Heterogenous impacts of substance use disorder telehealth on health outcomes: A causal machine learning approach

Several studies have investigated the adoption and effectiveness of telehealth services, especially after the increase in telehealth usage after the COVID-19 pandemic. Far fewer studies, however, have investigated its adoption and effectiveness for substance use disorder (SUD) – which is a pressing public health concern in the U.S. Compared with other forms of medical treatment, substance use treatment often requires frequent or extended in-person interaction (e.g., through intensive outpatient services, residential care, or daily methadone administration). We therefore investigate whether SUD-related public health outcomes, such as overdose hospitalizations and deaths, can be significantly reduced by telehealth adoption. In contrast with past studies on SUD telehealth, which often rely on small patient samples, we leverage a large, nationally representative dataset of SUD treatment centers and service offerings. We investigate whether rates of telehealth adoption at the county level predict lower incidence rates of substance-related overdose deaths and hospitalizations. Importantly, building on research that uses ML to understand SUD treatment (Baird, Cheng, & Xia, 2023; Baird, Cheng, & Xia 2022), we also leverage causal machine learning (ML) analysis to investigate whether the impact of telehealth adoption on SUD public health outcomes varies based on community characteristics. Our preliminary results suggest that increased telehealth adoption at the county level predicts lower substance-related hospitalization rates. Notably, the impact of telehealth adoption is strongest in areas with high levels of access to in-person treatment. This suggests that telehealth may complement, rather than substitute for, traditional in-person treatment in the context of SUD, in contrast with prior research suggesting a substitutive effect (Cao, Chen, & Smith, 2023). Our planned causal ML analysis aims to uncover and explore further moderators, such as communities’ social determinants of health, substance use patterns, and access to other forms of healthcare.