This cross-sectional study examined whether patient portal message content and writing style explain disparities in care team response rates across racial, educational, and insurance groups, using NLP analysis of 3.6 million medical advice request messages sent to primary care clinicians from 2021–2023.
Messages from Black patients had a 3.7 percentage point lower target clinician response rate than White patients (11.6% relative reduction; 95% CI, −4.1 to −3.3 pp; P < .001). Writing style—not message content—accounted for ~48% of the Black-White gap, ~35% of the Hispanic-White gap, ~61% of the high school vs. college education gap, and ~43% of the Medicaid vs. commercially insured gap.
Single health system or network limits generalizability; cross-sectional design cannot establish causation; race/ethnicity and education inferred or self-reported with potential misclassification; unmeasured confounders (e.g., clinical complexity, panel size) may remain.
Writing style differences—not clinical content—drive roughly half of the response disparity for marginalized patients; care teams and health systems should audit message triage workflows and consider tools that reduce style-based bias in prioritizing portal messages.
