Development of a Predictive Model for Tuberculosis (TB) in a Small Town Like Kapurthala, Punjab Using AI

Abstract

Tuberculosis (TB) continues to be a big public health challenge in India, and it is important to have readily available strategies to recognize people who may have TB and refer them. The objective of this study is to propose a simple and easy-to-implement Artificial Intelligence model, which could help identify potential TB cases at a very early stage based on the routine collected demographic, clinical, behavioural, and environmental parameters in Kapurthala, Punjab. These were considered as potential predictors: age, sex, residence, BMI, cough, fever, night sweats, loss of weight, loss of appetite, fatigue, prior TB, exposure to TB in the home, diabetes, biomass smoke exposure, and lack of ventilation. Different algorithms such as logistic regression, decision tree, random forest, and XGBoost were tested. The illustrative dataset consisted of 40 synthetic participant records: 12 (30.0%) of which were confirmed with TB and 28 (70.0%) were not confirmed with TB. Cough was recorded in 22 (55.0%), fever in 19 (47.5%), weight loss in 17 (42.5%), and night sweats in 12 (30.0%) records. The variables that best discriminated between outcome groups were night sweats, loss of appetite, fatigue, loss of weight, fever, and cough. The accuracy, sensitivity, specificity, precision and F1-score of the logistic regression model that was selected were 0.925, 1.00, 0.893, 0.800, and 0.889, respectively, at a classification threshold of 0.50. It was illustrated with an ROC-AUC value of 1.00. Additionally, the performance of subgroups by age, sex, and place of residence was examined. While these results show the proposed analytical and reporting sequence, they do not prove clinical effectiveness due to the fact that they are synthetically generated. Prospective recruitment should be done, and independent validation/calibration with representative Kapurthala participants. A final model should facilitate screening for confirmatory testing rather than supplant clinical judgment and microbiological diagnosis.