Patient Portals in Primary Care: Impacts on Patient Health and Physician Productivity*

Interest in innovative health care delivery models has increased due to measures such as the Affordable Care Act, which is designed to expand insurance coverage and contain health care costs. The goal of these innovations is to increase physician productivity without sacrificing quality of care. One innovation that has been forwarded as a low-cost alternative to physician office visits is “e-visits,” or secure messaging between patients and physicians via patient portals. We evaluate the effect of e-visit adoption on patient health and physician productivity using a six-year panel dataset from a large primary care provider in the United States. The main challenge in evaluating e-visits is that there could be unobservable selection: in particular, a näıve analysis regressing the number of office visits (our measure of physician productivity) on patient e-visit adoption is biased downward if patients adopt e-visits because they are already on a trajectory toward improved health. We address this selection problem by implementing an instrumental variable analysis that leverages the differential propensities of physicians to adopt e-visits over time, since this variation affects patient e-visit adoption arguably independent of their health conditions. We find large selection effects: our instrumental variable estimates show that e-visit adoption doubles the number of annual office visits from 3 to 6, whereas a näıve analysis predicts an effect of the opposite sign. Meanwhile, we find no evidence that e-visits affect patient health as measured by emergency room visit frequency, blood cholesterol (LDL), or blood glucose (HbA1c) levels. We supplement our analysis with an event study design, and also conduct robustness checks using data on telephone visits.