Daily Ards Research Analysis
Three studies advance ARDS science across measurement, monitoring, and prediction: a multi-database cohort shows SpO2/FiO2 can misclassify ARDS severity versus PaO2/FiO2; a prospective pediatric feasibility study demonstrates bedside EIT can individualize PEEP; and a Transformer-based EHR model (TECO) outperforms conventional methods for ICU mortality prediction with external validation in ARDS and sepsis cohorts.
Summary
Three studies advance ARDS science across measurement, monitoring, and prediction: a multi-database cohort shows SpO2/FiO2 can misclassify ARDS severity versus PaO2/FiO2; a prospective pediatric feasibility study demonstrates bedside EIT can individualize PEEP; and a Transformer-based EHR model (TECO) outperforms conventional methods for ICU mortality prediction with external validation in ARDS and sepsis cohorts.
Research Themes
- Oxygenation metrics and ARDS severity classification
- Bedside monitoring and individualized ventilation (EIT-guided PEEP)
- AI-driven prognostication from longitudinal EHR data
Selected Articles
1. Limitations of SpO2/FiO2 for ARDS classification
Across three ICU databases totaling 708 ARDS patients, SpO2/FiO2 frequently misclassified ARDS severity compared with reference measures, indicating systematic limitations of pulse oximetry-based oxygenation indices. Findings support cautious use of SpO2/FiO2 for severity stratification and trial enrollment.
Impact: This multi-database analysis challenges routine reliance on SpO2/FiO2 for ARDS severity assessment and could influence guidelines, monitoring strategies, and clinical trial design.
Clinical Implications: Where feasible, prioritize PaO2/FiO2 from arterial blood gas for ARDS severity classification and use SpO2/FiO2 with caution, accounting for oximetry bias and patient-specific factors (skin pigmentation, perfusion, vasopressors). Re-evaluate trial eligibility criteria and ventilatory targets that rely solely on SpO2.
Key Findings
- In 708 ARDS patients from three high-resolution ICU databases, SpO2/FiO2 frequently misclassified ARDS severity compared with reference standards.
- Time-matched SpO2 data showed systematic limitations that can bias severity stratification.
- Findings question the use of SpO2/FiO2 as a surrogate for PaO2/FiO2 in ARDS classification.
Methodological Strengths
- Multi-database cohort including ICU Cockpit, MIMIC-IV, and SICdb
- Time-matched analysis of oxygenation metrics across datasets
Limitations
- Retrospective observational design limits causal inference
- Potential database coding/measurement variability and unmeasured confounders
Future Directions: Prospective studies comparing SpO2/FiO2 and PaO2/FiO2 under standardized calibration and diverse patient phenotypes; development of corrected SpO2-based indices integrating perfusion and sensor bias.
BACKGROUND: The ratio of pulse-oximetric peripheral oxygen saturation to fraction of inspired oxygen (SpO METHODS: Observational cohort study of ARDS patients from three high-resolution Intensive Care Unit databases, including our own database ICU Cockpit, MIMIC-IV (Version 3.0) and SICdb (Version 1.0.6). Patients with ARDS were identified based on the Berlin criteria or ICD 9/10-codes. Time-matched datapoints of SpO RESULTS: Overall, 708 ARDS patients were included in the analysis. ARDS severity was misclassified by SpO CONCLUSIONS: The use of SpO
2. A deep learning model for clinical outcome prediction using longitudinal inpatient electronic health records.
The TECO Transformer model trained on 2,579 COVID-19 inpatients achieved AUC 0.89–0.97 for ICU mortality prediction and outperformed EDI, RF, and XGBoost. External validation in ARDS (n=2,799) and sepsis (n=6,622) cohorts from MIMIC-IV showed superior performance (AUC 0.65–0.76) and identified clinically interpretable predictors.
Impact: Demonstrates generalizable AI leveraging longitudinal EHR to predict ICU mortality, with external validation in ARDS and sepsis; could reshape early warning and resource allocation.
Clinical Implications: If prospectively validated and integrated, TECO could augment early warning systems for ARDS and sepsis, enabling earlier escalation, targeted monitoring, and improved triage. Implementation requires model governance, local recalibration, and bias auditing.
Key Findings
- TECO achieved AUC 0.89–0.97 for ICU mortality prediction in development COVID-19 cohort and outperformed EDI, RF, and XGBoost.
- External validation in ARDS (n=2,799) and sepsis (n=6,622) cohorts showed higher AUC (0.65–0.76) than RF and XGBoost.
- The model identified clinically interpretable predictors correlated with mortality.
Methodological Strengths
- Large development cohort with two external validations (ARDS and sepsis)
- Transformer architecture leveraging longitudinal, time-dependent EHR variables
Limitations
- Preprint not yet peer-reviewed; potential overfitting despite external validation
- Generalizability to non-MIMIC settings and prospective performance remain uncertain
Future Directions: Prospective, multi-center impact evaluation; fairness and drift monitoring; integration with clinician-in-the-loop workflows; head-to-head comparison with calibrated risk scores in ARDS.
OBJECTIVE: Recent advances in deep learning show significant potential in analyzing continuous monitoring electronic health records (EHR) data for clinical outcome prediction. We aim to develop a Transformer-based, Encounter-level Clinical Outcome (TECO) model to predict mortality in the intensive care unit (ICU) using inpatient EHR data. MATERIALS AND METHODS: TECO was developed using multiple baseline and time-dependent clinical variables from 2579 hospitalized COVID-19 patients to predict ICU mortality, and was validated externally in an ARDS cohort (n=2799) and a sepsis cohort (n=6622) from the Medical Information Mart for Intensive Care (MIMIC)-IV. Model performance was evaluated based on area under the receiver operating characteristic (AUC) and compared with Epic Deterioration Index (EDI), random forest (RF), and extreme gradient boosting (XGBoost). RESULTS: In the COVID-19 development dataset, TECO achieved higher AUC (0.89-0.97) across various time intervals compared to EDI (0.86-0.95), RF (0.87-0.96), and XGBoost (0.88-0.96). In the two MIMIC testing datasets (EDI not available), TECO yielded higher AUC (0.65-0.76) than RF (0.57-0.73) and XGBoost (0.57-0.73). In addition, TECO was able to identify clinically interpretable features that were correlated with the outcome. DISCUSSION: TECO outperformed proprietary metrics and conventional machine learning models in predicting ICU mortality among COVID-19 and non-COVID-19 patients. CONCLUSIONS: TECO demonstrates a strong capability for predicting ICU mortality using continuous monitoring data. While further validation is needed, TECO has the potential to serve as a powerful early warning tool across various diseases in inpatient settings.
3. EIT guided evaluation of regional ventilation distributions in neonatal and pediatric ARDS: a prospective feasibility study.
In 26 neonatal/pediatric ARDS or PLD patients (40 EIT measurements), EIT-guided PEEP was feasible and tended to be lower than both clinician-set and ARDSnet-recommended PEEP. EIT-guided individualized PEEP may optimize regional ventilation and potentially reduce VILI, including during ECMO.
Impact: Provides prospective evidence supporting bedside EIT to individualize PEEP in neonatal/pediatric ARDS, a population lacking robust monitoring tools.
Clinical Implications: Consider EIT-guided PEEP titration in pediatric ARDS to balance overdistension and collapse at the bedside, potentially lowering PEEP compared with ARDSnet tables and clinician defaults, even during ECMO.
Key Findings
- EIT-guided PEEP determination was feasible and safe in 26 neonatal/pediatric patients (40 measurements), including during ECMO.
- Median EIT-derived PEEP (11 mbar) was lower than clinician-set PEEP (11.5 mbar, p<0.001) and ARDSnet-recommended PEEP (14 mbar, p=0.018).
- In nARDS/PLD, EIT-PEEP was 3 mbar below clinician-set and 11 mbar below ARDSnet recommendations.
Methodological Strengths
- Prospective design with predefined decremental PEEP trials and continuous EIT monitoring
- Trial registration (retrospective) and quantitative comparison with ARDSnet and clinician-set PEEP
Limitations
- Single-center feasibility study with small sample size
- Retrospective trial registration and no clinical outcome endpoints powered for efficacy
Future Directions: Multi-center randomized or adaptive trials testing EIT-guided PEEP against standard care in pediatric ARDS with VILI and outcome endpoints.
BACKGROUND: Despite international guidelines for lung protective ventilation in neonatal or pediatric acute respiratory distress syndrome (nARDS/ pARDS), prospective data on bedside monitoring tools for regional ventilation distribution and lung mechanics are still rare. As a bedside and radiation-free procedure, electrical impedance tomography (EIT) offers a practical and safe approach for analyzing regional ventilation distributions. Recent trials in adults have shown the efficacy of an individualized EIT guided strategy for the improvement of ventilator induced lung injury (VILI). METHODS: We performed a single-center prospective feasibility study from November/2021 to December/2023 in the department of neonatal and pediatric intensive care medicine at the University Children´s Hospital in Bonn. All patients with diagnosis of nARDS (or history of perinatal lung disease-PLD)/ pARDS were screened for study inclusion. In all patients a decremental PEEP (positive end-expiratory pressure) trial was performed with a continuous EIT monitoring for an individual analysis of the EIT guided pixel compliance (C RESULTS: Overall, 40 EIT measurements were performed in 26 neonatal and pediatric patients (nARDS/PLD, n = 6; and pARDS, n = 20) within a predefined decremental PEEP trial. Thirteen patients were classified as having severe nARDS (PLD)/ pARDS with an Oxygen Saturation Index (OSI) > 12 or Oxygenation Index (OI) > 16. In-hospital mortality rate was 27% in the overall cohort. The median EIT-PEEP (11mbar) was calculated as lowest, as compared to the clinically set PEEP (11.5mbar, p < 0.001), and the ARDSnetwork PEEP table recommendation (ARDSnet-PEEP, 14mbar, p = 0.018). In patients with nARDS/PLD, the EIT-PEEP was calculated 3mbar below the clinically set PEEP (p = 0.058) and 11 mbar below the ARDSnet-PEEP (p = 0.01). In the linear regression analysis, EIT-PEEP and the dynamic compliance (C CONCLUSION: EIT is feasible and can be performed safely in patients with diagnosis of nARDS/PLD and pARDS, even during ongoing extracorporeal membrane oxygenation (ECMO) support. An individualized PEEP finding strategy according to the EIT compliance might optimize regional ventilation distribution in these patients and can potentially decrease VILI. CLINICAL TRIAL REGISTRATION: The study was registered at the German Clinical Trials Register (GCT; trial number: DRKS 00034905, Registration Date 15.08.2024). The registration was performed retrospectively after inclusion of the last patient.