Cost-Sensitive Machine Learning Classification for Mass Tuberculosis Screening

Active screening for Tuberculosis (TB) is needed to optimize detection and treatment. However, current algorithms for verbal screening perform poorly, causing misclassification that leads to missed cases and unnecessary and costly laboratory tests for false positives. We investigated the role of machine learning to improve the predefined one-size-fits-all algorithm used for scoring the verbal screening questionnaire. We present a cost-sensitive machine learning classification for mass tuberculosis screening. We compared score-based classification defined by clinicians to machine learning classification such as SVM-RBF, logistic regression, and XGBoost. We restricted our analyses to data from adults, the population most affected by TB, and investigated the difference between untuned and unweighted classifiers to the cost-sensitive ones. Predictions were compared with the corresponding GeneXpert MTB/Rif results. After adjusting the weight of the positive class to 40 for XGBoost, we achieved 96.64% sensitivity and 35.06% specificity. As such, sensitivity of our identifier increased by 1.26% while specificity increased by 13.19% in absolute value compared to the traditional score-based method defined by our clinicians. Our approach further demonstrated that only 2000 data points were sufficient to enable the model to converge. Our results indicate that even with limited data we can actually devise a better method to identify TB suspects from verbal screening. This approach may be a stepping stone towards more effective TB case identification, especially in primary health centres, and foster better detection and control of TB.