Predicting treatment adherence in patients with drug-resistant tuberculosis: insights from socioeconomic, demographic, and clinical factors of patients in the rural Eastern Cape

Abstract

Background: Drug-resistant tuberculosis (DR-TB) is a formidable challenge to global health. Patients are compelled to adhere to intricate medication regimens over extended periods, and any failure to comply with these treatment protocols can lead to treatment failure, increased mortality rates, and a heightened risk of developing further drug resistance. This study identifies the key factors that influence treatment adherence among patients with DR-TB. Furthermore, it rigorously evaluates the predictive accuracy of machine learning models in assessing treatment adherence, with a strong focus on socioeconomic, demographic, and clinical factors. Methods: A retrospective analysis was conducted on patients with DR-TB in rural Eastern Cape. Data were collected from medical records. Four different models were developed and tested to evaluate their effectiveness in predicting treatment adherence: Random Forest, Logistic regression, Support Vector Machine (SVM), and Gradient Boosting. Results: The Random Forest model achieved an accuracy of 53.3% in predicting treatment adherence. An analysis of feature importance indicated that age, income, education, social history, patient category, and comorbidities were the most significant factors influencing adherence. Patients with higher incomes, higher levels of education, and fewer comorbidities were more likely to follow their treatment plans. Conclusion: Socioeconomic and clinical factors, such as income, education level, and the presence of comorbidities, significantly influence adherence to DR-TB treatment. These findings indicate that machine-learning models, particularly Random Forest algorithms, can effectively assist in clinical decision-making by identifying patients who may be at risk of not adhering to their treatment.