Trajectories of the PaO2/FiO2-to-PaCO2 ratio and mortality in mechanically ventilated patients with sepsis: a multicenter retrospective cohort study

Background This study examines the relationship between changes in the PaO 2 /FiO 2 -to-PaCO 2 ratio (PFP) over the first 48 h and prognosis in mechanically ventilated sepsis patients, building on previous findings that PFP is prognostic in this population. Methods This retrospective cohort study analyzed adults with sepsis on invasive mechanical ventilation using the MIMIC-IV and eICU-CRD databases. PFP parameters were assessed at ICU admission, 0–24 h, and 24–48 h, employing a group-based trajectory model to characterize changes. Kaplan–Meier and Cox proportional hazards regression analyzed the relationship between PFP trajectories and post-landmark 28- and 90-day mortality. External validation utilized data from the Affiliated Hospital of Nanjing University of Chinese Medicine. ROC curves compared discrimination for mortality among PFP trajectories, static PFP, PF trajectories and PaCO 2 trajectories, and against standard critical-care prognostic scores. Nested Cox models with IDI and NRI were applied to evaluate incremental prognostic value of PFP trajectories over APS-III, SAPS-II and Charlson. Results This study analyzed 1,880 patients from the MIMIC-IV database, identifying three PFP trajectory patterns: a decline from high values (23.2%), a moderate increase (43.5%), and an increase from low values (33.3%). Cox regression and Kaplan–Meier analyses revealed that the moderate increase group had the lowest 28-day and 90-day mortality. Subgroup analyses indicated that hypertension and diabetes influenced the relationship between trajectory classification and mortality risk. External validation with 593 patients from the eICU-CRD database and clinical validation with 296 patients confirmed the moderate-rise trajectory consistently exhibited the lowest mortality risk, while prognostic rankings for the two high-risk trajectory subgroups showed notable cross-cohort heterogeneity. A multivariate group-based trajectory model (GBMTM) was developed to characterize P/F and ventilatory function dynamics, identifying a tripartite progression structure. The PFP trajectory model produced a slightly higher AUC for mortality prediction than static PFP(0.561 vs. 0.543), with a small difference detected by the DeLong test, however, it does not provide additional predictive value. Conclusion Three PFP trajectories were identified in ICU sepsis patients on mechanical ventilation. A trajectory with stabilization at moderate PFP levels was linked to lower mortality risk and no enhanced predictive value for mortality.