Improving retrospective ARDS case-finding using a simple 72-h physiologic persistence rule

Area of Science:

• Critical Care Medicine
• Respiratory Medicine
• Health Informatics

Background:

• Retrospective studies often use single-time-point Berlin criteria (PaO2/FiO2 < 300 with PEEP ≥ 5) to identify acute respiratory distress syndrome (ARDS).
• Transient hypoxemia is common in critically ill patients and may lead to overestimation of ARDS prevalence.
• Current screening methods may yield cohorts inconsistent with clinical trial populations.

Purpose of the Study:

• To determine if a 72-hour persistence criterion for hypoxemia improves ARDS case identification accuracy.
• To evaluate the additional value of radiology keyword searches and ICD codes in ARDS case finding.
• To assess the impact of varying hypoxemia persistence durations on ARDS prevalence.

Main Methods:

• Retrospective cohort study using MIMIC-IV and a UK ICU dataset for derivation and validation.
• Identified patients meeting Berlin criteria for at least 72 hours.
• Expert adjudication of 2000 patients from MIMIC-IV based on clinical notes, imaging, and echocardiography.
• Sensitivity analyses performed for 24-h and 48-h persistence criteria.
• Compared diagnostic performance of radiology keywords and ICD codes against expert adjudication.

Main Results:

• Of 3940 patients meeting the 72-h persistence threshold, expert adjudication confirmed ARDS in 49.7% (MIMIC-IV) and 56% (UK cohort).
• ARDS prevalence decreased significantly with shorter persistence: 21% (48h), 8% (24h), and 6% (single measurement).
• Radiology keywords showed limited sensitivity (49%) and moderate specificity (76%), while ICD codes had higher sensitivity (76%) but low specificity (47%) within the 72-h enriched sample.

Conclusions:

• Single-time-point Berlin criteria are inadequate for retrospective ARDS identification.
• A ≥72-hour persistence rule enhances cohort enrichment but does not definitively diagnose ARDS, with residual misclassification.
• Persistence criteria should be considered a pragmatic enrichment strategy, not a definitive retrospective ARDS label.