Julia Adler-Milstein

Background: The Age‐Friendly Health Systems (AFHS) movement, guided by the 4Ms framework has been widely adopted to improve care for older adults. While specialized inpatient models such as Acute Care for Elders (ACE) units and Geriatrics Consults aim to improve outcomes, gaps remain in understanding how these models align with 4Ms adherence and their relationship to patient outcomes. Methods: We conducted a retrospective observational study of 28,833 hospital encounters for patients aged ≥ 65 at an academic health system. Encounters were categorized into four groups: Usual Care, Geriatrics Consult, ACE Unit, and ACE Unit and Consult. Adherence to each M and a composite 4Ms score was derived from 16 process measures. Outcomes included falls, length of stay, discharge to a facility, and 30‐day readmissions. Multivariable regression was completed to assess the relationship between 4Ms adherence and health outcomes. Results: Patients receiving any Geriatrics clinical model had higher overall 4Ms adherence than usual care. High 4Ms adherence was associated with lower readmissions in the Consult Only group but higher readmissions in the ACE + Consult group. Facility discharge rates increased with high adherence in Consult and ACE + Consult groups. No group differences were observed in falls. ACE Unit patients with high adherence showed shorter lengths of stay, whereas Consult groups had longer stays. Conclusion: Inpatient Geriatrics models improve adherence to Age‐Friendly care practices, but their effects on outcomes are mixed.

Background: The 4Ms (what matters, medication, mentation, and mobility) is a widely adopted age‐friendly care framework. However, there is no evidence on inpatient 4Ms adherence thresholds associated with improved outcomes. Methods: In a retrospective cross‐sectional analysis, we measured encounter‐level adherence to the 4Ms for inpatients 65+ at the University of California San Francisco between January 1, 2021 and December 31, 2024. Outcomes included 30‐day unplanned readmissions, 30‐day emergency department (ED) visits, length of stay (LOS), and community admissions discharged to a facility. Using fixed effects regression models, we estimated the effect on outcomes of adherence to 100% of 4Ms assessments (what matters, medication, mentation, and mobility screening) plus 25%, 50%, and 75% of shifts with full adherence to 4Ms Act Ons (actions, such as de‐prescribing high‐risk medications, ambulating). We conducted subgroup analyses by age and medical versus surgical admissions. Results: Our study sample included 34,923 inpatient encounters for 22,949 unique patients aged 65+ (mean age 75.3; 49.0% female). We observed lower 30‐day readmissions at Act On adherence of 50% (4.1% reduction; 95% CI: 1.8–6.4) and 25% (1.9%; 0.3–3.6). We also found lower 30 day ED visits at 50% Act On adherence in the overall group (3.3%; 0.9–5.8), and in those 85+ (5.0%; 0.4–9.7). Act On adherence and LOS demonstrated a dose–response‐like relationship, with 1.0 (0.8–1.1), 1.5 (1.3–1.7), and 1.9 (1.6–2.2) day reductions associated with 25%, 50%, and 75% Act On adherence respectively. No associations were observed between Act On adherence and discharge to facility. Conclusions: Higher inpatient adherence to 4Ms care is associated with meaningful reductions in post‐acute care utilization and LOS. Our analyses offer clinical leaders guidance on deploying limited resources to positively influence outcomes. They suggest that fully adherent assessments and 25%–50% of shifts with fully adherent Act Ons represent target thresholds.

Diagnostic errors are a substantial source of patient harm. As artificial intelligence (AI) integrates into clinical workflows, opportunities are emerging to assess their impacts on diagnostic excellence (DxEx). The Coordinating Center for Diagnostic Excellence (CODEX) at the University of California San Francisco established the Action Incubator to translate research advances in DxEx into tangible strategies for improving diagnosis. The September 2025 in-person inaugural Action Incubator convened 30 multidisciplinary stakeholders representing health systems, patient advocacy, industry, and policy groups. Through structured discussions and breakout sessions, participants identified AI scribes as a near-term, scalable use case for evaluating AI’s impact on diagnosis not only because of their widespread adoption, but – as supported by cognitive load theory – because of their potential to reduce cognitive burden and allow clinicians to focus more on diagnosis. A modified Delphi process was used to prioritize candidate measures based on feasibility, and impact. Participants generated 17 candidate measures of AI scribe impact on DxEx. Consensus was reached on two priority metrics as functions of AI scribe usage rates by primary care physicians: (1) Timely follow-up of abnormal test results related to breast and colorectal cancer screening and (2) Patient-reported diagnostic experience. Participating health systems will pilot these measures using electronic health record audit logs and patient surveys. The first CODEX Action Incubator developed a pragmatic, consensus-driven framework for measuring the impact of AI on DxEx. Future annual Action Incubators will take up timely, actionable topics related to DxEx.

There is currently much interest in screening patients for social drivers of health (SDOHs) to address health-related social needs (HRSNs), prompting clinicians to increasingly document HRSNs in the electronic health record (EHR). SDOH data in the EHR create many opportunities to address patients’ needs, tailor medical care to account for social circumstances, exchange these data across organizations, and enhance population health improvement efforts. This review synthesizes current efforts to document SDOH data in structured and unstructured methods, to make use of community-level SDOH data, and to develop national standards driving EHR SDOH documentation and exchange. We describe barriers faced by organizations related to data collection burden, patient privacy, data quality, and health equity, and we discuss future research priorities to make the collection and use of EHR SDOH data more accurate, actionable, and equitable.

What We Wanted to Find Out: We wanted to learn what patients at a city hospital for people with lower incomes think about artificial intelligence (AI) in their healthcare. We looked at what they knew about AI, if they trusted it, and what good or bad things they thought it might bring. How We Did It: We asked 313 patients between November 2024 and January 2025 to fill out a survey. We asked them about their knowledge of AI, if they would trust AI to help doctors make decisions, and if they wanted things to be clear and supervised. We used numbers to see if knowing more about AI made them think it was more helpful. We also read their written comments to understand what built their trust. What We Found: Most patients (84%) knew about AI from everyday things like phones or smart speakers. But less than half knew about AI being used in healthcare to help doctors. Patients who knew more about AI also thought it was more helpful. Most patients said they needed AI to be clear (92%), doctors to check it (82%), and for it to be proven safe and effective to trust it. What This Means: This study is among the first to examine how patients at a safety-net hospital feel about AI in healthcare. Patients generally like the idea of AI helping doctors, but they really want it to be clear how AI is used and for doctors to still be in charge. Our Main Point: To make sure AI in healthcare is fair and trustworthy for everyone, it’s very important to teach patients about it and involve them in how it’s used.

To evaluate the association between perceived and actual changes in physician documentation time (DocTime) following implementation of an artificial intelligence (AI) scribe and to determine whether physicians with higher baseline DocTime experience greater reductions in DocTime from AI scribe use. Retrospective assessment of AI scribe use among 310 ambulatory physicians across specialties who chose to adopt a commercial tool at a large academic medical center. We utilized data from a postimplementation user feedback survey and electronic health record audit log measures of scribe use and DocTime. We used an ordered logit model to assess adjusted associations between perceived and actual changes in DocTime in the 12 weeks after AI scribe adoption for the 252 physicians (81.3%) with survey data. Multivariate regression models assessed whether baseline DocTime modified the relationship between level of AI scribe use (percentage of weekly encounters) and DocTime. Although the majority of physicians perceived reductions in DocTime (86.5%) following AI scribe adoption, there was no overall association between perceived reductions and actual changes in DocTime (OR, 0.975; P = .144). In multivariate models, higher levels of AI scribe use were associated with lower DocTime. For each additional 10% of encounters with AI scribe use, DocTime decreased by just over 30 seconds per scheduled hour (P < .001). This effect was modified by baseline DocTime, with less-efficient physicians realizing the majority of time savings. Although most physicians perceived DocTime reductions from AI scribe use, those realizing the majority of actual time savings were those with higher relative baseline DocTime.

AI scribes are transforming clinical documentation at scale. However, their real-world performance remains understudied, especially regarding their impacts on patient safety. To this end, we initiate a mixed-methods study of patient safety issues raised in feedback submitted by AI scribe users (healthcare providers) in a large U.S. hospital system. Both quantitative and qualitative analysis suggest that AI scribes may induce various patient safety risks due to errors in transcription, most significantly regarding medication and treatment; however, further study is needed to contextualize the absolute degree of risk.