Daily Ards Research Analysis
A Transformer-based EHR model (TECO) showed superior performance for ICU mortality prediction, including external validation in ARDS-related cohorts. Preterm neonatal RDS severity correlated with higher circulating MIF and GDF-15 and with risk alleles in MIF and GDF-15 genes. A bedside case demonstrated that EIT-guided individualized PEEP during VV-ECMO improved V/Q matching in asymmetric lung disease with contralateral pulmonary embolism.
Summary
A Transformer-based EHR model (TECO) showed superior performance for ICU mortality prediction, including external validation in ARDS-related cohorts. Preterm neonatal RDS severity correlated with higher circulating MIF and GDF-15 and with risk alleles in MIF and GDF-15 genes. A bedside case demonstrated that EIT-guided individualized PEEP during VV-ECMO improved V/Q matching in asymmetric lung disease with contralateral pulmonary embolism.
Research Themes
- AI-driven prognostication in critical care
- Biomarkers and genetics in neonatal respiratory distress
- Bedside physiological imaging to individualize ventilation
Selected Articles
1. A deep learning model for clinical outcome prediction using longitudinal inpatient electronic health records.
TECO, a Transformer-based model trained on 2,579 hospitalized COVID-19 patients, consistently outperformed EDI, RF, and XGBoost for ICU mortality prediction and generalized to external ARDS-related MIMIC datasets. It also surfaced clinically interpretable, outcome-correlated features, suggesting utility as an early warning system across inpatient conditions.
Impact: Provides an externally validated, interpretable deep learning approach that outperforms common baselines for ICU mortality, a key endpoint in ARDS and critical care.
Clinical Implications: Hospitals could deploy TECO-like systems to flag high-risk ICU patients earlier, potentially guiding staffing, triage, and escalation of supportive therapies in ARDS and related conditions.
Key Findings
- In development (COVID-19 cohort, n=2,579), TECO achieved AUC 0.89–0.97, surpassing EDI (0.86–0.95), RF (0.87–0.96), and XGBoost (0.88–0.96).
- In two external MIMIC test datasets, TECO yielded AUC 0.65–0.77, higher than RF (0.59–0.75) and XGBoost (0.59–0.74).
- The model identified clinically interpretable features correlated with mortality risk, supporting transparency and potential bedside adoption.
Methodological Strengths
- External validation across multiple ICU datasets including ARDS-related cohorts
- Transformer architecture leveraging longitudinal EHR with interpretability
Limitations
- Proprietary EDI comparator unavailable in MIMIC, limiting head-to-head comparison
- Retrospective EHR design; prospective impact and generalizability require further validation
Future Directions: Prospective, multi-site deployment studies to assess clinical impact, calibration drift monitoring, fairness auditing, and integration with clinician workflows.
OBJECTIVES: 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: The TECO model 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 acute respiratory distress syndrome cohort ( 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 2 MIMIC testing datasets (EDI not available), TECO yielded higher AUC (0.65-0.77) than RF (0.59-0.75) and XGBoost (0.59-0.74). In addition, TECO was able to identify clinically interpretable features that were correlated with the outcome. DISCUSSION: The TECO model outperformed proprietary metrics and conventional machine learning models in predicting ICU mortality among patients with COVID-19, widespread inflammation, respiratory illness, and other organ failures. CONCLUSION: The TECO model 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.
2. Association of macrophage migration-inhibitory factor gene and growth differentiation factor 15 gene polymorphisms and their circulating levels with respiratory distress syndrome among preterm neonates.
In 90 preterm neonates, severe RDS showed markedly higher serum MIF and GDF-15 than mild/moderate RDS, and risk alleles in MIF rs755622 and GDF-15 rs4808793 were more frequent in cases than controls. Findings support inflammatory and maturation-related pathways as contributors to neonatal RDS risk and severity.
Impact: Links circulating biomarkers and genetic polymorphisms with neonatal RDS severity, offering targets for risk stratification and potential therapeutic pathways.
Clinical Implications: Serum MIF and GDF-15 and genotyping of MIF rs755622 and GDF-15 rs4808793 may inform early risk stratification and monitoring strategies in preterm neonates at risk for RDS.
Key Findings
- Severe RDS had higher median serum MIF (17.32 μg/L) and GDF-15 (3.19 pg/mL) than mild–moderate RDS (5.50 μg/L and 0.71 pg/mL; both P<0.05).
- The mutant C allele of MIF rs755622 was more frequent in cases (37.5%) vs controls (13.3%) (P=0.001; OR 0.256; 95% CI 0.112–0.589).
- The mutant G allele of GDF-15 rs4808793 was more frequent in cases (49.2%) vs controls (30%) (OR 0.443; 95% CI 0.229–0.856).
Methodological Strengths
- Biomarker quantification via ELISA with genotype analysis by RFLP-PCR
- Severity stratification (mild–moderate vs severe) within cases
Limitations
- Single-center, small sample size limits generalizability
- Case-control design precludes causal inference; potential population stratification
Future Directions: Larger, multi-ethnic cohorts with longitudinal follow-up to validate predictive value and explore mechanistic links to lung development and inflammation.
BACKGROUND: In preterm newborns, neonatal respiratory distress syndrome (RDS) is among the main causes of respiratory failure and mortality. However, the effect of macrophage migration-inhibitory factor (MIF) on neonatal developmental lung disease is not well documented in the literature. Moreover, little is known about the effects of growth differentiation factor-15 (GDF-15) on lung maturity in preterm infants. PURPOSE: To evaluate serum MIF and GDF-15 levels in preterm infants with and without RDS and analyze the genetic profile of single nucleotide polymorphisms (SNPs) for MIF rs755622 G>C and GDF-15 rs4808793 C>G. METHODS: In this case-control study, 90 preterm newborns were categorized into 3 groups: group A included 30 preterm newborns with mild to moderate RDS, group B included 30 preterm newborns with severe RDS, and group C included 30 healthy preterm newborns. Enzyme-linked immunosorbent assay methods were used to measure serum MIF and GDF-15 levels. The MIF rs755622 G>C and GDF-15 rs4808793 C>G SNPs were analyzed by restriction fragment length polymorphism-polymerase chain reaction. RESULTS: Significantly higher median MIF and GDF-15 blood levels were noted among neonates with severe RDS (17.32 μg/L and 3.19 pg/mL, respectively) versus those with mild to moderate RDS (5.50 μg/L and 0.71 pg/mL, respectively) (P<0.05 for both). A significantly higher frequency of a mutant C-allele of MIF rs755622 G>C was noted among cases (37.5%) versus controls (13.3%) (P=0.001; odds ratio [OR], 0.256; 95% confidence interval [CI], 0.112-0.589). A significantly higher frequency of a mutant G-allele of GDF-15 rs4808793 C>G SNPs was noted among cases (49.2%) versus controls (30%) (OR, 0.443; 95% CI, 0.229-0.856). CONCLUSION: These findings suggest that serum MIF and GDF-15 levels are strongly associated with RDS severity among preterm neonates. Moreover, polymorphisms of MIF and GDF-15 could be genetic risk factors for the development of neonatal RDS among preterm babies.
3. Personalized ventilation guided by electrical impedance tomography with increased PEEP improves ventilation-perfusion matching in asymmetrical airway closure and contralateral pulmonary embolism during veno-venous extracorporeal membrane oxygenation: A case report.
Bedside EIT revealed profound regional V/Q mismatch in a VV-ECMO patient with unilateral pneumonia and contralateral PE. Titrating PEEP above the injured lung’s airway opening pressure (to 20 cmH₂O) improved recruitment, stabilized EELI, enhanced V/Q matching, and improved oxygenation without hemodynamic compromise.
Impact: Demonstrates real-time V/Q EIT to individualize PEEP in complex physiology, highlighting a practical pathway to precision ventilation during ECMO.
Clinical Implications: EIT can guide PEEP titration to exceed airway opening pressure in recruitable lung regions and help anticipate malperfusion (e.g., PE), supporting individualized ventilation strategies in severe hypoxemia and ECMO.
Key Findings
- At PEEP 12 cmH₂O, EIT showed left-lung–dominant ventilation and right-lung–dominant perfusion, prompting PE suspicion later confirmed by contrast CT.
- The injured right lung had an airway opening pressure of 16 cmH₂O; increasing PEEP to 20 cmH₂O improved recruitment, stabilized EELI, and enhanced V/Q matching.
- Higher PEEP improved oxygenation with increased cardiac output and reduced pulmonary vascular resistance, without hemodynamic instability.
Methodological Strengths
- Real-time bedside EIT for simultaneous regional ventilation and perfusion assessment
- Physiology-based intervention (PEEP titration above airway opening pressure) with objective EELI tracking
Limitations
- Single case report limits generalizability
- No long-term outcomes or comparative control conditions
Future Directions: Prospective studies testing EIT-guided PEEP strategies versus standard care in ARDS and ECMO populations, including safety and outcome endpoints.
We report the case of a 54-year-old man with right-lung pneumonia and contralateral pulmonary embolism (PE) conditioning severe refractory hypoxemia requiring veno-venous extracorporeal membrane oxygenation. Electrical impedance tomography (EIT) was used to assess recruitability and regional ventilation and perfusion. At a clinical positive-end expiratory pressure (PEEP) of 12 cmH₂O, EIT revealed predominant ventilation in the left lung and predominant perfusion in the right lung. Reduced perfusion in the left lung raised suspicion of PE, confirmed by contrast-enhanced computed tomography. The clinical PEEP was insufficient to maintain recruitment of the pneumonia-affected right lung, which showed an airway opening pressure (AOP) of 16 cmH₂O. Therefore, PEEP was increased to 20 cmH₂O to exceed the AOP of the injured lung, improving lung recruitment, stabilizing end expiratory lung impedance (EELI), and increasing V/Q matching. Oxygenation improved, following an increased cardiac output, and reduced pulmonary vascular resistance. Despite increasing ventilation pressures, the higher PEEP was well-tolerated hemodynamically, optimizing V/Q coupling in this case of unilateral shunt and contralateral dead space. This case highlights the utility of ventilation/perfusion EIT in optimizing ventilatory strategies, in anticipating the presence of pulmonary malperfusion at bedside, and demonstrating the importance of individualized, physiology-based interventions in complex critical care scenarios.