Association Between Billing Patient Portal Messages as e-Visits and Patient Messaging Volume

Following the onset of the COVID-19 pandemic, the Centers for Medicare & Medicaid Services expanded billing options for telemedicine including “e-visits,” defined as asynchronous patient portal messages that require medical decision-making and at least 5 minutes of clinician time over a 7-day period. During the pandemic, patient messaging increased by more than 50%, requiring a substantial amount of uncompensated clinician time.^1,2 With the option to bill for a subset of messages, health care organizations began implementing workflows to bill qualifying patient messages as e-visits.³ However, level of clinician adoption and association with patient messaging behavior is unknown. We analyzed the implementation of clinician-initiated e-visit billing at a large academic medical center.

At UCSF Health, initially patients self-selected into an e-visit on sending a message. After minimal uptake, this process was retired in favor of clinicians determining when patient-initiated message threads qualify for billing as an e-visit. Patients are informed via the portal before sending a message that it may result in a bill, with information about cost sharing.⁴ Simultaneously, on November 14, 2021, an upgrade changed the patient portal user interface to encourage within-thread replies.

To capture clinician adoption of e-visit billing, we measured total patient medical advice message threads (each representing a back-and-forth exchange) and the subset billed as e-visits each week, and calculated the percentage of threads billed.^2,5 Our sample covered November 1, 2020, through October 22, 2022, with November 1, 2020, as the start of patient-initiated e-visits and November 14, 2021, as the switchover to clinician-initiated e-visit billing. Next, we conducted interrupted time-series analyses with Newey-West standard errors⁶ for 5 week-level measures: (1) billed e-visits to assess clinician adoption, (2) patient message threads to assess patient response in terms of initiating new threads, (3) patient messages (both initial messages and responses) to assess patient response in terms of overall messaging volume, and (4-5) scheduled ambulatory visits and unscheduled telephone calls to assess a substitution effect. Analyses were conducted using Stata v17, with statistical significance set at 2-sided α of .05. Our study was deemed exempt from needing patient consent by the UCSF institutional review board.

Prior to implementation of clinician-initiated e-visit billing, UCSF Health billed a mean of 50.6 (SD, 15.8) e-visits per week (0.26% of patient message threads) and received a mean of 19 739 (SD, 2003.5) patient threads and 59 648 (SD, 5486.9) patient messages per week. Following implementation, mean weekly e-visits increased to 235.7 (SD, 31.3) (1.40% of patient message threads), while mean patient message threads decreased to 16 838 (SD, 1491.6) and patient messages decreased to 57 925 (SD, 5580.4) per week (Figure). In interrupted time-series models, we found an immediate level change increase in weekly e-visits (β = 195.7 additional e-visits per week [95% CI, 171.7 to 219.6]; P < .001) and decreases in weekly patient message threads (β = −3388.2 message threads per week [95% CI, −4781.7 to −1994.7]; P < .001) and patient messages (β = −4594.2 messages per week [95% CI, −9025.3 to −163.1]; P = .04) (Table). No trend changes were observed. We found no significant level or trend changes in scheduled visits or unscheduled telephone calls.

This study found an association between implementation of clinician-initiated billing and an increase in e-visits, but adoption was low. Multiple factors likely contributed, including lack of awareness of e-visit workflow, perception that additional steps to complete e-visits were not worth the reimbursement, and concerns about negative patient perceptions of charging for messages.

Nonetheless, an association with a reduction in patient portal messaging (both threads and individual messages) was observed that may be attributable to awareness of the possibility of being billed. The reduction in threads may be partially due to the simultaneous change in the portal interface, but this should not have affected patient message volume. Study limitations include lack of assessment of e-visit profitability; inability to assess what proportion of patient-initiated message threads required medical decision-making and at least 5 minutes of time to meet e-visit billing criteria; clinician satisfaction with e-visit billing; and whether any reduction in messaging resulted in worse patient outcomes. Future research should investigate overall costs under different payment models and the effect of billing for messaging on outcomes, health equity, and patient and clinician satisfaction.

Accepted for Publication: December 21, 2022.

Published Online: January 6, 2023. doi:10.1001/jama.2022.24710

Corresponding Author: A. Jay Holmgren, PhD, MHI, University of California, San Francisco, 10 Koret Way, Office 327A, San Francisco, CA 94131 (a.holmgren@ucsf.edu).

Author Contributions: Dr Holmgren had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: All authors.

Acquisition, analysis, or interpretation of data: Holmgren, Byron, Grouse.

Drafting of the manuscript: Holmgren.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Holmgren.

Administrative, technical, or material support: Holmgren, Byron, Grouse.

Supervision: Adler-Milstein.

Conflict of Interest Disclosures: Dr Holmgren reported grants from the American Medical Association and the Healthcare Leadership Council outside the submitted work. Dr Byron reported receiving royalties from University of California, copyright SF2015-137 and holding patent SF2016-239. Dr Grouse reported receiving personal fees from Siemens for research in natural language processing and from American Academy of Neurology Continuum Audio for work as an interviewer, outside the submitted work. No other disclosures were reported.

Data Sharing Statement: See the Supplement.