Differences in how patients write portal messages may explain disparities in target clinician response, whereas differences in message content may contribute less to response.
Investigators used the Mass General Brigham Epic electronic health record system to analyze 3.6 million patient-initiated medical advice message threads from 511,020 adult patients sent from 2021 through 2023 to nontrainee physicians, nurse practitioners, or physician assistants at primary care practices. For a random sample of 1 million threads, the investigators characterized the message content and writing style with natural language processing. After 10,398 non-English messages were excluded, 989,602 threads remained for the text-based analysis.
The primary outcomes included whether a message thread received a response from any member of the care team or from the target clinician. Secondary outcomes included responses within 3 business days and expanded definitions of response that incorporated telephone calls and outpatient visits.
Following adjustment for patient clinical characteristics, message timing, practice, and target clinician, patients from historically marginalized groups had lower rates of care team response. Compared with White patients, Black patients had a 3.7–percentage point lower target clinician response rate, representing a 12% reduction relative to the 32% mean response rate. The difference in response from any care team member was smaller, at 1.1 percentage points.
Message content accounted for relatively little response differences across demographics. For target clinician response, message content accounted for 6% of the baseline gap among Black patients compared with White patients, 13% among patients with a high school education compared with those with a college education, and 10% among Medicaid beneficiaries compared with commercially insured patients. Among Hispanic patients, differences in content narrowed rather than widened the disparity.
Writing style had greater influence over response differences, accounting for 48% of the target clinician response gap among Black patients, 35% among Hispanic patients, 61% among patients with a high school education, and 43% among Medicaid beneficiaries.
Several writing characteristics were associated with higher target clinician response rates, including using the clinician’s name in the opening, positive sentiment, question or exclamation marks, and higher word counts. Target clinician response was higher among messages that used their last name in the opening (39%) and those that contained 4 to 25 (21%) or 101 to 200 (43%) words.
The contribution of individual writing characteristics varied across demographic groups. Message length and expressive punctuation were consistent contributors, with patients from historically marginalized groups more likely to send shorter messages and messages without expressive punctuation. Among Medicaid beneficiaries and patients with a high school education, message opening and sentiment contributed meaningfully to differences in target clinician response.
The study had several limitations. Data came from a single academic health system, potentially limiting generalizability to practices with different workflows, staffing models, and patient populations. The text analysis excluded non-English messages, and the evaluated features could not capture every aspect of message content or writing style. Care team responses outside the electronic health record also may not have been completely captured. The cross-sectional design did not permit causal conclusions, and the investigators could not determine whether writing characteristics drove prioritization or reflected unmeasured confounding factors.
“[W]riting style was a potential explanation for approximately half of the observed disparities in care team response,” wrote lead study author Mitchell Tang, PhD, of the Harvard Kenneth C. Griffin Graduate School of Arts and Sciences at Harvard University, and colleagues.
Full disclosures of the study authors can be found in the study.
Source: JAMA Network Open
