Patient-facing tech

Kelly Gleason, Thomas Kidu +3

How can health systems test large language models on real patient portal messages without touching live EHR workflows? This technical feasibility tutorial describes a Python 3 web interface and modular backend running inside the institutional firewall on an NVIDIA GRID T4-1Q GPU, supporting single-message and batch tasks: authorship identification, categorization, criticality flagging, and response drafting with zero-, one-, and few-shot prompting. A deidentification pipeline validated against 110 manually adjudicated entities achieved 95.1% sensitivity and 82.1% precision. Use cases drew on an IRB-approved dementia-relevant corpus of 6941 medical advice request messages from 497 patients; token-based cost readouts were included. No comparative performance effect sizes are reported.

Khushi Kohli, Stephanie Wang +6

Do US counties with limited physical healthcare capacity also lack the broadband needed for telemedicine to substitute? This cross-sectional ecological analysis linked 3,133 counties across the 2017 National Neighborhood Data Archive (outpatient care centers, diagnostic labs, nursing/residential care), 2022 FCC Mapping Broadband Health in America data (split at the median 9.8% of households without broadband), and 2022 American Community Survey covariates, using t-tests and multivariable linear regression. Low-broadband counties had fewer outpatient care centers (10.46 vs. 11.91 per 100,000) and diagnostic labs (1.91 vs. 3.95 per 100,000; both P<0.001), plus higher poverty and rurality. Adjusted associations persisted (β = -0.045, -0.024, and -0.089).

Caglar Onal, Sezen O Onal

Is engagement with clinical digital health tools associated with psychological distress? This cross-sectional analysis pooled Health Information National Trends Survey cycles (HINTS 5, 2017-2020; HINTS 6, 2022; HINTS 7, 2024) covering 23 682 US adults (mean age 55 years; 59% female), with a composite engagement index spanning secure messaging, online test results, portal access, wellness apps, and device data transmission, and distress measured by the PHQ-4. In survey-weighted regression, higher engagement was associated with higher distress (\u03b2 = 0.51; 95% CI, 0.27-0.75; P < .001), against a mean PHQ-4 of 2.0. Associations were strongest for active behaviors—clinician messaging and app use—and persisted among those reporting good or better health.

Anita A Vashi, Liam Rose +3

Does centralizing appointment scheduling and adding same-day virtual clinician evaluation improve access after nurse triage? This retrospective quasi-experimental evaluation used difference-in-differences and event-study analyses of the VA Health Connect rollout across 18 regions from October 2018 to September 2024, drawing on 11,118,916 encounters (4,560,677 pre-, 6,558,239 post-modernization) from VA Corporate Data Warehouse, Telecare, CRM, and VSignals survey data. Same-day scheduling rose 14.3 percentage points (95% CI 10.1-18.5) and time from call to scheduled appointment fell 0.37 days, though time to completed appointment rose 2.9 days. Callers with no 7-day follow-up declined 2.3 points; ED visits, admissions, and costs were unchanged.

Liu, J. B., Chen, Y.-J. +4

What drives nonresponse to routinely collected patient-reported outcome measures? This retrospective cohort study used iterative mixed-effects logistic regression on all adults seen at five Mass General Brigham radiation oncology clinics over one year (12,214 patients, 71 providers, five clinics), modeling failure to ever complete the portal-administered PROMIS Global-10. Patient- and appointment-level response rates were 35.4% and 10.9%, with patient-level response varying nearly fivefold across clinics (12.8% to 66.2%). After adding provider- and clinic-level factors, sex, education, and employment became nonsignificant, while recent surgery (aOR 1.97) and time since diagnosis >12 months (aOR 0.46) persisted; later program launch (aOR 0.29) and higher historical collection rate (aOR 0.79) predicted lower nonresponse, and academic versus community setting did not.

Elizabeth Woodcock, Chris Profeta, Molly Siegel

What determines whether automated waitlists—tools that notify patients of earlier appointment openings—succeed in improving ambulatory access? A convergent, multisite mixed methods study surveyed 127 US health systems, 90 of which reported automated waitlist usage data, plus qualitative and quantitative data from 10 purposively sampled systems, analyzed using the Consolidated Framework for Implementation Research. High performers filled 38.8% (IQR 36.2%-45.7%) of appointments offered through the waitlist, and missed appointment rates were lower for waitlist-scheduled visits (3.1%, IQR 2.5%-4.8%) than for all appointments (6.6%, IQR 4.1%-9.9%). Flexible configuration, cross-functional governance, and leadership endorsement facilitated sustained use; specialty gatekeeping, clinician capacity, insurance requirements, and digital inequities limited reach.