Causal Estimation of Telehealth Modality Impact Using Simulated Patient Data and Non-Traditional Machine Learning

Abstract

Telehealth has emerged as an essential means of delivering healthcare, but the causal effect of various modalities. video, telephone, and chatbots is poorly understood because of confounding and limitations in observational data. Employing a simulated dataset of 1,000 teleconsultation visits, we compared these modalities on recovery outcomes at 14 days. The data contained demographic, clinical, and care process variables intended to approximate true-world patterns of modality use. We estimated Individual Treatment Effects (ITE) and Average Treatment Effects (ATE) adjusting for confounders through Causal Forests in a Double Machine Learning setting. Outcome showed that video calls had a very large improvement in recovery compared to phone calls (ATE = −0.285, 95% CI. [−0.470,−0.100]) while chatbot outcomes did not differ significantly from video calls. These results underscore the significance of modality selection in virtual care and illustrate the utility of causal inference strategies for the evaluation of digital health interventions when randomized controlled trials are impossible.

Ähnliche Arbeiten