Abstract
BACKGROUND: Optimal timing for treatment initiation in Mycobacterium avium complex pulmonary disease (MAC-PD) remains unclear due to lack of established rules for predicting treatment response. METHODS: A retrospective observational study was conducted to develop a prediction model for treatment failure at 2 Japanese university hospitals between 2012 and 2023. Participants were 135 patients with MAC-PD who received macrolides and ethambutol-containing regimens over 1 year. Treatment failure was defined as nonachievement culture conversion at 1 year. We selected model components as cavity (categorized by diameter) and bronchiectasis (categorized by modified Reiff score) on pretreatment computed tomography. Their combinations of each category were scored based on number of lobes involved and compared by average areas under the curve calculated using k-fold cross-validation. RESULTS: Forty-three (31.9%) of the 135 patients failed in treatment. Number of lobes with cavities > 2cm or bronchiectasis with varicose or cystic changes was designated as the prediction model, with an average area under the curve of 0.798, and was named the Bronchiectasis and Cavity Distribution score. The representative metrics were sensitivity of 0.907 at the cutoff of 2 and specificity of 0.913 at the cutoff of 4 points. The patients were stratified into low-risk (0-1 points), intermediate-risk (2-3 points), and high-risk (4-6 points) groups. The treatment failure rates were 8.0%, 35.6%, and 69.2% in the respective groups. CONCLUSIONS: With simple assessment of computed tomography findings, the Bronchiectasis and Cavity Distribution score predicted treatment failure. Although validation studies are warranted, this score may provide guidance for treatment of MAC-PD.