Association between plasma transfusion and in-hospital mortality in severe pneumonia: a propensity score-matched retrospective cohort study

BACKGROUND: Plasma transfusion is used in patients with severe pneumonia who develop coagulopathy or bleeding, but recipients are often sicker than non-recipients. We evaluated the association between plasma transfusion and in-hospital mortality after adjustment for available measured confounders, while recognizing that unmeasured confounding from transfusion indications and illness severity may persist. METHODS: This single-center retrospective cohort included 156 adults with severe pneumonia admitted between January 2021 and May 2024; 62 received plasma and 94 did not. The matched mortality comparison using propensity score matching (PSM) and McNemar analysis was prespecified as the primary analysis. Supportive and exploratory analyses included Cox regression, Kaplan-Meier analysis, IPTW, overlap weighting, Fine-Gray modeling, E-value analysis, and exploratory coagulation-enhanced, logistic, microbiological/coagulation, and time-dependent exposure sensitivity analyses. RESULTS: Before matching, mortality was higher among plasma recipients than controls (64.5% vs. 28.0%; crude odds ratio [OR] 4.69, 95% confidence interval [CI] 2.35-9.34; p < 0.001), with substantial baseline imbalance. After matching, mortality remained numerically higher in the plasma group (50.0% vs. 36.1%) but was not statistically significant (matched OR 1.62, 95% CI 0.67-3.92; McNemar p = 0.383). Cox models yielded attenuated HRs after adjustment, but the fully adjusted model violated the global proportional hazards assumption; therefore, Cox estimates were interpreted as time-weighted associations over follow-up rather than constant effects. Exploratory sensitivity estimates varied in direction across estimands: coagulation-enhanced matching yielded OR 1.50 (95% CI 0.53-4.21), full-cohort logistic regression yielded adjusted OR 3.19 (95% CI 1.12-9.11), approximate adjusted time-dependent Cox analysis yielded HR 1.73 (95% CI 0.85-3.51), and strict verified-date adjusted time-dependent Cox analysis yielded HR 2.67 (95% CI 1.38-5.16). CONCLUSIONS: The attenuation of the crude mortality difference after propensity score matching is consistent with baseline severity and confounding by indication accounting for a substantial component of the unadjusted association; however, residual confounding cannot be excluded. Because residual confounding, immortal-time bias, related time-dependent exposure bias, and proportional-hazards violations remain important limitations, these data do not establish a definitive beneficial or harmful effect of plasma transfusion. The non-significant matched result reflects limited statistical power (approximately 25% for a paired OR of 2.0) and should not be interpreted as evidence of no effect.