Publications

OBJECTIVES: To develop and validate a machine-learning algorithm to predict fatal overdose using Pennsylvania Prescription Drug Monitoring Program (PDMP) data.

METHODS: The training/testing (n = 3020,748) and validation (n = 2237,701) cohorts included Pennsylvania residents with a prescription dispensing from February 2018-September 2021. Potential predictors (n = 222) were measured in the 6 months prior to a random index date. Using a gradient boosting machine, we developed a 20-variable model to predict risk of fatal drug overdose in the 6 months after the index date.

RESULTS: Beneficiaries in the training (n = 1,812,448), testing (n = 1,208,300), and validation (n = 2,237,701) samples had similar age, with low rates of fatal overdose during 6-month follow up (0.12%, 0.12%, 0.04%, respectively). The validation c-statistic was 0.86 for predicting fatal overdose using 20 PDMP variables. When ranking individuals based on risk score, the prediction model more accurately identified fatal overdose at 6 months compared to using opioid dosage or opioid/benzodiazepine overlap, although the percentage of individuals in the highest risk percentile who died at 6 months was less than 1%.

CONCLUSIONS AND POLICY IMPLICATIONS: A gradient boosting machine algorithm predicting fatal overdose derived from twenty variables performed well in discriminating risk across testing and validation samples, improving on single factor risk measures like opioid dosage.

INTRODUCTION: The opioid epidemic has exacted a significant toll in rural areas, yet adoption of medications for opioid use disorder (MOUD) lags. The Rural Access to Medication Assisted Treatment in Pennsylvania (RAMP) Project facilitated adoption of MOUD in rural primary care clinics. The purpose of this study was to gain a better understanding of the barriers and facilitators operating at multiple levels to access or provide MOUD in rural Pennsylvania.

METHODS: In total, the study conducted 35 semi-structured interviews with MOUD patients and MOUD providers participating in RAMP. Qualitative analysis incorporated both deductive and inductive approaches. The study team coded interviews and performed thematic analysis. Using a modified social-ecological framework, themes from the qualitative interviews are organized in five nested levels: individual, interpersonal, health care setting, community, and public policy.

RESULTS: Patients and providers agreed on many barriers (e.g., lack of providers, lack of transportation, insufficient rapport and trust in patient-provider relationship, and cost, etc.); however, their interpretation of the barrier, or indicated solution, diverged in meaningful ways. Patients described their experiences in broad terms pointing to the social determinants of health, as they highlighted their lives outside of the therapeutic encounter in the clinic. Providers focused on their professional roles, responsibilities, and operations within the primary care setting.

CONCLUSIONS: Providers may want to discuss barriers to treatment related to social determinants of health with patients, and pursue partnerships with organizations that seek to address those barriers. The findings from these interviews point to potential opportunities to enhance patient experience, increase access to and optimize processes for MOUD in rural areas, and reduce stigma against people with opioid use disorder (OUD) in the wider community.

INTRODUCTION: In the United States, dentists frequently prescribe hydrocodone. In October 2014, the US Drug Enforcement Administration rescheduled hydrocodone from controlled substance schedule III to II, introducing more restricted prescribing and dispensing regulations, which may have changed dental prescribing of opioids.

OBJECTIVE: The study aim was to evaluate the impact of the hydrocodone rescheduling on dental prescribing of opioids in the United States.

METHODS: This was a cross-sectional study of opioids prescribed by dentists between October 2012 and October 2016, using the IQVIA Longitudinal Prescription Dataset. Monthly dentist-based opioid prescribing rate (opioid prescription [Rx]/1,000 dentists) and monthly average opioid dosages per prescription (mean morphine milligram equivalent per day [MME/d]) were measured in the 24 mo before and after hydrocodone rescheduling in October 2014 (index or interruption). An interrupted time-series analysis was conducted using segmented ordinary least square regression models, with Newey-West standard errors to handle autocorrelation.

RESULTS: Dentists prescribed 50,412,942 opioid prescriptions across the 49 mo. Hydrocodone was the most commonly prescribed opioid pre- and postindex (74.9% and 63.8%, respectively), followed by codeine (13.8% and 21.6%), oxycodone (8.1% and 9.5%), and tramadol (2.9% and 4.8%). At index, hydrocodone prescribing immediately decreased by -834.8 Rx/1,000 dentists (95% confidence interval [CI], -1,040.2 to -629.4), with increased prescribing of codeine (421.9; 95% CI, 369.7-474.0), oxycodone (85.3; 95% CI, 45.4-125.2), and tramadol (111.8; 95% CI, 101.4-122.3). The mean MME increased at index for all opioids except for hydrocodone, and dosages subsequently decreased during the postindex period.

CONCLUSION: Following the rescheduling, dentist prescribing of hydrocodone declined while prescribing of nonhydrocodone opioids increased. Understanding the impact of this regulation informs strategies to ensure appropriate prescribing of opioids for dental pain.

KNOWLEDGE TRANSFER STATEMENT: The study findings can be used by policy makers to make informed decisions in developing future risk mitigation strategies aimed to regulate opioid prescribing behaviors. Furthermore, dentist-specific resources and guidelines are needed subsequent to these policies in order to meet the dental population needs.

BACKGROUND: Atrial fibrillation (AF) is a common arrhythmia, the management of which includes anticoagulation for stroke prevention. Although disparities in anticoagulant prescribing have been well documented for individual socioeconomic factors, less is known about the association of neighborhood-level disadvantage and anticoagulation for AF.

OBJECTIVE: To assess the association between neighborhood disadvantage and anticoagulant initiation for patients with incident AF.

DESIGN: Retrospective cohort study.

PARTICIPANTS: A cohort of patients enrolled in the Veterans Health Administration (VA) with incident AF from January 2014 through December 2020 from the Race, Ethnicity, and Anticoagulant CHoice in Atrial Fibrillation (REACH-AF) Study.

MAIN MEASURES: The primary exposure was neighborhood disadvantage quantified using area deprivation index (ADI), classified by quintiles (Q). The outcomes were initiation of any anticoagulant therapy (warfarin or direct oral anticoagulant, DOAC) within 90 days of AF diagnosis and DOAC use among initiators. We used mixed effects logistic regression to assess the association between ADI and anticoagulant therapy, incorporating a fixed effect for treatment site and baseline patient, provider, and facility covariates.

KEY RESULTS: Among 161,089 patients, 105,489 (65.5%) initiated any anticoagulant therapy, and 78,903 (74.8%) used DOACs. Any anticoagulant therapy increased 3.2 percentage points (63.0% to 66.2%; p<.001) from Q1 to Q5, whereas DOAC use decreased 8.2 percentage points (79.4% to 71.2%; p<.0001) across quintiles. The adjusted odd ratios of any anticoagulant therapy were non-significantly different for Q2-Q5 than Q1. The adjusted odds of DOAC use decreased progressively from 0.89 (95% CI, 0.84-0.94) in Q2 to 0.77 (95% CI, 0.73-0.83) in Q5 compared to Q1 (p<.0001).

CONCLUSIONS: Among Veterans with incident AF, we observed similar initiation of any anticoagulant, though neighborhood deprivation was associated with decreased DOAC use among anticoagulant initiators. Future interventions to improve pharmacoequity in anticoagulant prescribing for AF should consider the role of neighborhood-level determinants of health inequities.

IMPORTANCE: Improvements in control of factors associated with diabetes risk in the US have stalled and remain suboptimal. The benefit of continually improving goal achievement has not been evaluated to date.

OBJECTIVE: To quantify potential gains in life expectancy (LE) among people with type 2 diabetes (T2D) associated with lowering glycated hemoglobin (HbA1c), systolic blood pressure (SBP), low-density lipoprotein cholesterol (LDL-C), and body mass index (BMI) toward optimal levels.

DESIGN, SETTING, AND PARTICIPANTS: In this decision analytical model, the Building, Relating, Assessing, and Validating Outcomes (BRAVO) diabetes microsimulation model was calibrated to a nationally representative sample of adults with T2D from the National Health and Nutrition Examination Survey (2015-2016) using their linked short-term mortality data from the National Death Index. The model was then used to conduct the simulation experiment on the study population over a lifetime. Data were analyzed from January to October 2021.

EXPOSURE: The study population was grouped into quartiles on the basis of levels of HbA1c, SBP, LDL-C, and BMI. LE gains associated with achieving better control were estimated by moving people with T2D from the current quartile of each biomarker to the lower quartiles.

MAIN OUTCOMES AND MEASURES: Life expectancy.

RESULTS: Among 421 individuals, 194 (46%) were women, and the mean (SD) age was 65.6 (8.9) years. Compared with a BMI of 41.4 (mean of the fourth quartile), lower BMIs of 24.3 (first), 28.6 (second), and 33.0 (third) were associated with 3.9, 2.9, and 2.0 additional life-years, respectively, in people with T2D. Compared with an SBP of 160.4 mm Hg (fourth), lower SBP levels of 114.1 mm Hg (first), 128.2 mm Hg (second), and 139.1 mm Hg (third) were associated with 1.9, 1.5, and 1.1 years gained in LE in people with T2D, respectively. A lower LDL-C level of 59 mg/dL (first), 84.0 mg/dL (second), and 107.0 mg/dL (third) were associated with 0.9, 0.7, and 0.5 years gain in LE, compared with LDL-C of 146.2 mg/dL (fourth). Reducing HbA1c from 9.9% (fourth) to 7.7% (third) was associated with 3.4 years gain in LE. However, a further reduction to 6.8% (second) was associated with only a mean of 0.5 years gain in LE, and from 6.8% to 5.9% (first) was not associated with LE benefit. Overall, reducing HbA1c from the fourth quartile to the first is associated with an LE gain of 3.8 years.

CONCLUSIONS AND RELEVANCE: These findings can be used by clinicians to motivate patients in achieving the recommended treatment goals and to help prioritize interventions and programs to improve diabetes care in the US.

OBJECTIVE: The Medicare Part D Senior Savings Model (SSM) took effect on 1 January 2021. In this study we estimated the number of beneficiaries who would benefit from SSM and the long-term health and economic consequences of implementing this new policy.

RESEARCH DESIGN AND METHODS: Data for Medicare beneficiaries with diabetes treated with insulin were extracted from the 2018 Medical Expenditure Panel Survey. A validated diabetes microsimulation model estimated health and economic impacts of the new policy for the 5-year initial implementation period and a 20-year extended policy horizon. Costs were estimated from a health system perspective.

RESULTS: Of 4.2 million eligible Medicare beneficiaries, 1.6 million (38.3%) would benefit from the policy, and out-of-pocket (OOP) costs per year per beneficiary would decrease by 61% or $500 on average. Compared with non-White subgroups, the White population subgroups would have a higher proportion of SSM enrollees (29.6% vs. 43.7%) and a higher annual OOP cost reduction (reduction of $424 vs. $531). Among the SSM enrollees, one-third (605,125) were predicted to have improved insulin adherence due to lower cost sharing and improved health outcomes. In 5 years, the SSM would 1) avert 2,014 strokes, 935 heart attacks, 315 heart failure cases, and 344 end-stage renal disease cases; 2) gain 3,220 life-years and 3,381 quality-adjusted life-years (QALY); and 3) increase insulin cost and total medical cost by $3.5 billion and $2.8 billion. In 20 years, the number of avoided clinical outcomes, number of life-years and QALY gained, and the total and insulin cost would be larger.

CONCLUSIONS: The Medicare SSM may reduce the OOP costs for approximately one-third of the Medicare beneficiaries treated with insulin, improving health outcomes via increased insulin adherence. However, the SSM will also increase overall Medicare spending for insulin and overall medical costs, which may impact future premiums and benefits. Our findings can inform policy makers about the potential impact of the new Medicare SSM.

IMPORTANCE: In the US, cannabis use has nearly doubled during the past decade, in part because states have implemented recreational cannabis laws (RCLs). However, it is unclear how legalization of adult-use cannabis may affect alcohol consumption.

OBJECTIVE: To estimate the association between implementation of state RCLs and alcohol use among adults in the US.

DESIGN, SETTINGS, AND PARTICIPANTS: This was a cross-sectional study of 4.2 million individuals who responded to the Behavioral Risk Factor Surveillance System in 2010 to 2019. A difference-in-differences approach with demographic and policy controls was used to estimate the association between RCLs and alcohol use, overall and by age, sex, race and ethnicity, and educational level. Data analyses were performed from June 2021 to March 2022.

EXPOSURES: States with RCLs, as reported by the RAND-University of Southern California Schaeffer Opioid Policy Tools and Information Center.

MAIN OUTCOMES AND MEASURES: Past-month alcohol use, binge drinking, and heavy drinking.

RESULTS: Of 4.2 million respondents (median age group, 50-64 years; 2 476 984 [51.7%] women; 2 978 467 [58.3%] non-Hispanic White individuals) in 2010 through 2019, 321 921 individuals lived in state-years with recreational cannabis laws. Recreational cannabis laws were associated with a 0.9 percentage point (95% CI, 0.1-1.7; P = .02) increase in any alcohol drinking but were not significantly associated with binge or heavy drinking. Increases in any alcohol use were primarily among younger adults (18-24 years) and men, as well as among non-Hispanic White respondents and those without any college education. A 1.4 percentage point increase (95% CI, 0.4-2.3; P = .006) in binge drinking was also observed among men, although this association diminished over time.

CONCLUSIONS AND RELEVANCE: This cross-sectional study and difference-in-differences analysis found that recreational cannabis laws in the US may be associated with increased alcohol use, primarily among younger adults and men.

This cross-sectional study investigates trends in out-of-pocket costs for unmixed and novel glucagon formulations among patients with Medicare Advantage and commercial insurance from 2010 to 2020.

IMPORTANCE: In the US, cannabis use has nearly doubled during the past decade, in part because states have implemented recreational cannabis laws (RCLs). However, it is unclear how legalization of adult-use cannabis may affect alcohol consumption.

OBJECTIVE: To estimate the association between implementation of state RCLs and alcohol use among adults in the US.

DESIGN, SETTINGS, AND PARTICIPANTS: This was a cross-sectional study of 4.2 million individuals who responded to the Behavioral Risk Factor Surveillance System in 2010 to 2019. A difference-in-differences approach with demographic and policy controls was used to estimate the association between RCLs and alcohol use, overall and by age, sex, race and ethnicity, and educational level. Data analyses were performed from June 2021 to March 2022.

EXPOSURES: States with RCLs, as reported by the RAND-University of Southern California Schaeffer Opioid Policy Tools and Information Center.

MAIN OUTCOMES AND MEASURES: Past-month alcohol use, binge drinking, and heavy drinking.

RESULTS: Of 4.2 million respondents (median age group, 50-64 years; 2 476 984 [51.7%] women; 2 978 467 [58.3%] non-Hispanic White individuals) in 2010 through 2019, 321 921 individuals lived in state-years with recreational cannabis laws. Recreational cannabis laws were associated with a 0.9 percentage point (95% CI, 0.1-1.7; P = .02) increase in any alcohol drinking but were not significantly associated with binge or heavy drinking. Increases in any alcohol use were primarily among younger adults (18-24 years) and men, as well as among non-Hispanic White respondents and those without any college education. A 1.4 percentage point increase (95% CI, 0.4-2.3; P = .006) in binge drinking was also observed among men, although this association diminished over time.

CONCLUSIONS AND RELEVANCE: This cross-sectional study and difference-in-differences analysis found that recreational cannabis laws in the US may be associated with increased alcohol use, primarily among younger adults and men.