External Validation of the BACES Score in Canadian Patients With Nontuberculous Mycobacterial Pulmonary Disease.

BACKGROUND:The clinical course of nontuberculous mycobacterial pulmonary disease (NTM-PD) can be variable and difficult to predict. Recently, the BACES score was developed as a tool to predict all-cause mortality in patients with NTM-PD. This score is calculated based on five patient characteristics (BMI, age, cavity, erythrocyte sedimentation rate, and sex), and higher scores portend worse prognosis. Although the BACES score has been validated in a cohort of South Korean patients, it has not yet been validated in other settings or ethnic groups. RESEARCH QUESTION:How well does the BACES mortality score perform in a cohort of Canadian patients with NTM-PD? STUDY DESIGN AND METHODS:We performed a single-center retrospective chart review. Patients who were seen between July 2003 and June 2021 were eligible for inclusion if they met guideline-based diagnostic criteria for NTM-PD and were excluded if any component of the BACES score was missing. To assess the model's discriminatory performance, we compared Kaplan-Meier curves between risk groups and calculated Harrell's C index. To assess calibration, we used a graphical calibration curve. RESULTS:The cohort included 435 patients with a median follow-up of 5.8 years. The median age was 64 years and 74% were female. Based on the BACES scores, patients were classified into three risk groups: low, moderate, or high. Survival curves showed clear separation of the risk groups. Harrell's C index was 0.733 in the study cohort, indicating moderate to good discriminatory performance, although this was lower than the value reported in the derivation cohort (0.812). The graphical calibration curve showed a tendency of the BACES model to underpredict mortality. INTERPRETATION:The BACES model was evaluated in a multicultural cohort of Canadian patients and demonstrated good discriminatory performance but suboptimal calibration, which may be due to population differences, the use of dichotomized variables in model construction, or both.