Development and validation of an interpretable machine learning model for risk stratification among individuals with latent tuberculosis infection: a multicenter case-control study
Abstract
Background: Latent tuberculosis infection (LTBI) affects a quarter of the global population. Conventional LTBI risk screening tools have limited sensitivity, and existing machine learning prediction models mostly rely on costly laboratory biomarkers, restricting their application in grassroots healthcare and public health. This study aimed to develop an interpretable machine learning model to distinguish pulmonary tuberculosis (PTB) from LTBI and stratify individuals with LTBI to prioritize further evaluation and preventive interventions.
Methods: This multicenter case-control study enrolled 782 participants from 21 designated tuberculosis institutions in Henan Province, China, including 401 individuals with LTBI and 381 patients with newly diagnosed, treatment-naive PTB. Epidemiological predictors were collected via standardized face-to-face questionnaires. The dataset was randomly split into a training set and a test set at a 7:3 ratio. Eleven predictors were screened via ten-fold cross-validated LASSO regression. Seven machine learning algorithms were constructed and evaluated using multiple metrics including AUC, F1 score, Brier score, calibration curves, and decision curve analysis. A three-tier risk framework was defined based on a percentile strategy. SHapley Additive exPlanations (SHAP) was applied to interpret the optimal model, which was further deployed as a free web calculator.
Results: The eXtreme Gradient Boosting (XGB) model achieved the highest test-set performance, with an AUC of 0.818 (95% CI 0.763–0.873), an F1 score of 0.764 (95% CI 0.698–0.822), and a Brier score of 0.176 (95% CI 0.148–0.204), with favorable calibration (Hosmer-Lemeshow P > 0.05). DCA confirmed a stable positive net benefit. Using cutoffs of > 0.763 (high-risk) and < 0.328 (low-risk), the model stratified participants into three tiers, with the proportion of PTB cases increasing from 20.00% in the low-risk tier to 84.29% in the high-risk tier. SHAP analysis identified male sex, low body mass index, advanced age, unintentional weight loss within six months, and low annual household income as the most influential features.
Conclusions: The interpretable XGB model based on low-cost epidemiological indicators can distinguish prevalent PTB from LTBI and may help prioritize individuals with LTBI for further evaluation. The web calculator is a research prototype; its clinical utility requires external validation and prospective evaluation before it can guide preventive-treatment decisions.
Related
- Development and internal validation of an interpretable mammography-based model for malignancy risk stratification in BI-RADS 4 breast lesions
- Development and Validation of an Interpretable Machine‐Learning Model for Predicting Suboptimal Clinical Response After Sleeve Gastrectomy: A Multicenter Study
- Feature-token clinical transformer for 24-h landmark mortality risk stratification in critically ill patients with heart failure: development and multicenter external validation
- Development and validation of a risk stratification model for mild cognitive impairment among older adults with possible sarcopenia integrating traditional Chinese medicine constitution: a multicenter cross-sectional study
- Preoperative risk stratification for postoperative inflammatory–nutritional deterioration after colorectal cancer surgery: an interpretable machine learning study
BibTeX
@article{wang2026development,
title = {{Development and validation of an interpretable machine learning model for risk stratification among individuals with latent tuberculosis infection: a multicenter case-control study}},
author = {Weidong Wang and Dingyong Sun and Yanqiu Zhang and Xiaowan Xu and Wenjing Chang and Danwei Zheng and Mengya He and Chunyu Xie and Yunfeng Qi and Yafei Xu and Jianhui Zhu and Linqi Diao},
journal = {BMC Public Health},
year = {2026},
month = oct,
doi = {10.1186/s12889-026-30066-9},
url = {https://link.springer.com/article/10.1186/s12889-026-30066-9},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Development and validation of an interpretable machine learning model for risk stratification among individuals with latent tuberculosis infection: a multicenter case-control study
AU - Weidong Wang
AU - Dingyong Sun
AU - Yanqiu Zhang
AU - Xiaowan Xu
AU - Wenjing Chang
AU - Danwei Zheng
AU - Mengya He
AU - Chunyu Xie
AU - Yunfeng Qi
AU - Yafei Xu
AU - Jianhui Zhu
AU - Linqi Diao
PY - 2026
DA - 2026/10/22
JO - BMC Public Health
DO - 10.1186/s12889-026-30066-9
UR - https://link.springer.com/article/10.1186/s12889-026-30066-9
AB - Background: Latent tuberculosis infection (LTBI) affects a quarter of the global population. Conventional LTBI risk screening tools have limited sensitivity, and existing machine learning prediction models mostly rely on costly laboratory biomarkers, restricting their application in grassroots healthcare and public health. This study aimed to develop an interpretable machine learning model to distinguish pulmonary tuberculosis (PTB) from LTBI and stratify individuals with LTBI to prioritize further evaluation and preventive interventions. Methods: This multicenter case-control study enrolled 782 participants from 21 designated tuberculosis institutions in Henan Province, China, including 401 individuals with LTBI and 381 patients with newly diagnosed, treatment-naive PTB. Epidemiological predictors were collected via standardized face-to-face questionnaires. The dataset was randomly split into a training set and a test set at a 7:3 ratio. Eleven predictors were screened via ten-fold cross-validated LASSO regression. Seven machine learning algorithms were constructed and evaluated using multiple metrics including AUC, F1 score, Brier score, calibration curves, and decision curve analysis. A three-tier risk framework was defined based on a percentile strategy. SHapley Additive exPlanations (SHAP) was applied to interpret the optimal model, which was further deployed as a free web calculator. Results: The eXtreme Gradient Boosting (XGB) model achieved the highest test-set performance, with an AUC of 0.818 (95% CI 0.763–0.873), an F1 score of 0.764 (95% CI 0.698–0.822), and a Brier score of 0.176 (95% CI 0.148–0.204), with favorable calibration (Hosmer-Lemeshow P > 0.05). DCA confirmed a stable positive net benefit. Using cutoffs of > 0.763 (high-risk) and < 0.328 (low-risk), the model stratified participants into three tiers, with the proportion of PTB cases increasing from 20.00% in the low-risk tier to 84.29% in the high-risk tier. SHAP analysis identified male sex, low body mass index, advanced age, unintentional weight loss within six months, and low annual household income as the most influential features. Conclusions: The interpretable XGB model based on low-cost epidemiological indicators can distinguish prevalent PTB from LTBI and may help prioritize individuals with LTBI for further evaluation. The web calculator is a research prototype; its clinical utility requires external validation and prospective evaluation before it can guide preventive-treatment decisions.
SN - 1471-2458
ER -