Portal triage tracks prose style, not clinical need

Clinical takeaway: A message's greeting, length, and punctuation tracked with whether or not it reached the intended clinician. Clinicians reviewing inbox workflows may want to consider whether their triage step is sorting on prose rather than acuity. 

Most portal messages addressed to you never land in a physician's inbox unaltered. In many primary care practices they enter a shared pool first, where a nurse or medical assistant decides in seconds whether the question needs you or can be closed out by the team. That decision is made with little clinical context and under the pressure of a high volume that has not eased since the pandemic. What the triager has to work with, mostly, is the way the patient wrote. 

Portal messaging is now a standing channel for medical advice rather than an occasional convenience, and the labor it generates falls on practices that were already at capacity. Patients from historically marginalized groups use portals less to begin with, and prior work has found that when they do write, they are less likely to get a timely reply and less likely to hear from a physician rather than a nurse. Why that happens has not been clear, and the potential explanations point to very different fixes. 

Across the full sample, 77.5% of threads got a reply from someone on the care team. Only 32.0% got one from the clinician the patient intended to reach. Nearly everyone heard back but far fewer heard back from their own clinician, which was a source of disparity. Black patients were 3.7 percentage points less likely than White patients to reach the clinician they addressed, an 11.6% relative drop. The gap in getting any reply at all was 1.1 points. The same pattern held for Hispanic patients, patients who preferred a language other than English, patients with a high school education, and Medicaid and Medicare-Medicaid dual-eligible patients. 

Content explained 5.7% of the gap for Black patients and 12.7% for patients with only a high school education, and for Hispanic patients it ran the other way, narrowing the gap rather than widening it. Writing style explained 48.0% of the Black-White gap, 34.9% for Hispanic patients, 60.5% for high school versus college education, and 42.8% for Medicaid versus commercial insurance. 

Style features sorted messages across the whole sample, not just by demographic group. Messages that opened by including the clinician's last name reached that person 38.7% of the time, compared with 25.7% for messages with no greeting. Length tracked the same way, 49.7% for messages of 201 words or more versus 21.0% for messages of four to 25 words. Question marks, exclamation marks, and positive sentiment all tracked with escalation, while messages containing "please" did worse than those without. Reading level, notably, did not sort messages at all. 

Researchers analyzed 3,619,390 medical advice request threads sent by 511,020 adults to primary care clinicians in one large academic health system between 2021 and 2023. Regressions held constant patient comorbidity, prior care use, send time, and the specific clinician and practice each message went to. Natural language processing extracted content topics and style features from about 990,000 English-language threads, and a decomposition method then split each demographic gap between the two. Non-English messages were excluded despite being a likely contributor to response disparities, and the single-system design means triage workflows elsewhere may sort differently. 

Style-based sorting is easier to fix than bias rooted in race or insurance status. Triage staff almost certainly do not know they are doing it, so naming the pattern in an inbox huddle is a real first step. Portal design can do the rest, prompting patients toward structured messages that narrow the stylistic spread before a recipient sees them. Also, health systems are adding AI to draft replies and summarize inboxes, and those tools sit on top of human triage, so it will either learn this pattern or filter it out. 

"One of the most important writing-style features was the message's salutation," said Mitchell Tang, PhD, assistant professor of health policy and management at Columbia University Mailman School of Public Health. "Simply beginning a message with 'Dear Dr. Tang' seemed to make it much more likely that Dr. Tang would end up responding." 

"The results could help health systems identify potential sources of bias in portal-message triage and develop more equitable approaches to digital communication," Tang continued. "But they are also a reminder to think about how communication style can influence the way we perceive and respond to people—not only in health care, but in our everyday lives." 

Source: Tang M, et al. (2026 Aug 24) JAMA Netw Open. Care Team Response to Patient Portal Message Content and Writing Style