Publications

OBJECTIVE: To develop and validate a machine-learning algorithm to improve prediction of incident OUD diagnosis among Medicare beneficiaries with ≥1 opioid prescriptions.

METHODS: This prognostic study included 361,527 fee-for-service Medicare beneficiaries, without cancer, filling ≥1 opioid prescriptions from 2011-2016. We randomly divided beneficiaries into training, testing, and validation samples. We measured 269 potential predictors including socio-demographics, health status, patterns of opioid use, and provider-level and regional-level factors in 3-month periods, starting from three months before initiating opioids until development of OUD, loss of follow-up or end of 2016. The primary outcome was a recorded OUD diagnosis or initiating methadone or buprenorphine for OUD as proxy of incident OUD. We applied elastic net, random forests, gradient boosting machine, and deep neural network to predict OUD in the subsequent three months. We assessed prediction performance using C-statistics and other metrics (e.g., number needed to evaluate to identify an individual with OUD [NNE]). Beneficiaries were stratified into subgroups by risk-score decile.

RESULTS: The training (n = 120,474), testing (n = 120,556), and validation (n = 120,497) samples had similar characteristics (age ≥65 years = 81.1%; female = 61.3%; white = 83.5%; with disability eligibility = 25.5%; 1.5% had incident OUD). In the validation sample, the four approaches had similar prediction performances (C-statistic ranged from 0.874 to 0.882); elastic net required the fewest predictors (n = 48). Using the elastic net algorithm, individuals in the top decile of risk (15.8% [n = 19,047] of validation cohort) had a positive predictive value of 0.96%, negative predictive value of 99.7%, and NNE of 104. Nearly 70% of individuals with incident OUD were in the top two deciles (n = 37,078), having highest incident OUD (36 to 301 per 10,000 beneficiaries). Individuals in the bottom eight deciles (n = 83,419) had minimal incident OUD (3 to 28 per 10,000).

CONCLUSIONS: Machine-learning algorithms improve risk prediction and risk stratification of incident OUD in Medicare beneficiaries.

INTRODUCTION: The objective of this study was to compare the risk of stroke in atrial fibrillation (AF) with adherent use of oral anticoagulation (OAC), non-adherent use, and non-use of OAC.

METHODS: Using 2013-2016 Medicare claims data, we identified patients newly diagnosed with AF in 2014-2015 and collected prescriptions filled for OAC in the 12 months after AF diagnosis (n = 39,272). We categorized participants each day into three time-dependent exposures: adherent use (≥ 80% of the previous 30 days covered with OAC), non-adherent use (0-80% covered with OAC), and non-use (0%). We constructed Cox proportional hazards models to estimate the association between time-dependent exposures and time to stroke, adjusting for demographics and clinical characteristics.

RESULTS: The sample included 39,272 patients. Study participants spent 35.0% of the follow-up period in the adherent use exposure category, 10.9% in the non-adherent category, and 54.0% in the non-use category. OAC adherent use [hazard ratio (HR) 0.62; 95% confidence interval (CI) 0.52-0.74] and non-adherent use (HR 0.74; 95% CI 0.57-0.95) were associated with lower hazards of stroke than non-use. Adherent use of DOAC (HR 0.54; 95% CI 0.42-0.69) and warfarin (HR 0.70; 95% CI 0.56-0.89) was associated with lower risk of stroke than non-use, but the risk of stroke did not statistically differ between DOAC and warfarin adherent use (HR 0.77; 95% CI 0.56-1.04).

DISCUSSION: Although adherence to OAC reduces stroke risk by nearly 40%, newly diagnosed AF patients in Medicare adhere to OAC on average only one third of the first year after AF diagnosis.

BACKGROUND: In 2018, the Department of Veterans Affairs (VA) issued Notice 2018-08 requiring facilities to complete "case reviews" for Veterans identified in the Stratification Tool for Opioid Risk Mitigation (STORM) dashboard as high risk for adverse outcomes among patients prescribed opioids. Half of the facilities were randomly assigned to a Notice version including additional oversight. We evaluated implementation strategies used, whether strategies differed by randomization arm, and which strategies were associated with case review completion rates.

METHODS: Facility points of contact completed a survey assessing their facility's use of 68 implementation strategies based on the Expert Recommendations for Implementing Change taxonomy. We collected respondent demographic information, facility-level characteristics, and case review completion rates (percentage of high-risk patients who received a case review). We used Kruskal-Wallis tests and negative binomial regression to assess strategy use and factors associated with case reviews.

RESULTS: Contacts at 89 of 140 facilities completed the survey (64%) and reported using a median of 23 (IQR 16-31) strategies. The median case review completion rate was 71% (IQR 48-95%). Neither the number or types of strategies nor completion rates differed by randomization arm. The most common strategies were using the STORM dashboard (97%), working with local opinion leaders (80%), and recruiting local partners (80%). Characteristics associated with case review completion rates included respondents being ≤ 35 years old (incidence rate ratio, IRR 1.35, 95% CI 1.09-1.67) and having < 5 years in their primary role (IRR 1.23; 95% CI 1.01-1.51), and facilities having more prior academic detailing around pain and opioid safety (IRR 1.40, 95% CI 1.12-1.75). Controlling for these characteristics, implementation strategies associated with higher completion rates included (1) monitoring and adjusting practices (adjusted IRR (AIRR) 1.40, 95% CI 1.11-1.77), (2) identifying adaptations while maintaining core components (AIRR 1.28, 95% CI 1.03-1.60), (3) conducting initial training (AIRR 1.16, 95% CI 1.02-1.50), and (4) regularly sharing lessons learned (AIRR 1.32, 95% CI 1.09-1.59).

CONCLUSIONS: In this national evaluation of strategies used to implement case reviews of patients at high risk of opioid-related adverse events, point of contact age and tenure in the current role, prior pain-related academic detailing at the facility, and four specific implementation strategies were associated with case review completion rates, while randomization to additional centralized oversight was not.

TRIAL REGISTRATION: This project is registered at the ISRCTN Registry with number ISRCTN16012111. The trial was first registered on May 3, 2017.

This cross-sectional study assesses drug coverage, formulary restrictions, median retail prices, and annual out-of-pocket costs associated with commonly used insulin products across Medicare Part D plans.

IMPORTANCE: Most studies that have examined drug prices have focused on list prices, without accounting for manufacturer rebates and other discounts, which have substantially increased in the last decade.

OBJECTIVE: To describe changes in list prices, net prices, and discounts for branded pharmaceutical products for which US sales are reported by publicly traded companies, and to determine the extent to which list price increases were offset by increases in discounts.

DESIGN, SETTING, AND PARTICIPANTS: Retrospective descriptive study using 2007-2018 pricing data from the investment firm SSR Health for branded products available before January 2007 with US sales reported by publicly traded companies (n = 602 drugs). Net prices were estimated by compiling company-reported sales for each product and number of units sold in the US.

EXPOSURES: Calendar year.

MAIN OUTCOMES AND MEASURES: Outcomes included list and net prices and discounts in Medicaid and other payers. List prices represent manufacturers' price to wholesalers or direct purchasers but do not account for discounts. Net prices represent revenue per unit of the product after all manufacturer concessions are accounted for (including rebates, coupon cards, and any other discount). Means of outcomes were calculated each year for the overall sample and 6 therapeutic classes, weighting each product by utilization and adjusting for inflation.

RESULTS: From 2007 to 2018, list prices increased by 159% (95% CI, 137%-181%), or 9.1% per year, while net prices increased by 60% (95% CI, 36%-84%), or 4.5% per year, with stable net prices between 2015 and 2018. Discounts increased from 40% to 76% in Medicaid and from 23% to 51% for other payers. Increases in discounts offset 62% of list price increases. There was large variability across classes. Multiple sclerosis treatments (n = 4) had the greatest increases in list (439%) and net (157%) prices. List prices of lipid-lowering agents (n = 11) increased by 278% and net prices by 95%. List prices of tumor necrosis factor inhibitors (n = 3) increased by 166% and net prices by 73%. List prices of insulins (n = 7) increased by 262%, and net prices by 51%. List prices of noninsulin antidiabetic agents (n = 10) increased by 165%, and net prices decreased by 1%. List price increases were lowest (59%) for antineoplastic agents (n = 44), but discounts only offset 41% of list price increases, leading to 35% increase in net prices.

CONCLUSIONS AND RELEVANCE: In this analysis of branded drugs in the US from 2007 to 2018, mean increases in list and net prices were substantial, although discounts offset an estimated 62% of list price increases with substantial variation across classes.

IMPORTANCE: Low-value care is associated with harm among patients and with wasteful health care spending but has not been well characterized in the Veterans Health Administration.

OBJECTIVES: To characterize the frequency of and variation in low-value diagnostic testing for 4 common conditions at Veterans Affairs medical centers (VAMCs) and to examine the correlation between receipt of low-value testing for each condition.

DESIGN, SETTING, AND PARTICIPANTS: This retrospective cohort study used Veterans Health Administration data from 127 VAMCs from fiscal years 2014 to 2015. Data were analyzed from April 2018 to March 2020.

EXPOSURES: Continuous enrollment in Veterans Health Administration during fiscal year 2015.

MAIN OUTCOMES AND MEASURES: Receipt of low-value testing for low back pain, headache, syncope, and sinusitis. For each condition, sensitive and specific criteria were used to evaluate the overall frequency and range of low-value testing, adjusting for sociodemographic and VAMC characteristics. VAMC-level variation was calculated using median adjusted odds ratios. The Pearson correlation coefficient was used to evaluate the degree of correlation between low-value testing for each condition at the VAMC level.

RESULTS: Among 1 022 987 veterans, the mean (SD) age was 60 (16) years, 1 008 336 (92.4%) were male, and 761 485 (69.8%) were non-Hispanic White. A total of 343 024 veterans (31.4%) were diagnosed with low back pain, 79 176 (7.3%) with headache, 23 776 (2.2%) with syncope, and 52 889 (4.8%) with sinusitis. With the sensitive criteria, overall and VAMC-level low-value testing frequency varied substantially across conditions: 4.6% (range, 2.7%-10.1%) for sinusitis, 12.8% (range, 8.6%-22.6%) for headache, 18.2% (range, 10.9%-24.6%) for low back pain, and 20.1% (range, 16.3%-27.7%) for syncope. With the specific criteria, the overall frequency of low-value testing across VAMCs was 2.4% (range, 1.3%-5.1%) for sinusitis, 8.6% (range, 6.2%-14.6%) for headache, 5.6% (range, 3.6%-7.7%) for low back pain, and 13.3% (range, 11.3%-16.8%) for syncope. The median adjusted odds ratio ranged from 1.21 for low back pain to 1.40 for sinusitis. At the VAMC level, low-value testing was most strongly correlated for syncope and headache (ρ = 0.56; P < .001) and low back pain and headache (ρ = 0.48; P < .001).

CONCLUSIONS AND RELEVANCE: In this cohort study, low-value diagnostic testing was common, varied substantially across VAMCs, and was correlated between veterans' receipt of different low-value tests at the VAMC level. The findings suggest a need to address low-value diagnostic testing, even in integrated health systems, with robust utilization management practices.

OBJECTIVES: Many older adults with limited life expectancy and/or advanced dementia (LLE/AD) are potentially overtreated for diabetes and may benefit from deintensification. Our aim was to examine the incidence and predictors of diabetes medication deintensification in older Veterans with LLE/AD who were potentially overtreated at admission to Veterans Affairs (VA) nursing homes (community living centers [CLCs]).

DESIGN: Retrospective cohort study using linked VA and Medicare clinical/administrative data and Minimum Data Set assessments.

SETTING: VA CLCs.

PARTICIPANTS: A total of 6960 Veterans with diabetes and LLE/AD admitted to VA CLCs in fiscal years 2009 to 2015 with hemoglobin (Hb)A1c measured within 90 days of admission.

MEASUREMENTS: We evaluated treatment deintensification (discontinuation or dose reduction for a consecutive 7-day period) among residents who were potentially overtreated (HbA1c ≤7.5% and receiving hypoglycemic medications). Competing risk models assessed 90-day cumulative incidence of deintensification.

RESULTS: More than 40% (n = 3056) of Veteran CLC residents with diabetes were potentially overtreated. The cumulative incidence of deintensification at 90 days was 45.5%. Higher baseline HbA1c values were associated with a lower likelihood of deintensification (e.g., HbA1c 7.0-7.5% vs <6.0%; adjusted risk ratio [aRR] = .57; 95% confidence interval [CI] = .50-.66). Compared with non-sulfonylurea oral agents (e.g., metformin), other treatment regimens were more likely to be deintensified (aRR = 1.31-1.88), except for basal insulin (aRR = .59; 95% CI = .52-.66). The only resident factor associated with increased likelihood of deintensification was documented end-of-life status (aRR = 1.12; 95% CI = 1.01-1.25). Admission from home/assisted living (aRR = .85; 95% CI = .75-.96), obesity (aRR = .88; 95% CI = .78-.99), and peripheral vascular disease (aRR = .90; 95% CI = .81-.99) were associated with decreased likelihood of deintensification.

CONCLUSION: Deintensification of treatment regimens occurred in less than one-half of potentially overtreated Veterans and was more strongly associated with low HbA1c values and use of medications with high risk for hypoglycemia, rather than other resident characteristics. J Am Geriatr Soc 68:736-745, 2020.

BACKGROUND/OBJECTIVES: Geriatric guidelines recommend against statin use in older adults with limited life expectancy (LLE) or advanced dementia (AD). This study examined resident and facility factors predicting statin discontinuation after nursing home (NH) admission in veterans with LLE/AD taking statins for secondary prevention.

DESIGN: Retrospective cohort study of Veterans Affairs (VA) bar code medication administration records, Minimum Data Set (MDS) assessments, and utilization records linked to Medicare claims.

SETTING: VA NHs, known as community living centers (CLCs).

PARTICIPANTS: Veterans aged 65 and older with coronary artery disease, stroke, or diabetes mellitus, type II, admitted in fiscal years 2009 to 2015, who met criteria for LLE/AD on their admission MDS and received statins in the week after admission (n = 13,110).

MEASUREMENTS: Residents were followed until statin discontinuation (ie, gap in statin use ≥14 days), death, or censoring due to discharge, day 91 of the stay, or end of the study period. Competing risk models assessed cumulative incidence and predictors of discontinuation, stratified by whether the resident had their end-of-life (EOL) status designated or used hospice at admission.

RESULTS: Overall cumulative incidence of statin discontinuation was 31% (95% confidence interval [CI] = 30%-32%) by day 91, and it was markedly higher in those with (52%; 95% CI = 50%-55%) vs without (25%; 95% CI = 24%-26%) EOL designation/hospice. In patients with EOL designation/hospice (n = 2,374), obesity, congestive heart failure, and admission from nonhospital settings predicted decreased likelihood of discontinuation; AD, dependency in activities of daily living, greater number of medications, and geographic region predicted increased likelihood of discontinuation. In patients without EOL designation/hospice (n = 10,736), older age and several specific markers of poor prognosis predicted greater discontinuation, whereas obesity/overweight predicted decreased discontinuation.

CONCLUSION: Most veterans with LLE/AD taking statins for secondary prevention do not discontinue statins following CLC admission. Designating residents as EOL status, hospice use, and individual clinical factors indicating poor prognosis may prompt deprescribing.