Daily Ards Research Analysis
An interpretable machine-learning model built from two large ICU databases achieved strong discrimination for ARDS mortality, supporting early risk stratification. An international mixed-methods study identified disease uncertainty as the dominant influence on ventilation decisions during COVID-19 ARDS. A phase I/IIa MSC infusion study in severe COVID-19 modulated inflammatory and neuroinjury biomarkers, suggesting potential immuno-neuroprotection.
Summary
An interpretable machine-learning model built from two large ICU databases achieved strong discrimination for ARDS mortality, supporting early risk stratification. An international mixed-methods study identified disease uncertainty as the dominant influence on ventilation decisions during COVID-19 ARDS. A phase I/IIa MSC infusion study in severe COVID-19 modulated inflammatory and neuroinjury biomarkers, suggesting potential immuno-neuroprotection.
Research Themes
- Interpretable predictive analytics for ARDS mortality
- Decision-making under uncertainty in pandemic ventilation management
- Cell-based immunomodulation and biomarker modulation in severe COVID-19
Selected Articles
1. An interpretable machine learning model for predicting mortality risk in adult ICU patients with acute respiratory distress syndrome.
Using MIMIC-IV and eICU-CRD, the authors developed and explained an ARDS (acute respiratory distress syndrome) mortality model, with XGBoost achieving AUC-ROC 0.887 and AUPRC 0.731. SHAP provided interpretable insights to support early identification of high-risk patients.
Impact: Large-scale, interpretable prognostic modeling across two ICU databases provides actionable risk stratification for ARDS. This can enhance triage, resource allocation, and timely interventions.
Clinical Implications: Implementable risk prediction could guide early escalation, ventilator strategies, and ICU resource prioritization for ARDS patients.
Key Findings
- Included 5,732 severe ARDS ICU patients with 20.4% mortality.
- XGBoost achieved AUC-ROC 0.887 (95% CI 0.863-0.909) and AUPRC 0.731 (95% CI 0.673-0.783).
- SHAP explained model decisions, improving interpretability for clinical use.
- Models were developed using two independent databases (MIMIC-IV and eICU-CRD).
Methodological Strengths
- Large, multi-database cohort with external validation context
- Model interpretability using SHAP with systematic feature selection and Bayesian optimization
Limitations
- Retrospective database study susceptible to residual confounding and missingness
- Generalizability beyond included health systems may be limited
Future Directions: Prospective validation, calibration drift monitoring, and integration into clinical workflows with impact evaluation are needed.
BACKGROUND: Acute respiratory distress syndrome (ARDS) is a clinical syndrome triggered by pulmonary or extra-pulmonary factors with high mortality and poor prognosis in the ICU. The aim of this study was to develop an interpretable machine learning predictive model to predict the risk of death in patients with ARDS in the ICU. METHODS: The datasets used in this study were obtained from two independent databases: Medical Information Mart for Intensive Care (MIMIC) IV and eICU Collaborative Research Database (eICU-CRD). This study used eight machine learning algorithms to construct predictive models. Recursive feature elimination with cross-validation is used to screen features, and cross-validation-based Bayesian optimization is used to filter the features used to find the optimal combination of hyperparameters for the model. The Shapley additive explanations (SHAP) method is used to explain the decision-making process of the model. RESULTS: A total of 5,732 patients with severe ADRS were included in this study for analysis, of which 1,171 patients (20.4%) did not survive. Among the eight models, XGBoost performed the best; AUC-ROC was 0.887 (95% CI: 0.863-0.909) and AUPRC was 0.731 (95% CI: 0.673-0.783). CONCLUSION: We developed a machine learning-based model for predicting the risk of death of critically ill ARDS patients in the ICU, and our model can effectively identify high-risk ARDS patients at an early stage, thereby supporting clinical decision-making, facilitating early intervention, and improving patient prognosis.
2. Uncertainty and decision-making in critical care: lessons from managing COVID-19 ARDS in preparation for the next pandemic.
A mixed-methods program (systematic review, interviews, and international survey) found that disease uncertainty was the foremost driver of ventilation decision-making in COVID-19 ARDS across countries and professions (p<0.001). Underconfidence was common and unrelated to experience, highlighting targets such as information sharing and teamworking for future preparedness.
Impact: Defines modifiable determinants of critical care decisions under uncertainty, informing system-level interventions for future pandemics.
Clinical Implications: Preparedness programs should prioritize real-time evidence sharing, multidisciplinary team processes, and resource allocation to mitigate uncertainty in ARDS ventilation decisions.
Key Findings
- Among 371 respondents, disease uncertainty ranked as the most important influence on COVID-19 ARDS ventilation decisions across regions and professions (p<0.001).
- Underconfidence in decision-making (median 9/20) was common and unaffected by experience (p=0.79) or profession (p=0.58).
- Qualitative analysis highlighted positive team factors and negative resource limitations impacting disease uncertainty.
- Literature synthesis showed patient factors were well-studied while uncertainty was understudied.
Methodological Strengths
- Multinational, multidisciplinary sample with mixed-methods triangulation
- Pre-specified thematic framework with appropriate nonparametric statistics
Limitations
- Survey self-selection and recall bias may affect generalizability
- Does not link decision processes to patient outcomes
Future Directions: Develop decision-support tools and rapid guidance mechanisms; evaluate whether improving information flow and team processes affects ARDS outcomes during surges.
PURPOSE: Coronavirus disease 2019 (COVID-19) acute respiratory distress syndrome (ARDS) was an emergent syndrome that led to high volumes of critically ill ventilated patients. We explored influences on decision-making regarding management of COVID-19 ARDS mechanical ventilation to identify modifiable factors to improve preparedness for future pandemics. METHODS: A systematic review and small group interviews informed the development of an international questionnaire (UK, Italy, Germany and Netherlands) on factors influencing COVID-19 ARDS ventilation decision-making in critical care professionals. Participants ranked four themes in order of importance: disease (uncertainties around COVID-19 ARDS), contextual (cognitive strain), environmental (structural logistics) and team factors. Participants also ranked the subthemes within each theme. Thematic analysis was used to derive findings from qualitative data. Kruskal-Wallis, Mann-Whitney U and Kendall's tau were used for quantitative data analysis. RESULTS: Patient factors (comorbidities, clinical/biochemical parameters) were the most studied influences in the extant literature on decision-making; uncertainty was one of the least studied. 371 critical care professionals responded to the questionnaire. Disease uncertainty (lack of applicable guidelines, unfamiliarity with pathophysiology) was ranked as the most important influence on ventilation decision-making for COVID-19 ARDS across regions, professions and experience levels (p<0.001). Participants expressed underconfidence in their decision-making (median score: 9/20); this was unaffected by experience (p=0.79) or profession (p=0.58). Qualitative findings supported and extended the initial proposed influences, including the impact of team factors (+ve) and resource limitations (-ve) on disease uncertainty. CONCLUSION: Future pandemic preparedness programmes should target modifiable influences such as information sharing, teamworking and resource limitations to mitigate against the negative influence of uncertainty and thereby improve decision-making overall.
3. Impact of mesenchymal stromal/stem cell infusions on circulating inflammatory biomarkers in COVID-19 patients: analysis of a phase I-IIa trial.
In a small phase I/IIa study of severe COVID-19 pneumonia, two allogeneic MSC infusions were safe and associated with favorable cytokine trajectories (attenuated IL1RA/IL18 rise, reduced IL-6) and prevention of neurofilament light chain surges versus controls. All MSC recipients were discharged on average 15 days after the second infusion.
Impact: Provides mechanistic biomarker evidence that MSCs may attenuate systemic and neuroinflammation in severe COVID-19, informing future trials and endpoints.
Clinical Implications: While not practice-changing yet, findings support testing MSCs in larger randomized trials with neurological injury biomarkers as secondary endpoints.
Key Findings
- Two MSC infusions were feasible and safe; all patients discharged on average 15 ± 3.7 days after the second infusion.
- Compared with controls, MSCs attenuated rises in IL1RA (P=0.044) and IL18 (P=0.032) and reduced IL-6 levels.
- Both groups had similar reductions in long pentraxin, but MSCs prevented neurofilament light chain surges seen in controls.
- Random-effects models showed distinct cytokine trajectories favoring MSC treatment.
Methodological Strengths
- Phase I/IIa interventional design with matched controls and longitudinal biomarker profiling
- Appropriate random-effects modeling to assess temporal cytokine trends
Limitations
- Very small sample size with nonrandomized design limits causal inference
- Historical matching may introduce temporal and treatment confounding
Future Directions: Conduct multicenter randomized trials powered for clinical outcomes and include neuroinjury biomarkers (e.g., NfL) to validate observed effects.
BACKGROUND AIMS: SARS-CoV-2 infection triggers respiratory inflammation with potentially fatal systemic effects. Mesenchymal stromal/stem cells (MSCs) are promising for treating severe COVID-19 due to their anti-inflammatory and regenerative capacities. This study investigates the effects of allogeneic MSCs in severe COVID-19 pneumonia. METHODS: In the phase I/IIa RESCAT trial (May 2021-Feb 2022), patients with severe COVID-19 pneumonia received two intravenous MSC infusions and were compared to a control group (CTRL). To assess cytokine and biomarker responses, the MSC group was matched 1:2 with standard care patients (mCTRL) by age, gender, BMI, and PaO2/FiO2 (Nov 2020-Feb 2021). Random-effects linear regression evaluated cytokine and biomarker trends over time between MSC and control groups. RESULTS: Seventeen patients (MSC = 5, CTRL = 2, mCTRL = 10) were analyzed. Two MSC infusions were feasible and safe, with all patients discharged on average 15 ± 3.7 days postsecond infusion. While IL1RA and IL18 levels significantly increased in CTRL-mCTRL patients (P = 0.044 and P = 0.032), MSC treatment averted these rises, showing a distinct trajectory, particularly for IL1RA. MSC treatment also reduced IL6 levels compared to CTRL-mCTRL, while both groups showed similar reductions in Long pentraxin. Furthermore, MSC infusions prevented the neurofilament light chain surge observed in CTRL patients. CONCLUSIONS: MSC in COVID-19 patients resulted safe and feasible, effectively modulating inflammatory cytokines, in particular mitigating brain damage related biomarker, suggesting both reduced inflammation and a potential neurological protection.