Machine Learning for Prediction of Relapses in Multiple Drug Resistant Tuberculosis Patients

Abstract

The objective of the study: to evaluate the possibility of using machine learning algorithms for prediction of relapses in multiple drug resistant tuberculosis (MDR TB) patients.Subjects and Methods. Сlinical, epidemiological, gender, sex, social, biomedical parameters and chemotherapy parameters were analyzed in 346 cured MDR TB patients. The tools of the scikit-learn library, Version 0.24.2 in the Google Colaboratory interactive cloud environment were used to build forecasting models.Results. Analysis of the characteristics of relapse prediction models in cured MDR TB patients using machine learning algorithms including decision tree, random forest, gradient boosting, and logistic regression using K-block stratified validation revealed high sensitivity (0.74 ± 0.167; 0.91 ± 0.17; 0.91 ± 0.14; 0.91 ± 0.16, respectively) and specificity (0.97 ± 0.03; 0.98 ± 0.02; 0.98 ± 0.02; 0.98 ± 0.02, respectively).Five main predictors of relapse in cured MDR-TB patients were identified: repeated courses of chemotherapy; length of history of tuberculosis; destructive process in the lungs; total duration of treatment less than 22 months; and use of less than five effective anti-TB drugs in the regimen of chemotherapy.