Patient portal messages are often presented as a direct line between patients and their doctors. In practice, however, a message may pass through several layers of a health care system before reaching the clinician a patient intended to contact. A new study from Columbia University Mailman School of Public Health suggests that the way patients write those messages can influence whether they receive a response—and whether that response comes from their own doctor or another member of the care team. The findings, published in JAMA Network Open, raise concerns that ordinary differences in communication style may be reinforcing racial, ethnic, educational and insurance-related inequities in digital health care.
The researchers examined 3,619,390 patient medical advice request messages sent between 2021 and 2023 by 511,020 adults. These messages were analyzed as part of a cross-sectional study, meaning the investigators looked for associations within a large body of existing records rather than following patients over time in a controlled experiment. The analysis compared responses received by different patient groups and examined two separate outcomes: whether anyone on the care team responded and whether the clinician originally intended to receive the message responded. The distinction matters because a nurse, medical assistant or covering clinician may answer a message even when the patient expected a personal response from a doctor.
Across the dataset, patients from groups that have historically experienced disadvantages in health care were generally less likely to receive responses. This included non-White patients, people who preferred a language other than English, patients with lower levels of formal education and people insured through Medicaid or through both Medicaid and Medicare. Compared with White patients, Black patients had a response rate from the originally intended clinician that was 3.7 percentage points lower. Although a percentage-point difference may appear modest in a single interaction, its implications become substantial when multiplied across millions of digital conversations, particularly for patients who depend heavily on portal messaging to obtain advice, renew medications or determine whether an in-person visit is necessary.
The study was motivated in part by the transformation of patient communication during the COVID-19 pandemic. Portal messaging volumes rose sharply as clinics limited face-to-face visits, and usage has remained high even after many services returned to normal. Yet patients may not realize that a message addressed to a physician usually does not arrive directly in that doctor’s personal inbox. Instead, many health systems use triage systems designed to manage the volume. Nurses, medical assistants and other staff members rapidly review incoming requests, assess their urgency and complexity, decide whether they can respond themselves and determine whether a physician should become involved.
That process creates a setting in which linguistic signals can affect clinical workflow. In the study, researchers used natural language processing, a branch of artificial intelligence that allows computers to quantify patterns in human language, to evaluate both what patients wrote and how they wrote it. Content-related measures included the apparent subject or medical issue raised in a message. Writing-style measures included message length, formality, sentiment, linguistic structure and the way the message began. Statistical models were then used to estimate how much these features could account for differences in response patterns between demographic and insurance groups.
The results suggested that message content alone explained relatively little of the disparities. Writing style, by contrast, accounted for a substantial share of the differences in responses from the intended clinician. Style features explained 48 percent of the response gap between Black and White patients, 35 percent of the gap between Hispanic and White patients, and 60.5 percent of the difference between patients whose highest education was high school and those with a college degree. Writing style also accounted for 43 percent of the difference between Medicaid beneficiaries and commercially insured patients. These figures do not mean that writing style caused the disparities, but they indicate that language patterns were strongly associated with how messages moved through the triage process.
One of the most influential features was the salutation—the opening greeting used in a message. According to Mitchell Tang, an assistant professor of Health Policy and Management at Columbia Mailman School of Public Health and the study’s lead investigator, beginning a message with a formal greeting such as “Dear Dr. Tang” appeared to make it more likely that the named physician would ultimately respond. A formal salutation may function as a social cue, signaling that the message is specifically directed to a doctor rather than presenting a general administrative request. It may also make the intended recipient easier for a busy triage worker or an automated routing system to identify.
The finding illustrates how seemingly neutral judgments can produce unequal outcomes. A message that is concise, formally structured and addressed to a physician may be interpreted as organized, urgent or worthy of escalation. A message written in an informal tone, containing emotional language, limited punctuation or a less conventional opening may be perceived as more complicated or less clearly directed—even when the underlying medical need is identical. Such differences can reflect education, language background, cultural norms, disability, stress or unequal familiarity with digital health systems. Patients are therefore not necessarily choosing a style strategically; they may simply be communicating in the way that feels natural or accessible to them.
The investigators emphasize that the findings should not be interpreted as evidence that patients are responsible for unequal responses or that everyone should be trained to write in a particular way. The study was observational and based on messages from a single health care environment, so it cannot establish that a specific writing feature directly caused a clinician to respond. Other factors may also influence triage, including staffing levels, clinician workload, message timing, prior relationships between patients and doctors, language translation and the urgency of the medical issue. Natural language processing systems can introduce their own limitations as well, especially when they interpret multilingual communication, culturally specific expressions or emotional language.
Instead, the research points toward possible changes in the design and oversight of digital communication systems. Health organizations could audit response rates by race, ethnicity, preferred language, education and insurance status, while also examining whether particular linguistic features affect routing decisions. Triage protocols might be developed to focus more heavily on clinical content and less on assumptions associated with tone, formality or grammar. Technology could help identify the intended recipient without rewarding a specific style of greeting, and staff training could encourage reviewers to recognize that communication conventions vary widely across communities. As portal messaging becomes a permanent part of medical care, the study suggests that equity will depend not only on who has digital access, but also on how health systems interpret the words patients use once they are online.
Subject of Research: Patient portal message responses, triage processes and communication-style-related disparities in health care.
Article Title: Care Team Response to Patient Portal Message Content and Writing Style
Web References: Columbia University Mailman School of Public Health; DOI: https://doi.org/10.1001/jamanetworkopen.2026.30379
References: JAMA Network Open, “Care Team Response to Patient Portal Message Content and Writing Style”; related research by Mitchell Tang and colleagues on racial, ethnic, insurance and language-related disparities in patient portal message triage.
Keywords: patient portals, digital health, health disparities, medical communication, natural language processing, message triage, racial inequities, health care access, clinician response, public health
