Comprehensive Summary
Zhang et al. investigated the use of machine learning (ML) models to predict early treatment outcomes for patients with multidrug resistant or rifampicin resistant tuberculosis (MDR/RR-TB). The study included an internal cohort of 744 patients from Beijing and an external validation cohort of 137 patients from Guangzhou, with data collected between 2017 and 2023. Researchers compared traditional logistic regression with seven ML models to predict sputum culture conversion at two and six months of treatment. Results showed that the artificial neural network (ANN) model consistently outperformed logistic regression and other ML methods, with high accuracy, sensitivity, and specificity. The ANN model was able to predict early treatment success with an area under the curve (AUC) of 0.82 at two months and 0.90 at six months, demonstrating strong stability and generalizability. Factors such as age, drug resistance type, medication adherence, comorbidities, and radiological findings were identified as key predictors of treatment response.
Outcomes and Implications
This research highlights the clinical potential of ML models, particularly ANN, in forecasting early treatment outcomes for MDR/RR-TB patients. Early prediction enables physicians to tailor therapies, strengthen monitoring for high-risk patients, and adjust regimens before treatment failure occurs. Given that MDR/RR-TB has a cure rate of only about 60% worldwide and prolonged therapy increases the risk of resistance, these predictive tools could play a pivotal role in improving patient outcomes and reducing transmission. Importantly, the ability to integrate routinely collected demographic and clinical data into predictive models makes this approach feasible in real-world healthcare settings, especially in resource-limited areas heavily burdened by TB. While further multicenter validation is necessary, these findings provide a foundation for adopting ML-based tools to support precision medicine and enhance global TB control efforts.