (ID: NCT05492344).

Reinforcement learning for intensive care medicine: actionable clinical insights from novel approaches to reward shaping and off-policy model evaluation.

BACKGROUND: Reinforcement learning (RL) holds great promise for intensive care medicine given the abundant availability of data and frequent sequential decision-making. But despite the emergence of promising algorithms, RL driven bedside clinical decision support is still far from reality. Major challenges include trust and safety. To help address these issues, we introduce cross off-policy evaluation and policy restriction and show how detailed policy analysis may increase clinical interpretability. As an example, we apply these in the setting of RL to optimise ventilator settings in intubated covid-19 patients. METHODS: With data from the Dutch ICU Data Warehouse and using an exhaustive hyperparameter grid search, we identified an optimal set of Dueling Double-Deep Q Network RL models. The state space comprised ventilator, medication, and clinical data. The action space focused on positive end-expiratory pressure (peep) and fraction of inspired oxygen (FiO2) concentration. We used gas exchange indices as interim rewards, and mortality and state duration as final rewards. We designed a novel evaluation method called cross off-policy evaluation (OPE) to assess the efficacy of models under varying weightings between the interim and terminal reward components. In addition, we implemented policy restriction to prevent potentially hazardous model actions. We introduce delta-Q to compare physician versus policy action quality and in-depth policy inspection using visualisations. RESULTS: We created trajectories for 1118 intensive care unit (ICU) admissions and trained 69,120 models using 8 model architectures with 128 hyperparameter combinations. For each model, policy restrictions were applied. In the first evaluation step, 17,182/138,240 policies had good performance, but cross-OPE revealed suboptimal performance for 44% of those by varying the reward function used for evaluation. Clinical policy inspection facilitated assessment of action decisions for individual patients, including identification of action space regions that may benefit most from optimisation. CONCLUSION: Cross-OPE can serve as a robust evaluation framework for safe RL model implementation by identifying policies with good generalisability. Policy restriction helps prevent potentially unsafe model recommendations. Finally, the novel delta-Q metric can be used to operationalise RL models in clinical practice. Our findings offer a promising pathway towards application of RL in intensive care medicine and beyond.

Does Reinforcement Learning Improve Outcomes for Critically Ill Patients? A Systematic Review and Level-of-Readiness Assessment.

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.

Red Blood Cell Transfusion in the Intensive Care Unit.

Importance: Red blood cell (RBC) transfusion is common among patients admitted to the intensive care unit (ICU). Despite multiple randomized clinical trials of hemoglobin (Hb) thresholds for transfusion, little is known about how these thresholds are incorporated into current practice. Objective: To evaluate and describe ICU RBC transfusion practices worldwide. Design, Setting, and Participants: International, prospective, cohort study that involved 3643 adult patients from 233 ICUs in 30 countries on 6 continents from March 2019 to October 2022 with data collection in prespecified weeks. Exposure: ICU stay. Main Outcomes and Measures: The primary outcome was the occurrence of RBC transfusion during ICU stay. Additional outcomes included the indication(s) for RBC transfusion (consisting of clinical reasons and physiological triggers), the stated Hb threshold and actual measured Hb values before and after an RBC transfusion, and the number of units transfused. Results: Among 3908 potentially eligible patients, 3643 were included across 233 ICUs (median of 11 patients per ICU [IQR, 5-20]) in 30 countries on 6 continents. Among the participants, the mean (SD) age was 61 (16) years, 62% were male (2267/3643), and the median Sequential Organ Failure Assessment score was 3.2 (IQR, 1.5-6.0). A total of 894 patients (25%) received 1 or more RBC transfusions during their ICU stay, with a median total of 2 units per patient (IQR, 1-4). The proportion of patients who received a transfusion ranged from 0% to 100% across centers, from 0% to 80% across countries, and from 19% to 45% across continents. Among the patients who received a transfusion, a total of 1727 RBC transfusions were administered, wherein the most common clinical indications were low Hb value (n = 1412 [81.8%]; mean [SD] lowest Hb before transfusion, 7.4 [1.2] g/dL), active bleeding (n = 479; 27.7%), and hemodynamic instability (n = 406 [23.5%]). Among the events with a stated physiological trigger, the most frequently stated triggers were hypotension (n = 728 [42.2%]), tachycardia (n = 474 [27.4%]), and increased lactate levels (n = 308 [17.8%]). The median lowest Hb level on days with an RBC transfusion ranged from 5.2 g/dL to 13.1 g/dL across centers, from 5.3 g/dL to 9.1 g/dL across countries, and from 7.2 g/dL to 8.7 g/dL across continents. Approximately 84% of ICUs administered transfusions to patients at a median Hb level greater than 7 g/dL. Conclusions and Relevance: RBC transfusion was common in patients admitted to ICUs worldwide between 2019 and 2022, with high variability across centers in transfusion practices.

Effect of lung ultrasound-guided fluid deresuscitation on duration of ventilation in intensive care unit patients (CONFIDENCE): protocol for a multicentre randomised controlled trial.

BACKGROUND: Fluid therapy is a common intervention in critically ill patients. It is increasingly recognised that deresuscitation is an essential part of fluid therapy and delayed deresuscitation is associated with longer invasive ventilation and length of intensive care unit (ICU) stay. However, optimal timing and rate of deresuscitation remain unclear. Lung ultrasound (LUS) may be used to identify fluid overload. We hypothesise that daily LUS-guided deresuscitation is superior to deresuscitation without LUS in critically ill patients expected to undergo invasive ventilation for more than 24 h in terms of ventilator free-days and being alive at day 28. METHODS: The "effect of lung ultrasound-guided fluid deresuscitation on duration of ventilation in intensive care unit patients" (CONFIDENCE) is a national, multicentre, open-label, randomised controlled trial (RCT) in adult critically ill patients that are expected to be invasively ventilated for at least 24 h. Patients with conditions that preclude a negative fluid balance or LUS examination are excluded. CONFIDENCE will operate in 10 ICUs in the Netherlands and enrol 1000 patients. After hemodynamic stabilisation, patients assigned to the intervention will receive daily LUS with fluid balance recommendations. Subjects in the control arm are deresuscitated at the physician's discretion without the use of LUS. The primary endpoint is the number of ventilator-free days and being alive at day 28. Secondary endpoints include the duration of invasive ventilation; 28-day mortality; 90-day mortality; ICU, in hospital and total length of stay; cumulative fluid balance on days 1-7 after randomisation and on days 1-7 after start of LUS examination; mean serum lactate on days 1-7; the incidence of reintubations, chest drain placement, atrial fibrillation, kidney injury (KDIGO stadium ≥ 2) and hypernatremia; the use of invasive hemodynamic monitoring, and chest-X-ray; and quality of life at day 28. DISCUSSION: The CONFIDENCE trial is the first RCT comparing the effect of LUS-guided deresuscitation to routine care in invasively ventilated ICU patients. If proven effective, LUS-guided deresuscitation could improve outcomes in some of the most vulnerable and resource-intensive patients in a manner that is non-invasive, easy to perform, and well-implementable. TRIAL REGISTRATION: ClinicalTrials.gov NCT05188092. Registered since January 12, 2022.

A structured diagnostic algorithm for patients with ARDS.

This article is one of ten reviews selected from the Annual Update in Intensive Care and Emergency Medicine 2023. Other selected articles can be found online at  https://www.biomedcentral.com/collections/annualupdate2023 . Further information about the Annual Update in Intensive Care and Emergency Medicine is available from  https://link.springer.com/bookseries/8901 .

Diagnosing acute kidney injury ahead of time in critically ill septic patients using kinetic estimated glomerular filtration rate.

INTRODUCTION: Accurate and actionable diagnosis of Acute Kidney Injury (AKI) ahead of time is important to prevent or mitigate renal insufficiency. The purpose of this study was to evaluate the performance of Kinetic estimated Glomerular Filtration Rate (KeGFR) in timely predicting AKI in critically ill septic patients. METHODS: We conducted a retrospective analysis on septic ICU patients who developed AKI in AmsterdamUMCdb, the first freely available European ICU database. The reference standard for AKI was the Kidney Disease: Improving Global Outcomes (KDIGO) classification based on serum creatinine and urine output (UO). Prediction of AKI was based on stages defined by KeGFR and UO. Classifications were compared by length of ICU stay (LOS), need for renal replacement therapy and 28-day mortality. Predictive performance and time between prediction and diagnosis were calculated. RESULTS: Of 2492 patients in the cohort, 1560 (62.0%) were diagnosed with AKI by KDIGO and 1706 (68.5%) by KeGFR criteria. Disease stages had agreement of kappa = 0.77, with KeGFR sensitivity 93.2%, specificity 73.0% and accuracy 85.7%. Median time to recognition of AKI Stage 1 was 13.2 h faster for KeGFR, and 7.5 h and 5.0 h for Stages 2 and 3. Outcomes revealed a slight difference in LOS and 28-day mortality for Stage 1. CONCLUSIONS: Predictive performance of KeGFR combined with UO criteria for diagnosing AKI is excellent. Compared to KDIGO, deterioration of renal function was identified earlier, most prominently for lower stages of AKI. This may shift the actionable window for preventing and mitigating renal insufficiency.

Lung Ultrasound Prediction Model for Acute Respiratory Distress Syndrome: A Multicenter Prospective Observational Study.

Rationale: Lung ultrasound (LUS) is a promising tool for diagnosis of acute respiratory distress syndrome (ARDS), but adequately sized studies with external validation are lacking. Objectives: To develop and validate a data-driven LUS score for diagnosis of ARDS and compare its performance with that of chest radiography (CXR). Methods: This multicenter prospective observational study included invasively ventilated ICU patients who were divided into a derivation cohort and a validation cohort. Three raters scored ARDS according to the Berlin criteria, resulting in a classification of "certain no ARDS," or "certain ARDS" when experts agreed or "uncertain ARDS" when evaluations conflicted. Uncertain cases were classified in a consensus meeting. Results of a 12-region LUS exam were used in a logistic regression model to develop the LUS-ARDS score. Measurements and Main Results: Three hundred twenty-four (16% certain ARDS) and 129 (34% certain ARDS) patients were included in the derivation cohort and the validation cohort, respectively. With an ARDS diagnosis by the expert panel as the reference test, the LUS-ARDS score, including the left and right LUS aeration scores and anterolateral pleural line abnormalities, had an area under the receiver operating characteristic (ROC) curve of 0.90 (95% confidence interval [CI], 0.85-0.95) in certain patients of the derivation cohort and 0.80 (95% CI, 0.72-0.87) in all patients of the validation cohort. Within patients who had imaging-gold standard chest computed tomography available, diagnostic accuracy of eight independent CXR readers followed the ROC curve of the LUS-ARDS score. Conclusions: The LUS-ARDS score can be used to accurately diagnose ARDS also after external validation. The LUS-ARDS score may be a useful adjunct to a diagnosis of ARDS after further validation, as it showed performance comparable with that of the current practice with experienced CXR readers but more objectifiable diagnostic accuracy at each cutoff.

Oxygenation thresholds for invasive ventilation in hypoxemic respiratory failure: a target trial emulation in two cohorts.

BACKGROUND: The optimal thresholds for the initiation of invasive ventilation in patients with hypoxemic respiratory failure are unknown. Using the saturation-to-inspired oxygen ratio (SF), we compared lower versus higher hypoxemia severity thresholds for initiating invasive ventilation. METHODS: This target trial emulation included patients from the Medical Information Mart for Intensive Care (MIMIC-IV, 2008-2019) and the Amsterdam University Medical Centers (AmsterdamUMCdb, 2003-2016) databases admitted to intensive care and receiving inspired oxygen fraction ≥ 0.4 via non-rebreather mask, noninvasive ventilation, or high-flow nasal cannula. We compared the effect of using invasive ventilation initiation thresholds of SF < 110, < 98, and < 88 on 28-day mortality. MIMIC-IV was used for the primary analysis and AmsterdamUMCdb for the secondary analysis. We obtained posterior means and 95% credible intervals (CrI) with nonparametric Bayesian G-computation. RESULTS: We studied 3,357 patients in the primary analysis. For invasive ventilation initiation thresholds SF < 110, SF < 98, and SF < 88, the predicted 28-day probabilities of invasive ventilation were 72%, 47%, and 19%. Predicted 28-day mortality was lowest with threshold SF < 110 (22.2%, CrI 19.2 to 25.0), compared to SF < 98 (absolute risk increase 1.6%, CrI 0.6 to 2.6) or SF < 88 (absolute risk increase 3.5%, CrI 1.4 to 5.4). In the secondary analysis (1,279 patients), the predicted 28-day probability of invasive ventilation was 50% for initiation threshold SF < 110, 28% for SF < 98, and 19% for SF < 88. In contrast with the primary analysis, predicted mortality was highest with threshold SF < 110 (14.6%, CrI 7.7 to 22.3), compared to SF < 98 (absolute risk decrease 0.5%, CrI 0.0 to 0.9) or SF < 88 (absolute risk decrease 1.9%, CrI 0.9 to 2.8). CONCLUSION: Initiating invasive ventilation at lower hypoxemia severity will increase the rate of invasive ventilation, but this can either increase or decrease the expected mortality, with the direction of effect likely depending on baseline mortality risk and clinical context.

Development and validation of a multivariable model predicting the required catheter dwell time among mechanically ventilated critically ill patients in three randomized trials.

BACKGROUND: The anatomic site for central venous catheter insertion influences the risk of central venous catheter-related intravascular complications. We developed and validated a predictive score of required catheter dwell time to identify critically ill patients at higher risk of intravascular complications. METHODS: We retrospectively conducted a cohort study from three multicenter randomized controlled trials enrolling consecutive patients requiring central venous catheterization. The primary outcome was the required catheter dwell time, defined as the period between the first catheter insertion and removal of the last catheter for absence of utility. Predictors were identified in the training cohort (3SITES trial; 2336 patients) through multivariable analyses based on the subdistribution hazard function accounting for death as a competing event. Internal validation was performed in the training cohort by 500 bootstraps to derive the CVC-IN score from robust risk factors. External validation of the CVC-IN score were performed in the testing cohort (CLEAN, and DRESSING2; 2371 patients). RESULTS: The analysis was restricted to patients requiring mechanical ventilation to comply with model assumptions. Immunosuppression (2 points), high creatinine > 100 micromol/L (2 points), use of vasopressor (1 point), obesity (1 point) and older age (40-59, 1 point; ≥ 60, 2 points) were independently associated with the required catheter dwell time. At day 28, area under the ROC curve for the CVC-IN score was 0.69, 95% confidence interval (CI) [0.66-0.72] in the training cohort and 0.64, 95% CI [0.61-0.66] in the testing cohort. Patients with a CVC-IN score ≥ 4 in the overall cohort had a median required catheter dwell time of 24 days (versus 11 days for CVC-IN score < 4 points). The positive predictive value of a CVC-IN score ≥ 4 was 76.9% for > 7 days required catheter dwell time in the testing cohort. CONCLUSION: The CVC-IN score, which can be used for the first catheter, had a modest ability to discriminate required catheter dwell time. Nevertheless, preference of the subclavian site may contribute to limit the risk of intravascular complications, in particular among ventilated patients with high CVC-IN score. Trials Registration NCT01479153, NCT01629550, NCT01189682.

Concise Versus Extended Lung Ultrasound Score to Monitor Critically Ill Patients With COVID-19.

BACKGROUND: Lung ultrasound (LUS) can be used to monitor critically ill patients with COVID-19, but the optimal number of examined lung zones is disputed. METHODS: This was a prospective observational study. The objective was to investigate whether concise (6 zones) and extended (12 zones) LUS scoring protocols are clinically equivalent in critically ill ICU subjects with COVID-19. The primary outcome of this study was (statistical) agreement between concise and extended LUS score index evaluated in both supine and prone position. Agreement was determined using correlation coefficients and Bland-Altman plots to detect systematic differences between protocols. Secondary outcomes were difference between LUS score index in supine and prone position using similar methods. RESULTS: We included 130 LUS examinations in 40 subjects (mean age 69.0 ± 8.5y, 75% male). Agreement between concise and extended LUS score index had no clinically relevant constant or proportional bias and limits of agreement were below the smallest detectable change. Across position changes, supine LUS score index was 8% higher than prone LUS score index and had limits above the smallest detectable change, indicating true LUS score index differences between protocols may occur due to the position change itself. Lastly, inter-rater and intra-rater agreement were very good. CONCLUSIONS: Concise LUS was equally informative as extended LUS for monitoring critically ill subjects with COVID-19 in supine or prone position. Clinicians can monitor patients undergoing position changes but must be wary that LUS score index alterations may result from the position change itself rather than disease progression or clinical improvement.