Enhancing Prediction of Treatment Duration in New Tuberculosis Cases: A Comprehensive Approach with Ensemble Methods and Medication Adherence

Tuberculosis (TB) remains a significant global health problem, with treatment duration varying among patients. TB patients have difficulty following a long-term treatment regimen. After the final diagnosis is determined, it is necessary to know the predicted duration of treatment for a patient. By increasing patient compliance with taking medication, the percentage of TB patients will increase, and this can reduce cases of multi-drug resistant patients and dropouts. This study aims to build a prediction model for the duration of treatment for new cases of Pulmonary TB patients by adding medication compliance parameters using the ensemble method. The research methodology uses CRISP-DM. This study begins with identifying problems and objectives, collecting data, preprocessing and analyzing data, modeling, evaluating, and validating models. The results showed that adding medication compliance parameters can improve model performance. However, the results of model exploration with feature selection techniques and various ensemble methods have not shown good performance. The medication adherence parameters used in this study are the number of medications swallowed in Phase I and Anti-Tuberculosis drug compliance in Phase I. These parameters had never been used in previous studies. The prediction model can be used as an early warning for a patient. If a patient is predicted to have a treatment duration of more than six months, then the patient will receive stricter drug intake supervision. Thus, this proposed model is expected to help achieve the target of eliminating Tuberculosis in 2030 to reduce the death rate by 90% compared to 2019.