Publications

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.

INTRODUCTION: Higher buprenorphine maintenance doses (>24 mg daily) may be more effective in improving opioid use disorder (OUD) treatment outcomes in the era of fentanyl predominance. Evidence on maintenance dose trends remains limited, especially among vulnerable populations, who are disproportionately affected by opioid overdose deaths. The objective of this study was to describe trends in maintenance dose of buprenorphine for OUD and the association of social vulnerability with maintenance dose.

METHODS: We identified buprenorphine treatment episodes initiated between January 2019 and June 2021 using the IQVIA Longitudinal Prescription database. Episodes were assigned to a maintenance dose category of <16, 16-24, or >24 mg, to reflect dosages lower than, consistent with, or higher than the guideline-recommended target dose range, respectively. County-level social vulnerability was measured with the Minority Health Index (MHI). We used multinomial logistic regression with generalized estimating equations to test the association between MHI (stratified into quartiles) and maintenance doses <16 mg and >24 mg (with 16-24 mg as the reference group), adjusting for patient, geographic, and time characteristics.

RESULTS: Of 1,044,460 treatment episodes among 543,326 individuals (mean [SD] age 41.9 [12.6] years, 43% female), the most common buprenorphine maintenance dose category was 16-24 mg (62.7%), followed by <16 mg (34.5%), then >24 mg (2.8%). The median (IQR) episode duration was 64 (109) days. The proportion of episodes in the <16 mg category increased from 34.1% in 2019 to 39.2% in 2021, whereas prevalence of the remaining categories decreased. A greater proportion of episodes in the >24 mg category were in rural versus urban counties; the magnitude of this difference increased over time. Relative to 16-24 mg, maintenance dose >24 mg was more likely among individuals in the most vulnerable MHI quartile compared to the least vulnerable quartile (aOR 1.28, 95% CI 1.19-1.38).

CONCLUSION: During a period of growing evidence supporting higher buprenorphine maintenance doses, the prevalence of treatment episodes with maintenance dose <16 mg increased. Individuals in the highest social vulnerability quartile tended to receive higher doses, suggesting that some higher-risk groups may receive more intensive treatment, even as important gaps remain. These results can inform efforts targeting equitable, high-quality buprenorphine prescribing.

OBJECTIVES: Obesity has a US prevalence of more than 40% and is associated with many comorbid conditions, posing a significant burden on employers. Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are recently available and effective weight loss agents. We examined characteristics and outcomes of employees with obesity and those using vs not using GLP-1 RAs.

STUDY DESIGN: Retrospective analysis of employee patients in Workpartners Research Reference Database from 2016 to 2023.

METHODS: Employees with obesity claims were identified and assigned to annual cohorts based on first year of obesity diagnosis (index). Study employees had at least 1 year of continuous data following their index diagnosis. Annual employee characteristics, comorbidities, absences, disability claims, and direct cost trends were explored for the year following diagnosis. Employees with obesity using and not using GLP-1 RAs were compared on the same metrics. Costs were inflation adjusted to December 2023 US$.

RESULTS: We identified 127,408 employees with obesity. Obesity prevalence increased during the study. Employees with obesity and type 2 diabetes decreased slightly, and other comorbidities were relatively stable during the time frame. Overall, 5.8% of employees with obesity (n = 7359) used a GLP-1 RA. GLP-1 RA use increased annually (3.6% in 2016 to 18.3% in 2023) and accounted for approximately 30% of the cohort's 2023 pharmacy costs. During the 12-month study period, compared with non-GLP-1 RA users, those using GLP-1 RAs had higher Charlson Comorbidity Index scores (difference = 0.71), higher proportions with all study comorbidities, $11,360 higher direct all-category costs (total medical costs were $12,092 [19.4%] higher), and 0.86 more absence days.

CONCLUSIONS: Prevalence of obesity is increasing, and use of GLP-1 RAs as the preferred antiobesity medication has increased as well. The long-term impact of this increased use warrants monitoring and management.

PURPOSE: We sought to describe how often and circumstances in which supply chain disruptions were associated with outpatient drug shortages within the Veterans Health Administration (VHA).

METHODS: We conducted a descriptive analysis of VHA purchasing data for drugs used to treat chronic conditions for outpatients with a supply chain report from 2017 to 2020. In primary analyses, a VHA drug shortage was defined as a ≥ 10% absolute decrease in percentage of doses ordered that were filled by the wholesaler and/or a ≥ 10% relative decrease in total number of doses filled, comparing the 3 months after the reporting date with the 6 months before. We also examined longitudinal ordering and filling data over 12 months before and after the reporting date by whether the medication resulted in a shortage to VHA.

RESULTS: Of 64 medications with supply chain disruptions, 67% (n = 43) resulted in a shortage to VHA. Bivariable analyses did not identify covariates significantly associated with VHA experiencing a shortage. For medications with a shortage, dosage units ordered sharply increased over the 4 months before the reporting date, peaked at the reporting date (336.0%), and then generally decreased over the remaining 11 months. The monthly dosage units filled slowly declined leading up to the reporting date, dropped more quickly to 60% at two months post-reporting date, and then increased.

CONCLUSIONS: Among ambulatory-focused medications for chronic conditions, a high proportion of those with supply chain disruptions were associated with shortages to VHA. Sharp increases in medication ordering may serve as an early signal of shortages.

BACKGROUND: Guidelines on primary prevention of atherosclerotic cardiovascular disease (ASCVD) recommend considering risk enhancing factors (REFs) to inform decisions on use of lipid-lowering therapies for the 44 million Americans with borderline or intermediate estimated ASCVD risk.

OBJECTIVE: To quantify REF burden and estimate whether REFs are associated with higher likelihood of statin use.

DESIGN: Cross-sectional using National Health and Nutrition Examination Survey pooled 2015 to March 2020.

PARTICIPANTS: Adults aged 40-75 without known ASCVD.

MAIN MEASURES: Nine REFs were measurable in NHANES including family history of premature ASCVD, rheumatoid arthritis, metabolic syndrome, premature menopause, chronic kidney disease and biomarkers of hypercholesteremia, hypertriglyceridemia, elevated low-density lipoprotein cholesterol and elevated C-reactive protein. Prevalence of REFs and association with statin use were calculated overall and across ASCVD treatment categories based on diabetes and 10-year ASCVD risk.  KEY RESULTS: In the sample of 3,111 participants (mean age 56 years; 52.6% female) weighted to represent 115.7 million US adults, 77% had at least one REF and 28% had at least three. The most common REFs were elevated high sensitivity C-reactive protein (49.6%) metabolic syndrome (48.3%), and hypertriglyceridemia (18.9%). The presence of any REF was associated with a 2.17 (95% CI, 1.42-3.33) greater odds of statin use. Metabolic syndrome, family history of premature ASCVD, hypertriglyceridemia and hypercholesteremia were associated with greater odds of statin use (aORs ranging from 1.5 to 2.3). The presence of any REF was associated with higher likelihood of statin use in participants with low (< 5%), intermediate (7.5-19%), and high ASCVD risk (≥ 20%), but not borderline ASCVD risk (5-7.4%) or diabetes categories.

CONCLUSIONS: REFs are highly prevalent and inconsistently associated with statin use. Achieving the potential benefit of individualizing ASCVD risk estimates will require clearer guidance on when and how to incorporate REFs into primary prevention prescribing decisions.

BACKGROUND: Risks related to long-term opioid therapy for chronic pain are high and may increase over time with aging. Deprescribing may be a beneficial intervention for older adults prescribed chronic opioids.

METHODS: Semi-structured interviews with hypothetical clinical cases of older adults prescribed opioids for chronic pain: (1) low-risk case: a patient prescribed low-dose opioids without concerns; (2) moderate-risk case: a patient with multimorbidity and concurrent benzodiazepine use prescribed moderate opioid doses; (3) high-risk case: a patient prescribed high-dose opioids with signs of an opioid use disorder (OUD). PCPs were asked, in an open-ended fashion, to discuss whether they would initiate a deprescribing conversation, how they would approach deprescribing, and how they would approach a patient who declined recommendations to deprescribe.

PARTICIPANTS AND SETTING: PCPs from a Massachusetts health system.

RESULTS: 18 PCPs participated (56% female, 78% academic). More than half of PCPs would initiate a deprescribing conversation across the three cases. PCPs' approach to deprescribing and mitigating risks differed based on clinical risk. In low and moderate-risk cases, PCPs emphasized a patient-directed taper plan and education on opioid risks. In the high-risk case, some PCPs were uncertain about initiating a deprescribing conversation due to concerns about the patient's mental health and the risk of illicit opioid use. Naloxone was infrequently recommended across the three cases, but in the high-risk case, approximately half of PCPs suggested medications for OUD.

CONCLUSIONS: PCPs reported that they would often initiate opioid deprescribing conversations with older adults, but were less confident in managing older adults with signs of OUD. PCPs require additional support to implement successful conversations on opioid deprescribing with older adults.

BACKGROUND: Healthcare utilization and cost impacts, of medication adherence above or below the 80% threshold remains unclear for cardiometabolic medications.

OBJECTIVE: To evaluate the differences in cardiovascular (CV)-related emergency department (ED) visits and total cost-of-care associated with changes in medication adherence around the 80% threshold.

METHOD: Retrospective observational analysis of claims spanning from Jan 1, 2021, to June 30, 2023. Patients aged 50-80 years with a history of cardiovascular disease (CVD) and specific cardiometabolic medications were followed for 12 months, and pre-post index-fill outcomes compared. Patients were categorized into pre- and post-index groups: 3 pre-index groups (Pre1-moderate, Pre2-high, Pre3-very-high) based on their pre-index 12-month adherence [PDC] and 4 post-index groups based on post-index PDC (Post0-low, Post1-moderate, Post2-high, Post3-very-high), for 4 medication classes (Antidiabetics, direct oral coagulants[DOACs], antiplatelets, and anti-heart-failure meds [HF]). Group definitions: Post0-low (PDC<=69%), Pre1-moderate or Post1-moderate (PDC 70 to <80%), Pre2-high or Post2-high (PDC 80 to <90%), and Pre3-very-high or Post3-very-high (PDC >=90). Outcomes included per-member-per-year [PMPY] total cost-of-care, and cardiovascular-related ED visits.

RESULTS: There were 55,934(antidiabetics), 46,290 (DOACs), 65,659 (antiplatelets), and 49, 670 (HF) patients in the final sample. Most of the patients in the HF (46-53%) and DOAC (51-57%) groups were in the 70+ age group. Among patients in the antidiabetic (45-47%) and antiplatelet (39-43%) groups, the majority were in the 60-69 age group. In general, patients who moved from a lower adherence group to a higher adherence group had lower total cost-of-care in almost all groups and medication classes. Moving from PDC >= 90 to PDC <90%, total cost-of-care was higher in all 4 medication classes.

CONCLUSION: Improving adherence to and beyond the traditional 80% target was associated with lower total cost-of-care.

BACKGROUND: Chronic benzodiazepine use remains common among older adults, despite limited evidence of benefit and substantial risks. Evidence-based approaches that are feasible in primary care settings are needed to support benzodiazepine deprescribing for older adults.

OBJECTIVES: To determine the feasibility, acceptability, and exploratory patient-centered outcomes of a team-based approach to benzodiazepine deprescribing in primary care.

DESIGN: Single-arm prospective clinical trial.

PARTICIPANTS: Adults age 65 and older prescribed long-term benzodiazepines recruited from four primary care clinics in an academic health system.

INTERVENTIONS: Ten-week virtual primary care embedded program consisting of pharmacist-guided tapering and three psychologist-led mindfulness-informed cognitive behavioral therapy (CBT) sessions.

MAIN MEASURES: Feasibility outcomes included enrollment, retention, and intervention adherence. Acceptability outcomes were collected through qualitative interviews. Exploratory efficacy outcomes included change in mean daily benzodiazepine dose, change in PROMIS anxiety score, and change in PROMIS sleep disturbance score.

KEY RESULTS: Seventeen participants (mean age 72, 29% female, mean 17 years of benzodiazepine use) enrolled and completed all six study visits (100% fidelity). Participants' mean (SD) baseline daily benzodiazepine dose was 9.0 (8.6) diazepam milligram equivalents. At completion, participants had a mean 60.5 percentage point reduction in benzodiazepine use (95% CI -69.9% to -51.3%; p < .001), with all participants reducing their dose and 3 stopping completely. Mean PROMIS anxiety scores decreased from 55.2 to 51.8 (-3.5 point change, 95% CI -6.5 to -0.7; p = 0.02) and mean PROMIS sleep disturbance scores were unchanged at week 10 (-1.7 point change, 95% CI -4.9 to 1.5; p = 0.30). Qualitative interviews indicated the program may target increased self-efficacy to reduce benzodiazepines and endorsed utility from both pharmacist- and psychologist-led components.

CONCLUSIONS: A primary care embedded virtual intervention involving pharmacist-led tapering and mindfulness-informed CBT sessions to support benzodiazepine deprescribing is feasible, acceptable, may reduce older adults' benzodiazepine use, and warrants multi-site testing.

GOV TRIAL NUMBER: NCT06119308.

GOV REGISTRATION DATE: 10/27/2023.

BACKGROUND: In older adults with osteoarthritis (OA) and hypertension (HTN), analgesic use may elevate blood pressure and cardiovascular risk. Whether comorbid HTN influences initial analgesic choice remains unclear; we examined initial analgesic use in Medicare beneficiaries with incident OA, comparing those with and without HTN.

METHODS: We conducted a retrospective cohort study using 2011-2022 nationally representative Medicare beneficiaries (≥ 65 years) with incident OA who initiated an analgesic within 30 days of diagnosis and had continuous enrollment for ≥ 365 days prior through ≥ 30 days post-index. Patients with baseline HTN were classified as OA + HTN; others as OA-only. We assessed overall analgesic trends using the Cochran-Armitage test and evaluated differences by HTN status using logistic regression with year as an interaction term. For stratified analyses by joint type, we applied weighted logistic regression.

RESULTS: Among 179,033 beneficiaries (mean age 75 ± 7.3 years; 62.7% women; 80.7% White), 57.1% had baseline HTN. Overall, the most commonly initiated analgesic classes were intra-articular injections (30.3%), and oral NSAIDs only (28.2%). Notable changes from 2012 to 2022 were increase in topical NSAIDs use (3.1%-5.7%) and decrease in opioid combination use (25.4%-13.9%), with no significant trend differences by HTN status. In joint-specific analyses, OA + HTN versus OA-only showed no differences in odds of initiating oral opioids (OR: 0.97, 95% CI: 0.92-1.03), intra-articular injections (OR: 1.01, 95% CI: 0.96-1.07) or topical NSAIDs (OR: 0.88, 95% CI: 0.78-1.01) versus oral NSAIDs.

CONCLUSION: Baseline HTN did not influence the choice of initial analgesic in incident OA patients. Safer, evidence-based alternatives are needed for older adults with comorbid HTN.