Early change in ventilatory ratio during prone positioning and 28-day mortality in patients with ARDS: a retrospective cohort study with exploratory eICU assessment

Prone positioning (PP) improves outcomes in moderate-to-severe acute respiratory distress syndrome (ARDS), but early bedside markers that reliably characterize physiological response remain uncertain. The ventilatory ratio (VR), calculated from minute ventilation, PaCO2, and predicted body weight, may capture ventilatory efficiency during PP. MIMIC-IV v3.1 was used as the primary retrospective cohort and eICU-CRD v2.0 as an exploratory cohort. Adult patients with moderate-to-severe ARDS receiving an initial PP session lasting at least 12 h were included. The primary exposure was exploratory VR response, defined as ∆VR% ≤ -20% using the nearest available post-PP VR within the predefined observation window. The minimum post-PP VR within 12 h was analyzed as a sensitivity definition. The primary MIMIC-IV cohort included 177 patients, with 73 deaths within 28 days. Only 6 patients were classified as VR responders under the primary nearest post-PP VR definition. Crude 28-day mortality was 33.3% among responders and 41.5% among non-responders; the unweighted Cox model did not show a statistically significant association (HR 0.79, 95% CI 0.20–3.08; P = 0.737). Adjusted and IPTW-weighted models were considered exploratory and not reliably estimable because the responder group contained only 6 patients and 2 deaths. Sensitivity analyses using the minimum post-PP VR definition produced a stronger but definition-sensitive signal, whereas analyses using absolute VR change or baseline/post-PP VR did not confirm a stable association. In eICU-CRD, 29 patients had complete core VR data and only 3 were responders. Early VR change during PP was not significantly associated with 28-day mortality under the primary nearest post-PP VR definition. This study supports the feasibility of extracting dynamic VR metrics from routinely collected ICU data, but does not establish VR response as a validated prognostic marker or clinical threshold.