This document discusses how predictive risk models can help evaluate interventions in observational studies. It notes challenges with observational studies, such as regression to the mean and lack of control groups. It proposes several solutions to address these challenges, including before-after studies, regression adjustment, matching controls, and regression discontinuity designs. Matched controls are described as a way to reduce dependence on regression model specification by "data pre-processing" to compare intervention patients to similar control patients. The document concludes by surveying the current state of telehealth studies, finding most are descriptive or use before-after designs, with few employing more rigorous controlled designs.