Pulmonary cavitation as a dominant prognostic indicator in tuberculosis complicated by chronic pulmonary aspergillosis: a retrospective cohort study.

BACKGROUND

Pulmonary tuberculosis (PTB) complicated by chronic pulmonary aspergillosis (CPA) is associated with substantial morbidity and mortality; however, its prognostic determinants remain incompletely characterized. This study aimed to identify factors associated with prognosis and unfavorable outcomes in patients with PTB complicated by CPA.

METHODS

Between January 2023 and June 2024, consecutive adult patients diagnosed with PTB complicated by CPA were retrospectively enrolled. Unfavorable outcomes were defined as massive hemoptysis requiring surgical intervention or death during follow-up. Univariable and multivariable Cox proportional hazards models were used to identify independent prognostic factors. SHapley Additive exPlanations (SHAP) analysis was applied to quantify the relative contribution of each predictor. Kaplan-Meier analysis was performed to compare event-free survival across risk strata.

RESULTS

Among 440 patients (median follow-up, 6.0 months), 88 (20.0%) experienced unfavorable outcomes. Multivariable analysis identified older age [hazard ratio (HR) =1.042 per year], diabetes mellitus (HR =1.888), pretreatment hemoptysis (HR =1.652), pulmonary cavitation (HR =2.069), and irregular medication adherence (HR =2.599) as independent predictors of poor prognosis. SHAP analysis ranked pulmonary cavitation as the strongest contributor to unfavorable outcomes. Kaplan-Meier curves demonstrated significantly lower event-free survival among patients with these risk factors (all P<0.05).

CONCLUSIONS

In patients with PTB complicated by CPA, unfavorable outcomes were independently associated with older age, diabetes mellitus, pretreatment hemoptysis, pulmonary cavitation, and irregular medication adherence, with pulmonary cavitation representing the dominant prognostic indicator. These findings support risk-stratified management strategies and highlight the value of interpretable machine learning for prognostic assessment.