Hemoptysis risk in hospitalized patients with bronchiectasis: development and external validation of a clinical prediction model

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background Hemoptysis is a complication of bronchiectasis, but risk prediction in hospitalized patients remains insufficiently characterized. Existing bronchiectasis severity scores were developed to predict mortality, hospitalization, or exacerbations rather than hemoptysis. We aimed to develop and externally validate a model for in-hospital hemoptysis risk and interpret its predictors within prespecified clinical domains. Methods This retrospective two-center study included 320 hospitalized adults with bronchiectasis. The derivation cohort comprised 215 patients from the Second Hospital of Hebei Medical University, and the external validation cohort comprised 105 patients from Shijiazhuang People's Hospital. Hemoptysis was the primary outcome. Candidate predictors were prespecified according to clinical relevance, clinical-domain representation, and consistency of definition and availability across centers. A core logistic regression model included age, sex, smoking history, asthma, serum albumin, hemoglobin, and log10(BNP/NT-proBNP+1); a prespecified secondary model additionally included chronic bronchitis. Discrimination, calibration, risk stratification, decision-curve analysis, and bootstrap validation were assessed. Results Hemoptysis occurred in 55 of 215 patients (25.6%) in the derivation cohort and 36 of 105 patients (34.3%) in the validation cohort. In the core model, higher serum albumin was independently associated with hemoptysis (odds ratio per 5-g/L increase, 1.95; 95% CI, 1.29–2.94), whereas asthma showed an inverse association (odds ratio, 0.17; 95% CI, 0.06–0.51). These associations were observational and should not be interpreted as causal or protective. The core model achieved an AUC of 0.747 in the derivation cohort and 0.820 in external validation, with a Brier score of 0.177. Bootstrap validation yielded an optimism-corrected AUC of 0.707. The bronchitis-extended model achieved AUCs of 0.763 and 0.746, respectively. Across low-, intermediate-, and high-risk tertiles of the core model, observed hemoptysis rates were 2.9%, 34.3%, and 65.7%. Calibration drift was evident in the low-risk group, where the mean predicted probability was 21.4% compared with an observed rate of 2.9%. Conclusion The core model showed good discrimination and risk stratification for in-hospital hemoptysis in bronchiectasis. The findings support a clinical-domain interpretation of bleeding susceptibility, while the albumin association should be regarded as observational. Because absolute risk was overestimated in the low-risk range, multicenter validation and local recalibration are required before probability-based clinical use.