Walid F. Gellad

Rationale: Sarcoidosis disproportionately affects U.S. military Veterans. Its rising incidence and prevalence among Veterans are higher than in civilians and vary by branch of service, suggesting a role for prior military exposures. Objectives: To investigate whether estimated service-related exposure to inorganic dust is associated with increased risk of sarcoidosis. Methods: In a retrospective nested case-control study of Veterans enrolled in the Veterans Health Administration (VHA) who received VHA- or Medicare-covered care, sarcoidosis cases identified in electronic health records between 2002 and 2022 were each matched to five controls by propensity scores and incidence density sampling by calendar year. To estimate inorganic dust exposure, military occupational codes (MOCs) were linked to a job exposure matrix (JEM). Multivariable conditional logistic regression analyses were adjusted for demographics, geography, rurality, and service branch. The attributable fraction among the exposed was calculated. In secondary analyses, latency from service separation to diagnosis was assessed. In sensitivity analyses, exposure misclassification was assessed by using a modified JEM weighted for MOC coding precision and by examining only post–9/11 Veterans; service duration was also analyzed. Results: Among >16 million individuals evaluated, 3,019,636 Veterans had at least one MOC; 5,855 incident cases and 29,275 controls were analyzed. Inorganic dust exposure was associated with increased risk of sarcoidosis (conditional odds ratio [cOR]: 1.08; 95% confidence interval [CI], 1.003–1.170); the attributable fraction was 7.6%. Associations were stronger with a modified JEM score (cOR: 1.15; 95% CI 1.05–1.27) and when restricted to post-9/11 Veterans (cOR: 1.25; 95% CI 1.07–1.47). In latency analysis, sarcoidosis risk peaked 2–5 years post-service (cOR: 2.10; 95% CI: 1.20–3.58). Conclusion: Occupational and environmental exposure to inorganic dust during military service was associated with increased risk of sarcoidosis among Veterans, carrying potential implications for prevention and surveillance.

Objective: To investigate whether eligibility for Veterans Health Administration (VA)‐purchased community care, which expanded Veterans' access to care outside VA, was associated with increased polypharmacy or potentially inappropriate medication use among older adult Veterans. Study Setting and Design: Regression discontinuity design, leveraging the distance threshold for community care eligibility (residing > 40 miles from the nearest VA facility with ≥ 1 or more full‐time primary care physician), to examine the effects of community care eligibility on polypharmacy and potentially inappropriate medication use among Veterans aged ≥ 65 years. Data Sources and Analytic Sample: VA pharmacy data for all prescriptions filled at VA facilities, VA Program Integrity Tool files for prescriptions paid by VA and filled in community pharmacies, and Medicare Part D data. Analyses included annual cross‐sectional samples of Veterans 36–39 miles or 41–44 miles from their nearest VA facility during FY 2016–2019. Principal Findings: The sample included 399,250 Veteran‐year observations, of which 226,157 (56.6%) were 36–39 miles and 173,093 (43.4%) were 41–44 miles from the nearest eligible VA facility. Overall, we observed no discontinuities across the 40‐mile threshold in the number of unique medications filled annually (−0.06 medications; 95% confidence interval [CI], −0.15 to 0.03). There were no discontinuities in proportions of Veterans filling ≥ 5 unique medications (−0.25 percentage points [pp]; 95% CI, −0.84 to 0.34), ≥ 10 medications (−0.55 pp.; 95% CI, −1.24 to 0.14), ≥ 1 medication on the Beers list (−0.02 pp.; 95% CI, −0.63 to 0.59), or ≥ 1 high‐risk drug–drug interaction (−0.05 pp.; 95% CI, −0.17 to 0.07). Among Veterans with mental health conditions, exceeding the 40‐mile threshold was associated with a higher likelihood of filling ≥ 10 unique medications annually (2.06 pp.; 95% CI, 0.42 to 3.70). We did not observe clinically or statistically significant discontinuities in other subgroups. Conclusions: Overall, eligibility for VA‐purchased community care was not associated with increased polypharmacy or potentially inappropriate medication use among older adult Veterans.

Introduction: Although drug shortages for outpatient chronic conditions commonly occur, population‐level data on how they impact patients' ability to refill prescriptions is scarce. We sought to identify distinct patterns of refill adherence following drug shortages and patient‐ and prescription‐level factors associated with adherence trajectories reflecting potential shortage‐related treatment disruption. Methods: We retrospectively analyzed panel data assembled from 2017 to 2020 Veterans Health Administration (VHA) electronic health record data. Patients were included if they were baseline users of medications subject to a shortage within VHA. Group‐based trajectory modeling was applied to users' monthly proportion of days covered (PDC) values from 6‐months before to 6‐months after the reported drug supply chain disruption. Patient demographics and medication characteristics were compared between identified trajectory groups using multivariable logistic regression. Results: Among 1.5 million episodes of medication use (representing 1.3 million unique Veterans) for 29 medications in shortage in VHA, 6.3% were for female patients and the mean age was 66.4 ± 12.8 years. A 4‐group trajectory model had the best fit: High Adherence (69.2% of observations), Moderate Adherence (14.1%), Potential Shortage‐Related Disruption (8.5%), and Pre‐Shortage Disruption (8.3%). Drug characteristics (drug class, number of manufacturers) were more strongly associated than patient characteristics with having Potential Shortage‐Related Treatment Disruption vs. High Adherence. Conclusions: We identified 4 trajectories of refill adherence for medications subject to VHA drug shortages, with 8.5% of users of affected drugs exhibiting a trajectory consistent with shortage‐related treatment disruption. Drug characteristics may modify whether drug shortages lead to treatment disruption in VHA.

Background: Opioid overdose remains a leading cause of preventable death in the United States. Existing approaches to identify individuals at elevated risk rely on imprecise rule-based criteria that misclassify patients’ risk of this serious health outcome. Machine learning (ML) algorithms can help improve prediction performance and can be combined with electronic health record (EHR) interventions to reduce overdose risk. Objective: The Machine Learning Prediction and Reducing Overdoses With EHR Nudges (mPROVEN) clinical trial integrates a validated ML overdose risk model with behavioral economics–informed EHR nudges to test whether the combination improves evidence-based prescribing behaviors associated with lower overdose risk and, ultimately, reduces overdose among elevated-risk patients. Methods: mPROVEN is a pragmatic cluster randomized controlled trial conducted in primary care practices within a large multistate integrated health system. Eligible patients are adults (≥18 years) identified by the ML algorithm as having elevated overdose risk and seen at a primary care visit during the study period. Primary care practices serve as the unit of randomization and will be randomized into three arms: (1) usual care; (2) elevated risk flag only, where clinicians see a noninterruptive EHR flag indicating elevated overdose risk; and (3) elevated risk flag + nudges, in which active choice and accountable justification alerts are embedded within the EHR in addition to the elevated risk flag. The trial will enroll a target cohort of 800 patients for the primary analysis. The intervention period is 4 months (or until the study ends, whichever occurs later). The primary outcome is a 3‑point composite measure of safer opioid prescribing at 4 months, awarding 1 point each for active naloxone prescription, average opioid dosage of 50 morphine milligram equivalents per day or less, and absence of opioid-benzodiazepine overlap. Secondary outcomes include the composite outcome at 6 months, individual score components, and all-cause and overdose-specific emergency department or inpatient visits. Outcomes will be compared across study arms using an intention‑to‑treat approach with linear mixed‑effects models accounting for clinic-level clustering. Results: Funded by the National Institutes of Health, in June 2022, enrollment began on March 10, 2025. Enrollment for the primary analysis cohort (n=798) was completed in May 2025 with additional participants enrolled for secondary analyses through December 2025 (n=1662). Primary cohort analyses began in January 2026, and results are expected by mid-2027. Conclusions: The mPROVEN study is among the first pragmatic randomized controlled trials to integrate ML‑based opioid overdose risk prediction with behavioral nudges within a large health system EHR. By combining advances in data science and behavioral economics, the study aims to reduce opioid overdose risk in primary care using a scalable and low-touch intervention to address a high-priority public health issue. Trial Registration: ClinicalTrials.gov NCT06806163; https://clinicaltrials.gov/study/NCT06806163 International Registered Report Identifier (IRRID): DERR1-10.2196/94007

The objective of this study was to evaluate changes in net and list prices of branded multiple sclerosis (MS) disease-modifying therapies (DMTs) between 2013 and 2021. Using several pharmaceutical pricing and utilization data sources, we estimated list and net (after rebates and discounts) prices for branded monoclonal antibody (MAb) and oral DMTs. We calculated the inflation-adjusted list and net prices for each DMT, the manufacturer discount as a percentage of list price, the average annual change (AAC) in prices, and the cumulative change in list price offset by discounts. From 2013 to 2021, oral DMT list prices increased from $73,924 to $104,372 (5.3% AAC) while net prices rose from $69,187 to $82,181 (1.5% AAC) because of increasing manufacturer discounts (6.4%-21.2%). From 2014 to 2021, MAb DMT list prices increased from $70,320 to $92,109 (4.1% AAC), with net prices rising from $55,109 to $79,396 (3.0% AAC). Discounts offset 51%-86% of cumulative list price increases for oral DMTs (fingolimod, teriflunomide) vs 0%-35% for MAb DMTs. The divergent net pricing trends between oral and MAb DMTs may reflect increasing brand and generic competition among oral DMTs and a lack of biosimilar options among MAb DMTs.

Background: Opioid use disorder (OUD) remains a critical public health crisis in the United States. Despite widespread policy and clinical interventions, early identification of individuals at risk for developing OUD remains challenging due to limitations in traditional screening approaches and a lack of individualized risk stratification methods. Machine learning (ML) methods offer an opportunity to develop timely, high-performing, and explainable predictive models that can enhance OUD prevention strategies in clinical settings. Objective: This study aims to develop and validate an ML model using electronic health record (EHR) data to predict the 3-month risk of incident OUD among adults initiating opioid therapy and to stratify patients into clinically actionable risk groups. Methods: This prognostic modeling study used 2017‐2022 OneFlorida+ EHR data to develop and validate ML algorithms predicting 3-month incident OUD risk. We included 182,083 adults (≥18 y) without cancer, overdose, or OUD or hospice history who received ≥1 outpatient, noninjectable opioid prescription. Using 183 predictors measured in sequential 3-month intervals, we developed an elastic net, least absolute shrinkage and selection operator, gradient boosting machine (GBM), and random forest models on randomly split training, testing, and validation sets. Model performance was assessed using C-statistics, predictive values, and number needed to evaluate, with patients stratified into risk deciles for clinical applicability. Model explainability was assessed using Shapley additive explanations, and fairness was evaluated using standard metrics. We externally validated the best-performing model using an independent cohort from the 2018‐2020 UPMC (formerly University of Pittsburgh Medical Center) health system. Results: In the validation sample (n=60,694), GBM (C-statistics=0.879, 95% CI 0.874‐0.884) and elastic net (C-statistics=0.872, 95% CI 0.867‐0.877) outperformed least absolute shrinkage and selection operator (C-statistics=0.846, 95% CI 0.840‐0.851) and random forest (C-statistics=0.798, 95% CI 0.792‐0.804), with GBM model requiring the fewest predictors (n=75) for predicting 3-month incident OUD. Using the GBM algorithm to predict the subsequent 3-month OUD risk, the top decile subgroup had a positive predictive value of 3.26%, a negative predictive value of 99.8%, and a number needed to evaluate of 31. The top decile (n=6696) captured ~68% of patients with OUD. Shapley additive explanations analysis identified age, number of outpatient visits, history of back and other pain conditions, comorbidity burden, and opioid prescribing patterns as the strongest predictors of incident OUD. Fairness assessment showed an acceptable false negative rate parity across race, age, and sex. In external validation on the UPMC cohort, the GBM model maintained good discrimination (C-statistics=0.756, 95% CI 0.750‐0.762) and effective risk stratification. Conclusions: An ML algorithm predicting incident OUD derived from OneFlorida+ EHR data performed well in external validation with data using UPMC. The algorithm might be valuable for incident OUD risk prediction and stratification across health systems, with potential to inform early intervention.

BACKGROUND Opioid overdose remains a leading cause of preventable death in the United States. Existing approaches to identify individuals at elevated risk rely on imprecise rule-based criteria that misclassify patients’ risk of this serious health outcome. Machine learning (ML) algorithms can help improve prediction performance and can be combined with electronic health record (EHR) interventions to reduce overdose risk. OBJECTIVE The Machine Learning Prediction and Reducing Overdoses With EHR Nudges (mPROVEN) clinical trial integrates a validated ML overdose risk model with behavioral economics–informed EHR nudges to test whether the combination improves evidence-based prescribing behaviors associated with lower overdose risk and, ultimately, reduces overdose among elevated-risk patients. METHODS mPROVEN is a pragmatic cluster randomized controlled trial conducted in primary care practices within a large multistate integrated health system. Eligible patients are adults (≥18 years) identified by the ML algorithm as having elevated overdose risk and seen at a primary care visit during the study period. Primary care practices serve as the unit of randomization and will be randomized into three arms: (1) usual care; (2) elevated risk flag only, where clinicians see a noninterruptive EHR flag indicating elevated overdose risk; and (3) elevated risk flag + nudges, in which active choice and accountable justification alerts are embedded within the EHR in addition to the elevated risk flag. The trial will enroll a target cohort of 800 patients for the primary analysis. The intervention period is 4 months (or until the study ends, whichever occurs later). The primary outcome is a 3‑point composite measure of safer opioid prescribing at 4 months, awarding 1 point each for active naloxone prescription, average opioid dosage of 50 morphine milligram equivalents per day or less, and absence of opioid-benzodiazepine overlap. Secondary outcomes include the composite outcome at 6 months, individual score components, and all-cause and overdose-specific emergency department or inpatient visits. Outcomes will be compared across study arms using an intention‑to‑treat approach with linear mixed‑effects models accounting for clinic-level clustering. RESULTS Funded by the National Institutes of Health, in June 2022, enrollment began on March 10, 2025. Enrollment for the primary analysis cohort (n=798) was completed in May 2025 with additional participants enrolled for secondary analyses through December 2025 (n=1662). Primary cohort analyses began in January 2026, and results are expected by mid-2027. CONCLUSIONS The mPROVEN study is among the first pragmatic randomized controlled trials to integrate ML‑based opioid overdose risk prediction with behavioral nudges within a large health system EHR. By combining advances in data science and behavioral economics, the study aims to reduce opioid overdose risk in primary care using a scalable and low-touch intervention to address a high-priority public health issue. CLINICALTRIAL ClinicalTrials.gov NCT06806163; https://clinicaltrials.gov/study/NCT06806163 INTERNATIONAL REGISTERED REPORT DERR1-10.2196/94007

Background/Objectives: The Developing and Evaluating a Machine Learning Opioid Prediction & Risk-Stratification E-Platform (DEMONSTRATE) trial aims to assess the usability, acceptability, feasibility, and effectiveness of implementing a machine learning (ML)-based clinical decision support (CDS) tool—the Overdose Prevention Alert—which predicts a patient’s risk of opioid overdose within three months. Methods: This single-arm study uses a pre–post implementation design with mixed-methods evaluation in 13 University of Florida Health, Gainesville, internal medicine and family medicine clinics. Eligible patients are aged ≥18 years, received an opioid prescription within the year prior to their upcoming primary care visit, are not receiving hospice care, do not have a malignant cancer diagnosis, and are identified by the ML algorithm as high risk for overdose. The Overdose Prevention Alert triggers when a primary care provider (PCP) signs an opioid order in electronic health records. We will evaluate effectiveness by comparing pre- and post-implementation outcomes using a composite patient-level measure defined by the presence of any of the following 6 favorable indicators: (1) evidence of naloxone access; (2) absence of opioid overdose diagnoses and naloxone administration; (3) absence of emergency department (ED) visits or hospitalizations due to opioid overdose or opioid use disorder (OUD); (4) absence of overlapping opioid and benzodiazepine use within a 7-day window; (5) absence of opioid use ≥50 morphine milligram equivalent daily average; (6) receipt of referrals to non-pharmacological pain management. Additional quantitative metrics will include alert penetration, usage patterns, and clinical actions taken. Usability and acceptability will be assessed using a 12-item questionnaire for PCPs and semi-structured interviews. Expected Results: The trial will provide insights into real-world ML-driven CDS implementation and inform future strategies to reduce opioid-related harm.

Drug overdose deaths, including those due to prescription opioids, represent a critical public health issue in the United States and worldwide. Artificial intelligence (AI) approaches have been developed and deployed to help prescribers assess a patient's risk for overdose-related death, but it is unknown whether public health experts can leverage similar predictions to make local resource allocation decisions more effectively. In this work, we evaluated how AI-based overdose risk assessment could be used to inform local public health decisions using a working prototype system. Experts from three health departments, of varying locations and sizes with respect to staff and population served, were receptive to the potential benefits of algorithmic risk prediction and of using AI-augmented visualization to connect across data sources. However, they also expressed concerns about whether the risk prediction model's formulation and underlying data would match the state of the overdose epidemic as it evolved in their specific locations. Our findings extend those of other studies on algorithmic systems in the public sector, and they present opportunities for future human-AI collaborative tools to support decision-making in local, time-varying contexts.