Application of causal forest double machine learning (DML) approach to assess tuberculosis preventive therapy’s impact on ART adherence

Adherence to antiretroviral therapy (ART) is critical for HIV treatment success, yet the impact of tuberculosis preventive therapy (TPT) remains inadequately understood. Using observational data from 4152 HIV patients in Ethiopia (2005-2024), we applied causal inference methods, including Adjusted Logistic Regression, Propensity Score Matching, and Causal Forest Double Machine Learning (DML), to estimate TPT's effect on ART adherence. The DML approach (leveraging Random Forests and orthogonalization) provided the most precise estimates after model comparison. We found TPT initiation reduced adherence probability by 3.14 percentage points on average (ATE = - 0.0314; 95% CI - 0.0373, - 0.0254; p < 0.001). While most patients experienced negligible effects, substantial heterogeneity existed: individuals with advanced WHO stage, longer ART duration, higher BMI, or older age showed better adherence responses, whereas those with higher CD4 counts, functional impairment, or cotrimoxazole prophylaxis use faced greater risks. Subgroup analyses revealed consistent effects across clinical strata but greater variability among non-TPT initiators. These findings support personalized TPT deployment, prioritizing patients with advanced disease while monitoring vulnerable subgroups and highlighting the need for adherence support. Future research should validate results in multi-site cohorts using longitudinal and psychosocial data.