Heterogeneity in the Association Between Pneumococcal Vaccination and the Risk of Severe Community-Acquired Pneumonia in Elderly Inpatients: A Causal Forest Analysis

Background: Community-acquired pneumonia (CAP) is a leading cause of morbidity and mortality in the elderly. While pneumococcal vaccination is a core preventive measure, it remains unclear whether its association with severe CAP is uniform across all elderly subgroups. Our study aimed to evaluate the overall association of pneumococcal vaccination with the risk of severe CAP in hospitalized patients aged ≥ 65 years and to explore potential heterogeneity in this association using a causal forest model. Methods: We conducted a retrospective cohort study of patients discharged between January 2023 and June 2025, aged ≥ 65 years, with a primary diagnosis of CAP. We used multivariable logistic regression to estimate the average association and a causal forest model to explore heterogeneous patterns in the conditional average treatment effect (CATE). Results: Among 1906 included patients (severe CAP: 924; non-severe CAP: 982), PPSV23 vaccination was independently associated with reduced odds of all-cause severe CAP (adjusted OR = 0.610, 95% CI: 0.401–0.930). The causal forest model yielded an average treatment effect (ATE) estimate of −0.112 (95% CI: −0.200 to −0.023), corresponding to an 11.2 percentage-point reduction in absolute risk. Exploratory analysis suggested potential heterogeneity: the association appeared most pronounced in patients aged 65–74 years (CATE = −0.122) and showed an attenuating trend in older groups. Age was the primary variable associated with heterogeneity, followed by hypertension, SARS-CoV-2 infection, and sex. Conclusions: In this observational cohort study, PPSV23 vaccination was associated with a reduced risk of severe CAP in elderly inpatients under strong assumptions of no unmeasured confounding. Exploratory analyses suggested potential heterogeneity in this association, which appeared to attenuate with advancing age and may be influenced by comorbidities. These hypothesis-generating findings indicate that further investigation is needed to determine whether prevention strategies should be tailored for the very old and those with specific chronic conditions.