Faculty Bibliography

BACKGROUND:Medication non-adherence contributes to the efficacy-effectiveness gap in real-world practice. Glucosuria, routinely measured on urinalysis, may serve as an objective proxy for assessing medication adherence to sodium-glucose cotransporter 2 (SGLT2) inhibitors, a class of medications that reduce the risks of kidney disease, heart failure, and mortality. METHODS:We leveraged two cohorts from the Optum Labs Data Warehouse real-world data (2014-2023): Cohort 1 included 3,987 patients with SGLT2 inhibitor pharmacy claims and a urinalysis performed both before and during active fill periods; cohort 2 included 45,711 patients prescribed SGLT2 inhibitors with urinalysis performed within 6 months after initiation. Glucosuria was defined as urine glucose 2+ or greater. Cohort 2 was followed for a mean of 3.3 years for all-cause mortality, heart failure hospitalization, end-stage kidney disease; fractures were used as a negative control outcome. RESULTS:In cohort 1, SGLT2 inhibitor use (vs. no use) was associated with 18.1-fold higher odds of glucosuria (95% CI, 16.4-20.0), adjusted for diabetes status. In cohort 2, 26,657 (58%) had glucosuria within 6 months of SGLT2 inhibitor initiation, suggesting adherence. After inverse probability of treatment weighting, glucosuria was associated with lower risks of all-cause mortality (HR, 0.78; 95% CI, 0.73-0.83), heart failure hospitalization (HR, 0.87; 95% CI, 0.78-0.98), and end-stage kidney disease (HR, 0.73; 95% CI, 0.58-0.91) compared to no glucosuria. No association was observed with fractures (HR, 1.00; 95% CI, 0.95-1.06). CONCLUSIONS:Glucosuria was associated with SGLT2 inhibitor adherence and, among patients prescribed SGLT2 inhibitors, is associated with reductions in risks of all-cause mortality, heart failure, and end-stage kidney disease.

OBJECTIVE/UNASSIGNED:To test whether AI-generated, clinician-edited after-visit summaries (AVS) are more patient-friendly than clinician-generated AVS, without compromising safety, when implemented in a live inpatient setting. PATIENTS AND METHODS/UNASSIGNED:grade reading levels, and presence of a "simplified" description of both the diagnosis and treatment/interventions. Our composite outcome was positive if all 3 of these outcomes were positive. We also surveyed patients, nurses, and clinicians on perceptions of safety, empathy, and understandability. RESULTS/UNASSIGNED:grade than the control group (8.5 vs 10.6, P <.001). The AI-generated, clinician-edited AVS were also more likely to contain simplified explanations of diagnoses (96.7% vs 12.9%, P <.001) and hospital interventions (86.7% vs 12.9%, P <.001). Our composite outcome favored AI-generated, clinician-edited summaries (13.3% vs 6.5%, p = 0.637). CONCLUSION/UNASSIGNED:In this small, randomized trial, exploratory and component measures suggest improved patient-friendliness in AI-generated, clinician-edited AVS, although no significant difference was observed in the primary outcome. CLINICALTRIALSGOV NUMBER/UNASSIGNED:NCT06711458.

BACKGROUND:Accurate measurement of pharmacy access is central to understanding spatial inequities in healthcare availability. The two-step floating catchment area (2SFCA) method is widely used to integrate proximity and availability, yet key challenges remain unresolved in dense urban settings, including catchment sensitivity, and the lack of uncertainty quantification despite reliance on survey-based population estimates. OBJECTIVE:To assess catchment sensitivity to specification using a mode-adjusted 2SFCA approach while propagating uncertainty from the American Community Survey (ACS). METHODS:We conducted a cross-sectional analysis of pharmacy access across 2222 census tracts in New York City using multiple distance- and time-based catchments. We applied a mode-adjusted 2SFCA method that incorporates walking and driving accessibility by decomposing populations demand using tract-level transportation proportions from the ACS. Uncertainty in access estimates was quantified via hierarchical Monte Carlo simulation propagating the margins of errors from the ACS. We used cross-tabulation to quantify agreement in classifying tracts as having fewer than one pharmacy per 10,000 residents across transportation modes. RESULTS:Relative pharmacy access patterns were stable across a wide range of catchment specifications, indicating limited sensitivity to reasonable changes in distance or travel-time thresholds in this dense urban setting. The mode-adjusted approach increased median access and introduced additional variability by accounting for transportation differences. Uncertainty in access estimates was generally negligible but may matter where driving and walking catchments do not overlap. CONCLUSIONS:These findings suggest that, in New York City, catchment size and ACS uncertainty have limited impact on access estimates, while transportation differences and border effects play a larger role.

Efforts to mitigate bias in artificial intelligence (AI) in healthcare have focused heavily on technical solutions such as diversifying training datasets, improving interpretability, and auditing models for fairness. While essential, these approaches overlook a critical dimension: the ways AI systems are implemented and used in practice. Even the most technically fair algorithms can produce inequitable outcomes if clinicians adopt, interpret, or apply them in patterned ways shaped by institutional norms, professional hierarchies, and patient characteristics. Drawing on social science research and our own experiences implementing AI models in healthcare settings, we argue that implementation is an important driver of bias in healthcare AI. We suggest that clinicians' discretionary use of AI could reproduce or even amplify inequities and that "human-in-the-loop" oversight may offer only limited safeguards against these risks. To ensure equitable outcomes, governance strategies must move beyond individual vigilance and adopt a sociotechnical perspective that accounts for the broader systems in which AI tools are embedded. We call for system-level monitoring, institutional accountability, and new frameworks for studying fairness in AI implementation.

OBJECTIVE:Using AI algorithms can exacerbate health disparities if care or resources are allocated away from underserved populations. We evaluated an algorithm for its potential to worsen health disparities across different clinical use cases. MATERIALS AND METHODS/METHODS:This was a retrospective study of patients with heart failure (HF) at an academic health system using an algorithm that predicts pharmacy fill nonadherence to evidence-based HF medications. We compared prediction performance metrics (accuracy, false positive rate, false negative rate), using rate-ratios (RRs), between subgroups with and without known HF care disparities: below vs above median neighborhood-level socioeconomic status (nSES) and Black vs White race. Results were then applied to 3 hypothetical clinical use cases. RESULTS:Among 34 697 patients (13% Black, 10% Hispanic, 65% White), algorithm accuracy was similar across nSES and racial subgroups. The algorithm assigned more false positives for medication nonadherence among low vs high nSES (RR [95%CI] 1.50 [1.44-1.56]) and Black vs White (2.05 [1.92-2.19]) subgroups. The algorithm also assigned fewer false negatives (0.63 [0.59-0.67]) to Black vs White subgroups. When applied to 3 hypothetical use cases, worsening of existing disparities was pertinent for clinical applications where false positives could be particularly harmful (e.g, if predictions of nonadherence prompted lower treatment priority). DISCUSSION/CONCLUSIONS:Although accuracy was similar across demographic groups, differences in false positive and false negative rates revealed that the same prediction may worsen disparities in some use cases, but not others. CONCLUSION/CONCLUSIONS:Evaluation of predictions in the context of clinical use is essential to avoid unintentionally worsening inequities.

Integrated electronic health record databases provide an unprecedented opportunity to enhance knowledge of disease prediction, prevention, and management in real-world settings. The Precision Medicine (PMED) Registry is a cohort of approximately 2 million patients seen at NYU Langone Health inpatient and outpatient centers, capturing data generated during clinical care from January 1, 2010, to the present, with regular data updates. Data have been used for several research investigations, including international meta-analyses, validation of disease identification algorithms, local evaluation of risk tools, testing analytical pipelines for imaging data, and the investigation of novel correlates of established risk prediction models. Additionally, the assessment of local practice has provided insights into clinical practice patterns and aided quality improvement efforts to assess and promote the uptake of guideline-directed therapies at the system and provider level. This study illustrates how real-world integrated electronic health record data with multi-modal clinical information can be leveraged to support research in prediction, diagnosis, prevention, and treatment optimization across health systems.

BACKGROUND:De-implementation-reducing low-value or harmful care-is critical but difficult in clinical practice. Clinical decision support (CDS) "nudges" in electronic health records (EHRs) aim to promote guideline-concordant deprescribing, but effects are inconsistent. In a pragmatic randomized controlled trial across a large health system, we tested a suite of EHR-based CDS nudges to support Choosing Wisely-aligned deprescribing of glycemic medications in older adults with type 2 diabetes. Although a prior pilot showed modest improvement in guideline concordance (5.1%), the full trial found no significant changes in prescribing; this process evaluation examines clinicians' comments on alerts to explain why. METHODS:We conducted a mixed-methods process evaluation of comments within EHR-based alerts from a null-result RCT that promoted Choosing Wisely deprescribing for older adults with type 2 diabetes. Among 66,634 alerts firing across EHR encounters (December 2016-July 2023), providers commented on 764 (1.2%). Two researchers independently coded comments using reflexive thematic analysis, identifying four themes (three negative). Exploratory logistic and multinomial regressions examined predictors of commenting, valence, and themes among acknowledged firings, adjusting for patient, provider, and encounter factors. RESULTS:Thematic analysis of comments revealed three barriers to deprescribing: (1) disagreement with Choosing Wisely guidelines (308 comments, e.g., perceived low overtreatment risk); (2) workflow misalignment (203 comments, e.g., wrong provider responsibility); and (3) patient preferences (69 comments). Logistic regression showed multiple concurrent OPAs reduced action odds by 31.6% (OR 0.684, 95% CI 0.560-0.835); comments were 2.57 times more likely to be negative than positive (OR 2.565, 95% CI 1.637-4.018). Disparities in engagement were found, with female providers, patients, and socially vulnerable individuals less likely to comment. CONCLUSION/CONCLUSIONS:This process evaluation demonstrates scalable real-time feedback for clinical decision support refinement in de-implementation, with regressions identifying context-specific predictors. Provider disagreement, alert firings misaligned to workflows, and patient resistance hinder effectiveness. Future work should refine clinical decision support design to address complexity, enhance guideline explainability to build provider concordance, align with provider roles and workflows, and include patient-centered approaches. TRIAL REGISTRATION/BACKGROUND:The NYU School of Medicine Institutional Review Board (i17-01308) approved the trial, which has the clinicaltrials.gov ID NCT04181307 (https://clinicaltrials.gov/study/NCT04181307), with a first record date of November 26, 2019.

BACKGROUND:Communication failures are a leading cause of sentinel events in U.S. healthcare, often due to unclear provider contact identification. The electronic health record (EHR) system offers a solution by enabling the discrete assignment of a first contact provider (FCP), who oversees and coordinates patient care. However, adoption of this practice is inconsistent across many hospital settings. This study describes the impact of continuous learning and improvement cycles to address this challenge. METHODS:Following the Plan-Do-Study-Act (PDSA) lifecycle, we completed five quality improvement cycles. Each PDSA cycle included a technological intervention accompanied by evolving operational expectations for clinical staff. We evaluated improvement after each PDSA by measuring the percent of a hospitalized patient's time with an assigned FCP. RESULTS:FCP coverage significantly improved from a baseline average of 5.1% to 59.0% after PDSA Cycle 1 (p < 0.001), 67.4% after Cycle 2 (p < 0.001), 79.7% after Cycle 3 (p < 0.001), 87.5% after Cycle 4 (p < 0.001), and 99.4% after Cycle 5 (p < 0.001). CONCLUSION/CONCLUSIONS:Having a reliable FCP at any point during a patient's hospital admission is an important safety practice. Continuous learning and improvement cycles, driven by a strong partnership between technology and operations, led to significant and sustained improvements in FCP assignments.

The effectiveness of COVID-19 vaccination in children and adolescents with prior SARS-CoV-2 infection remains unclear, particularly for Omicron subvariants. We evaluate vaccine effectiveness against reinfection with Omicron BA.1/BA.2, BA.4/BA.5, XBB, and later subvariants among 5- to 17-year-olds using data from the RECOVER initiative, a national electronic health record database covering 37 U.S. children's hospitals and health institutions. We emulate target trials by age group and variant period, comparing previously infected participants between January 2022 and August 2023. During the BA.1/BA.2 period, vaccination reduces the risk of reinfection, with effectiveness rates of 62% in children and 65% in adolescents. During the BA.4/BA.5 period, protection effectiveness in children was 57%, whereas no statistically significant protection is observed in adolescents. During the XBB and later period, no significant protection is observed in either group. In summary, COVID-19 vaccination provides protection against reinfection during the early and mid-Omicron periods in previously infected pediatric populations, but effectiveness declines for later variants.