Faculty Bibliography

OBJECTIVE:Long COVID (LC) remains poorly understood, and there is a critical need for advanced computational tools to better identify and characterize patients. In this study, we use summarized symptom reports by RECOVER-Adult cohort participants linked to EHR data to characterize patients and train a computable phenotype algorithm of LC. MATERIALS AND METHODS/METHODS:The study included adult participants with linked FHIR-sourced EHR data. We characterized EHR diagnoses, procedures, medications, lab tests, and vital sign features associated with LC. A computable phenotyping algorithm was trained and validated against patient-reported symptoms. MAIN OUTCOME AND MEASURES/METHODS:We assessed model discrimination and calibration in a held-out test set. We describe important model features and evaluate model discrimination and calibration. RESULTS:The study included 1,501 RECOVER-Adult cohort participants with linked EHR data. 376 (25%) met criteria for highly symptomatic LC based on the RECOVER Long COVID Research Index (LCRI). EHR features associated with LC included clinician diagnosis of shortness of breath, malaise and fatigue, and cardiac dysrhythmias; documented treatment with albuterol, gabapentin, or duloxetine; or elevated heart rate. The algorithm identifying patients with highly symptomatic LC had an AUROC of 0.80 (95% confidence interval (CI) 0.74-0.85), and AUPRC of 0.58 (95% CI, 0.47-0.69). CONCLUSION AND RELEVANCE/CONCLUSIONS:These findings demonstrate that, using EHR data, a machine-learning model can accurately select patients with sets of self-reported LC symptoms. The model could help identify patients within a health system with the highest probability of the condition and facilitate screening, recruitment for clinical trials, and etiologic studies.

OBJECTIVE:Long COVID (LC) remains poorly understood, and there is a critical need for advanced computational tools to better identify and characterize patients. In this study, we use summarized symptom reports by RECOVER-Adult cohort participants linked to EHR data to characterize patients and train a computable phenotype algorithm of LC. MATERIALS AND METHODS/METHODS:The study included adult participants with linked FHIR-sourced EHR data. We characterized EHR diagnoses, procedures, medications, lab tests, and vital sign features associated with LC. A computable phenotyping algorithm was trained and validated against patient-reported symptoms. MAIN OUTCOME AND MEASURES/METHODS:We assessed model discrimination and calibration in a held-out test set. We describe important model features and evaluate model discrimination and calibration. RESULTS:The study included 1,501 RECOVER-Adult cohort participants with linked EHR data. 376 (25%) met criteria for highly symptomatic LC based on the RECOVER Long COVID Research Index (LCRI). EHR features associated with LC included clinician diagnosis of shortness of breath, malaise and fatigue, and cardiac dysrhythmias; documented treatment with albuterol, gabapentin, or duloxetine; or elevated heart rate. The algorithm identifying patients with highly symptomatic LC had an AUROC of 0.80 (95% confidence interval (CI) 0.74-0.85), and AUPRC of 0.58 (95% CI, 0.47-0.69). CONCLUSION AND RELEVANCE/CONCLUSIONS:These findings demonstrate that, using EHR data, a machine-learning model can accurately select patients with sets of self-reported LC symptoms. The model could help identify patients within a health system with the highest probability of the condition and facilitate screening, recruitment for clinical trials, and etiologic studies.

IMPORTANCE/UNASSIGNED:Patients frequently forgo filling prescriptions due to high out-of-pocket costs. Real-time prescription benefit (RTPB) tools that recommend available lower-cost, clinically equivalent medications to prescribing clinicians may increase the likelihood that patients fill their prescriptions. OBJECTIVE/UNASSIGNED:To determine whether availability of an RTPB tool increases prescription fill rates. DESIGN AND SETTING/UNASSIGNED:This post hoc analysis of a cluster randomized clinical trial included medical practices within an urban ambulatory clinical network randomized to the RTPB tool between January and December 2021. Outpatient prescriptions that were eligible for an RTPB recommendation during the study period were analyzed. Data analyses were performed from October 18, 2022, to August 9, 2024. INTERVENTION/UNASSIGNED:An electronic health record-integrated RTPB tool that displays available lower-cost and clinically equivalent alternatives to the initiated prescription at the point of prescribing. MAIN OUTCOME AND MEASURE/UNASSIGNED:The primary outcome measured whether a prescription was filled. RESULTS/UNASSIGNED:Of 1 386 577 outpatient prescriptions at randomized practices during the trial period, 38 289 (2.8%) were included in the analytic sample. Across all orders, the availability of the RTPB tool did not impact the proportion of orders filled (54% and 55% in the control and RTPB groups, respectively; adjusted difference: 1.2 percentage points [pp]; 95% CI, -1.3 to 3.7 pp). However, in the quartile of drug classes with the highest out-of-pocket costs (average out-of-pocket cost for a 30-day fill of >$120.83), fill rates increased from 33% in the control group to 49% in the RTPB group (adjusted difference: 14.5 pp; 95% CI, 8.4-20.6 pp). Similar increases were not detected in lower-cost drug classes. Increases in fill rates within the highest out-of-pocket cost drug classes were largest for patients in the lowest-income communities served by the health system (30.3 pp; 95% CI, 19.5-41.1 pp) but not substantial in the highest-income communities (1.0 pp; 95% CI, -10.2 to 12.2 pp). CONCLUSIONS AND RELEVANCE/UNASSIGNED:In this post hoc analysis of a cluster randomized clinical trial, there was no change in overall prescription fill rates, but among high-cost drugs, the RTPB tool increased fill rates, particularly among patients from low-income communities. However, RTPB recommendations were made for a small proportion of orders, limiting the applicability of the findings to a narrow segment of the randomized population. TRIAL REGISTRATION/UNASSIGNED:ClinicalTrials.gov Identifier: NCT04940988.

BACKGROUND/UNASSIGNED:Skilled nursing facilities (SNFs) play a critical role in postacute recovery for older adults with heart failure (HF), yet the transition from SNF to home remains a vulnerable and understudied phase of care. Although discharge guidelines emphasize clear communication, HF-specific self-care education, medication management, and follow-up coordination, little is known about how these practices are implemented during SNF-to-home transitions. METHODS/UNASSIGNED:We conducted a convergent mixed-methods study across 4 nonprofit SNFs, integrating data from postdischarge patient and caregiver surveys, structured medical record abstraction of discharge instructions, and semi-structured staff interviews. Eligible patients were Medicare beneficiaries aged ≥65 years discharged from SNF to home following HF hospitalization, with SNF stays ≤60 days. Quantitative data were analyzed using descriptive statistics, while qualitative data underwent thematic and directed content analysis. Findings were triangulated across data sources to identify key challenges and actionable strategies. RESULTS/UNASSIGNED:Among 150 respondents, 59% reported receiving written discharge instructions; however, HF-specific self-care elements (eg, daily weight monitoring, low salt diet) were documented in only 15% to 41% of instructions. Although 87% reported receiving a medication list, only 53% had it reflected in the discharge instructions, and adherence support was infrequently addressed (24%). Follow-up coordination was similarly discordant: 37% of respondents reported a scheduled primary care appointment, compared with 13% documented in discharge instructions. Staff interviews revealed nonstandardized discharge workflows, workforce constraints, and reliance on verbal education, contributing to variability in patient preparation and communication across care settings. CONCLUSIONS/UNASSIGNED:SNF-to-home transitions after HF hospitalization are marked by discordance between patient-reported education and written documentation, as well as inconsistent medication and follow-up coordination. These gaps represent modifiable vulnerabilities during a high-risk recovery period. Standardized HF-focused discharge workflows and strengthened cross-setting communication may improve transitional care, while long-term solutions must address structural and workforce constraints.

Learning health systems (LHSs) integrate evidence generation, implementation, and evaluation into routine care, enabling health systems to continuously improve quality of care and patient safety. Advances in artificial intelligence (AI), data infrastructure, and rapid evaluation methods have created a pivotal opportunity to accelerate this transformation. Effective LHSs embed pragmatic trials, human-centered design, and continuous monitoring to deploy innovations efficiently while maintaining ethical and operational standards. Key challenges include premature deployment, poor ethical oversight, a lack of funding, clinician burden, and misaligned incentives for payers and providers. Strategic actions - investing in infrastructure, scaling learning, aligning collaborators, and leveraging AI - can bridge the gap between discovery and practice, ensuring that health systems translate innovation into measurable value.

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.

IMPORTANCE/UNASSIGNED:Colorectal cancer screening rates in the US remain suboptimal, particularly among low-income and minoritized populations, despite the availability of effective, low-cost options such as the fecal immunochemical test (FIT). Scalable outreach strategies are needed to improve uptake and reduce staff burden in safety-net settings. OBJECTIVE/UNASSIGNED:To evaluate whether a behavioral economics-informed, automated text messaging strategy was associated with increased FIT completion compared with nurse-led telephone call outreach. DESIGN, SETTING, AND PARTICIPANTS/UNASSIGNED:This quality improvement randomized clinical trial was conducted from April 7 to June 24, 2025, at 8 Federally Qualified Health Centers (FQHCs) in Brooklyn, New York, within the Family Health Centers at NYU Langone. Participants included adults (aged ≥18 years) with a new FIT order who listed English, Spanish, or Chinese (Mandarin or Cantonese) as their preferred language and had not opted out of text messaging. INTERVENTION/UNASSIGNED:Patients were randomized 1:1 to receive either 3 automated, 1-way text message reminders on days 2, 5, and 8 (intervention) or a single nurse-led telephone call reminder on day 8 (usual care). MAIN OUTCOMES AND MEASURES/UNASSIGNED:The primary outcome was FIT completion within 21 days of the test order, assessed from the electronic health record. Secondary outcomes included completion at 7 and 14 days. FIT completion at 7, 14, and 21 days was compared between groups using χ2 tests. RESULTS/UNASSIGNED:Among 1275 eligible randomized participants, 649 were assigned to the text group (418 female participants [64.4%]; mean [SD] age, 56.4 [9.3] years) and 626 to the telephone group (398 female participants [63.6%]; mean [SD] age, 56.7 [9.6] years). FIT completion within 21 days was higher in the text group (382 of 649 participants [58.9%]) compared with the telephone group (312 of 626 participants [49.8%]) with an absolute difference of 9.0 percentage points (95% CI, 3.6-14.5 percentage points; P = .001). Post hoc analyses found no evidence of differential effectiveness by age, sex, race and ethnicity, or patient portal use. CONCLUSIONS AND RELEVANCE/UNASSIGNED:In this quality improvement randomized clinical trial, a behaviorally informed text messaging strategy was associated with significantly improved FIT completion compared with usual nurse-led telephone outreach. Automated messaging may offer a scalable, low-cost strategy to promote preventive care and reduce staff burden in underserved populations. TRIAL REGISTRATION/UNASSIGNED:ClinicalTrials.gov Identifier: NCT06632054.

BACKGROUND:Healthcare systems use electronic patient reminders to encourage preventive care receipt. We used an innovative randomized testing approach to identify reminder design elements associated with improved patient responsiveness. AIM/OBJECTIVE:Improve patient response to electronic preventive care screening reminders. SETTING AND PARTICIPANTS/METHODS:Outpatient clinics affiliated with NYU Langone Health (NYULH) and their patients. Included patients were due for retinal, hemoglobin A1c (HbA1c), or mammography screening. PROGRAM/METHODS:Iterative EHR reminder redesign using a behaviorally informed framework, testing improvements based on randomized implementation. We randomized patients for each screening type to intervention (redesigned) or control (existing) reminders. Outcomes were scheduling or completion of indicated test. Plan-Do-Study-Act Cycle #1 tested all three redesigned reminders. Cycle #2 tested a personalized reminder versus the redesigned mammography reminder from Cycle #1. RESULTS:Cycle #1: Redesigned mammography reminders showed greater adherence (25/212 [11.8%] vs. 13/212 [6.1%]) compared to controls (p = 0.04). Redesigned retinal and HbA1c reminders showed no difference. Cycle #2: Scheduling adherence for mammography was similar for personalized and non-personalized reminders (p = 0.68). CONCLUSION/CONCLUSIONS:Iterative, randomized testing of re-designed electronic preventive care reminders was feasible within a large, complex healthcare system and showed the value of brief messages with a scheduling link for mammography.

IMPORTANCE/UNASSIGNED:Pancreatic ductal adenocarcinoma (PDAC) is a leading cause of cancer deaths in the US. Although early detection improves survival, the rarity of the disease has rendered population screening a difficult approach. OBJECTIVE/UNASSIGNED:To develop and validate a parsimonious, interpretable, and generalizable model predicting incident PDAC-termed PRIME (PDAC Risk Model for Earlier Detection)-using routinely available electronic health record (EHR) data. DESIGN, SETTING, AND PARTICIPANTS/UNASSIGNED:This cohort study used the Optum Labs Data Warehouse, a longitudinal, deidentified US EHR and claims database. Adults 40 years or older with an outpatient clinical encounter between 2016 and 2018 were included. Participants from 23 health systems (n = 4 859 833) comprised the training cohort; 31 additional systems (n = 5 619 091) served as validation. International validation was conducted in the UK Biobank (n = 498 754). Data analysis occurred July 2025 to January 2026. EXPOSURES/UNASSIGNED:Demographics, diagnosis codes, and routinely measured laboratory values were evaluated. Elastic-net regularization with 10-fold cross-validation selected the predictor set. MAIN OUTCOMES AND MEASURES/UNASSIGNED:Incident PDAC was identified by International Classification of Diseases, Ninth and Tenth Revisions (ICD-9/10) codes. Model performance was assessed using time-dependent area under the curve (AUC) and calibration metrics. RESULTS/UNASSIGNED:Overall, the study included more than 11 million adults (2.1% Asian individuals, 8.4% Black individuals, 4.3% Hispanic/Latino individuals, 82.7% White individuals, and 2.4% other race/ethnicity by EHR reporting). In the training cohort (mean [SD] age, 60.4 [11] years), 14 405 individuals were diagnosed with PDAC (incidence 55 per 100 000 person-years) over a mean (SD) of 5.4 (2.5) years; in the validation cohort, 11 693 individuals were diagnosed with PDAC (54 per 100 000 person-years) over a mean (SD) of 3.9 (2.5) years. PRIME retained 19 predictors including history of pancreatitis, gastrointestinal disorders, prior cancers, type 2 diabetes, elevated aspartate aminotransferase levels, smoking, non-type-O blood, and male sex. Discrimination was strong at the 36-month time horizon (AUC = 0.75 in both the training and validation cohorts) with good calibration. In the validation cohort, patients in the top 1% of predicted risk had substantially higher PDAC risk (HR, 7.63; 95% CI, 6.85-8.49) compared with average-risk patients. In the UK Biobank, PRIME achieved a 36-month AUC of 0.71 with good calibration. CONCLUSIONS AND RELEVANCE/UNASSIGNED:In this validation cohort study, PRIME was a transparent EHR-based model that effectively stratified PDAC risk across diverse US health systems and generalized internationally. Prospective studies should evaluate for EHR-guided PDAC case-finding and integration with blood-based early-detection assays.