Ventilatory Parameters and Adjunctive Therapies Associated with ICU Mortality in Severe Pneumonia: A Large Retrospective Cohort Study

Background: Severe pneumonia (SP) comprises clinically diverse infections with variable lung mechanics, host responses, and recruitability. Yet ventilatory settings and adjunctive therapies are frequently guided by ARDS-derived evidence. We aimed to examine the associations of serial ventilatory parameters, gas-exchange indices, and selected adjunctive therapies with ICU mortality among patients with SP receiving invasive mechanical ventilation. Methods: This single-center retrospective cohort study included 2480 adults with microbiologically confirmed SP who received invasive mechanical ventilation in the ICU from 2015 to 2025. Daily ventilatory, gas-exchange, vital-sign, laboratory, and management variables were collected during the early ventilation course. Multivariable time-dependent Cox regression models were stratified by ventilation mode and adjusted for demographics, non-pulmonary SOFA score, comorbidities, pathogen type, shock, gas exchange, inflammatory markers, organ dysfunction, and time. Exploratory subgroup analyses were performed according to viral or fungal infection, hypoxemia severity, and immunocompromised status. Results: ICU mortality was 38.23%. In primary imputed models, higher PEEP, systemic glucocorticoid use, and neuromuscular blocker use were each associated with higher ICU mortality (HR 1.04, 1.40–1.41, and 1.53–1.56, respectively; all p < 0.001). Complete-case sensitivity analyses attenuated the PEEP association, indicating sensitivity to missing-data handling. VT/kg and respiratory rate showed significant U/J-shaped non-linear associations with mortality (p = 0.005 and p = 0.027), with the lowest mortality risk at VT/kg 6–8 mL/kg and respiratory rate 18–26 breaths/min. Driving pressure, ventilatory ratio, minute ventilation, and [(4 × ΔP) + RR] were not significantly associated with mortality in linear models, although [(4 × ΔP) + RR] showed a non-linear association in sensitivity analyses. In severe hypoxemia, associations for PEEP, glucocorticoids, and neuromuscular blockers were not statistically significant. Conclusions: In this large SP cohort, ventilatory settings were generally within protective ranges. Mortality associations varied by parameter and subgroup, suggesting that physiology-guided ventilation and individualized PEEP titration may be reasonable approaches in light of the observed U-shaped (VT/kg) and J-shaped (respiratory rate) non-linear relationships, as well as a non-linear pattern for [(4 × ΔP) + RR], while adjunctive therapy signals require cautious interpretation given residual confounding.