Publications

BACKGROUND: There is systemic undercoding of medical comorbidities within administrative claims in the Department of Veterans Affairs (VA). This leads to bias when applying claims-based risk adjustment indices to compare outcomes between VA and non-VA settings. Our objective was to compare the accuracy of a medication-based risk adjustment index (RxRisk-VM) to diagnostic claims-based indices for predicting mortality.

METHODS: We modified the RxRisk-V index (RxRisk-VM) by incorporating VA and Medicare pharmacy and durable medical equipment claims in Veterans dually-enrolled in VA and Medicare in 2012. Using the concordance (C) statistic, we compared its accuracy in predicting 1 and 3-year all-cause mortality to the following models: demographics only, demographics plus prescription count, or demographics plus a diagnostic claims-based risk index (e.g., Charlson, Elixhauser, or Gagne). We also compared models containing demographics, RxRisk-VM, and a claims-based index.

RESULTS: In our cohort of 271,184 dually-enrolled Veterans (mean age = 70.5 years, 96.1% male, 81.7% non-Hispanic white), RxRisk-VM (C = 0.773) exhibited greater accuracy in predicting 1-year mortality than demographics only (C = 0.716) or prescription counts (C = 0.744), but was less accurate than the Charlson (C = 0.794), Elixhauser (C = 0.80), or Gagne (C = 0.810) indices (all P < 0.001). Combining RxRisk-VM with claims-based indices enhanced its accuracy over each index alone (all models C ≥ 0.81). Relative model performance was similar for 3-year mortality.

CONCLUSIONS: The RxRisk-VM index exhibited a high level of, but slightly less, accuracy in predicting mortality in comparison to claims-based risk indices.

IMPLICATIONS: Its application may enhance the accuracy of studies examining VA and non-VA care and enable risk adjustment when diagnostic claims are not available or biased.

LEVEL OF EVIDENCE: Level 3.

BACKGROUND: In the United States, there is well-documented regional variation in prescription drug spending. However, the specific role of physician adoption of brand name drugs on the variation in patient-level prescription drug spending is still being investigated across a multitude of drug classes. Our study aims to add to the literature by determining the association between physician adoption of a first-in-class anti-diabetic (AD) drug, sitagliptin, and AD drug spending in the Medicare and Medicaid populations in Pennsylvania.

METHODS: We obtained physician-level data from QuintilesIMS Xponent™ database for Pennsylvania and constructed county-level measures of time to adoption and share of physicians adopting sitagliptin in its first year post-introduction. We additionally measured total AD drug spending for all Medicare fee-for-service and Part D enrollees (N = 125,264) and all Medicaid (N = 50,836) enrollees with type II diabetes in Pennsylvania for 2011. Finite mixture model regression, adjusting for patient socio-demographic/clinical characteristics, was used to examine the association between physician adoption of sitagliptin and AD drug spending.

RESULTS: Physician adoption of sitagliptin varied from 44 to 99% across the state's 67 counties. Average per capita AD spending was $1340 (SD $1764) in Medicare and $1291 (SD $1881) in Medicaid. A 10% increase in the share of physicians adopting sitagliptin in a county was associated with a 3.5% (95% CI: 2.0-4.9) and 5.3% (95% CI: 0.3-10.3) increase in drug spending for the Medicare and Medicaid populations, respectively.

CONCLUSIONS: In a medication market with many choices, county-level adoption of sitagliptin was positively associated with AD spending in Medicare and Medicaid, two programs with different approaches to formulary management.

INTRODUCTION: Opioid prescribing safety among individuals with cancer is poorly understood. This study estimates the prevalence of Pharmacy Quality Alliance opioid measures among individuals with cancer undergoing or not undergoing active treatment versus those without cancer.

METHODS: Pennsylvania Medicaid data (2016) were analyzed in 2018 to identify adults aged 18-64 years with and without cancer diagnoses who had 2 or more opioid prescriptions. Active cancer treatment, defined as having chemotherapy, radiotherapy, cancer surgery, or hospitalization with a primary diagnosis of cancer, was evaluated from October 2015 to December 2016 allowing a ≥3-month look-back period for cancer diagnoses observed in the first quarter of 2016. Opioid dosages (>120 morphine milligram equivalents for ≥90 consecutive days), multiple providers (4 or more prescribers and 4 or more pharmacies), and opioid and benzodiazepines overlapping ≥30 days were evaluated.

RESULTS: The sample with opioid prescriptions included 111,491 enrollees without cancer diagnoses and 12,819 with cancer, 58.8% of whom were not in active cancer treatment. Among enrollees undergoing cancer treatment, with cancer but not in active treatment, and without cancer, the prevalence of high morphine milligram equivalents was 7.1%, 6.0%, and 4.7% (p<0.001), respectively. The corresponding prevalence of multiple providers was 6.7%, 4.1%, and 3.4% (p<0.001). Concurrent opioid and benzodiazepine prescriptions occurred in 28.6%, 30.5%, and 26.8% (p<0.001), respectively.

CONCLUSIONS: Individuals with cancer, regardless of treatment status, had higher-risk opioid use based on Pharmacy Quality Alliance measures versus those without cancer. Their systematic exclusion from opioid quality surveillance could create missed opportunities to identify patients at high risk of adverse opioid-related outcomes.

BACKGROUND: The continued escalation of opioid use disorder (OUD) calls for heightened vigilance to implement evidence-based care across the US. Rural care providers and patients have limited resources, and a number of barriers exist that can impede necessary OUD treatment services. This paper reports the design and protocol of an implementation study seeking to advance availability of medication assisted treatment (MAT) for OUD in rural Pennsylvania counties for patients insured by Medicaid in primary care settings.

METHODS: This project was a hybrid implementation study. Within a chronic care model paradigm, we employed the Framework for Systems Transformation to implement the American Society for Addiction Medicine care model for the use of medications in the treatment of OUD. In partnership with state leadership, Medicaid managed care organizations, local care management professionals, the Universities of Pittsburgh and Utah, primary care providers (PCP), and patients; the project team worked within 23 rural Pennsylvania counties to engage, recruit, train, and collaborate to implement the OUD service model in PCP practices from 2016 to 2019. Formative measures included practice-level metrics to monitor project implementation, and outcome measures involved employing Medicaid claims and encounter data to assess changes in provider/patient-level OUD-related metrics, such as MAT provider supply, prevalence of OUD, and MAT utilization. Descriptive statistics and repeated measures regression analyses were used to assess changes across the study period.

DISCUSSION: There is an urgent need in the US to expand access to high quality, evidence-based OUD treatment-particularly in rural areas where capacity is limited for service delivery in order to improve patient health and protect lives. Importantly, this project leverages multiple partners to implement a theory- and practice-driven model of care for OUD. Results of this study will provide needed evidence in the field for appropriate methods for implementing MAT among a large number of rural primary care providers.

Potable water can be a source of transmission for legionellosis and nontuberculous mycobacterium (NTM) infections and diseases. Legionellosis is caused largely by Legionella pneumophila, specifically serogroup 1 (Sg1). Mycobacterium avium, Mycobacterium intracellulare, and Mycobacterium abscessus are three leading species associated with pulmonary NTM disease. The estimated rates of these diseases are increasing in the United States, and the cost of treatment is high. Therefore, a national assessment of water disinfection efficacy for these pathogens was needed. The disinfectant type and total chlorine residual (TClR) were investigated to understand their influence on the detection and concentrations of the five pathogens in potable water. Samples (n = 358) were collected from point-of-use taps (cold or hot) from locations across the United States served by public water utilities that disinfected with chlorine or chloramine. The bacteria were detected and quantified using specific primer and probe quantitative-PCR (qPCR) methods. The total chlorine residual was measured spectrophotometrically. Chlorine was the more potent disinfectant for controlling the three mycobacterial species. Chloramine was effective at controlling L. pneumophila and Sg1. Plotting the TClR associated with positive microbial detection showed that an upward TClR adjustment could reduce the bacterial count in chlorinated water but was not as effective for chloramine. Each species of bacteria responded differently to the disinfection type, concentration, and temperature. There was no unifying condition among the water characteristics studied that achieved microbial control for all. This information will help guide disinfectant decisions aimed at reducing occurrences of these pathogens at consumer taps and as related to the disinfectant type and TClR.IMPORTANCE The primary purpose of tap water disinfection is to control the presence of microbes. This study evaluated the role of disinfectant choice on the presence at the tap of L. pneumophila, its Sg1 serogroup, and three species of mycobacteria in tap water samples collected at points of human exposure at locations across the United States. The study demonstrates that microbial survival varies based on the microbial species, disinfectant, and TClR.

PURPOSE: Little is known about the longitudinal patterns of buprenorphine adherence among pregnant women with opioid use disorder, especially when late initiation, nonadherence, or early discontinuation of buprenorphine during pregnancy may increase the risk of adverse outcomes. We aimed to identify distinct trajectories of buprenorphine use during pregnancy, and factors associated with these trajectories in Medicaid-enrolled pregnant women.

METHODS: A retrospective cohort study included 2361 Pennsylvania Medicaid enrollees aged 15 to 46 having buprenorphine therapy during pregnancy and a live birth between 2008 and 2015. We used group-based trajectory models to identify buprenorphine use patterns in the 40 weeks prior to delivery and 12 weeks postdelivery. Multivariable multinomial logistic regression models were used to identify factors associated with specific trajectories.

RESULTS: Six distinct trajectories were identified. Four groups initiated buprenorphine during the first trimester of the pregnancy (early initiators): 31.6% with persistently high adherence, 15.1% with moderate-to-high adherence, 10.5% with declining adherence, and 16.7% with early discontinuation. Two groups did not initiate buprenorphine until midsecond or third trimester (late initiators): 13.5% had moderate-to-high adherence and 12.6% had low-to-moderate adherence. Factors significantly associated with late initiation and discontinuation were younger age, non-white race, residents of rural counties, fewer outpatient visits, more frequent emergency department visits and hospitalizations, and lower buprenorphine daily dose.

CONCLUSIONS: Six buprenorphine treatment trajectories during pregnancy were identified in this population-based Medicaid cohort, with 25% of women initiating buprenorphine late during pregnancy. Understanding trajectories of buprenorphine use and factors associated with discontinuation/nonadherence may guide integration of behavioral treatment with obstetrical/gynecological care to improve buprenorphine treatment during pregnancy.

BACKGROUND: Patterns of inpatient opioid use and their associations with postdischarge opioid use are poorly understood.

OBJECTIVE: To measure patterns in timing, duration, and setting of opioid administration in opioid-naive hospitalized patients and to examine associations with postdischarge use.

DESIGN: Retrospective cohort study using electronic health record data from 2010 to 2014.

SETTING: 12 community and academic hospitals in Pennsylvania.

PATIENTS: 148 068 opioid-naive patients (191 249 admissions) with at least 1 outpatient encounter within 12 months before and after admission.

MEASUREMENTS: Number of days and patterns of inpatient opioid use; any outpatient use (self-report and/or prescription orders) 90 and 365 days after discharge.

RESULTS: Opioids were administered in 48% of admissions. Patients were given opioids for a mean of 67.9% (SD, 25.0%) of their stay. Location of administration of first opioid on admission, timing of last opioid before discharge, and receipt of nonopioid analgesics varied substantially. After adjustment for potential confounders, 5.9% of inpatients receiving opioids had outpatient use at 90 days compared with 3.0% of those without inpatient use (difference, 3.0 percentage points [95% CI, 2.8 to 3.2 percentage points]). Opioid use at 90 days was higher in inpatients receiving opioids less than 12 hours before discharge than in those with at least 24 opioid-free hours before discharge (7.5% vs. 3.9%; difference, 3.6 percentage points [CI, 3.3 to 3.9 percentage points]). Differences based on proportion of the stay with opioid use were modest (opioid use at 90 days was 6.4% and 5.4%, respectively, for patients with opioid use for ≥75% vs. ≤25% of their stay; difference, 1.0 percentage point [CI, 0.4 to 1.5 percentage points]). Associations were similar for opioid use 365 days after discharge.

LIMITATION: Potential unmeasured confounders related to opioid use.

CONCLUSION: This study found high rates of opioid administration to opioid-naive inpatients and associations between specific patterns of inpatient use and risk for long-term use after discharge.

PRIMARY FUNDING SOURCE: UPMC Health System and University of Pittsburgh.

INTRODUCTION: Identifying Alzheimer's disease (AD) pharmacologic treatment options that effectively reduce the risk of mortality and hospitalization in real-world settings is critical.

METHODS: We compared donepezil, galantamine, memantine, oral rivastigmine, and transdermal rivastigmine with regard to all-cause mortality and all-cause hospitalization risk among fee-for-service Medicare beneficiaries with AD (aged ≥ 65 years) using a retrospective cohort study design. Our primary analysis was based on intention to treat (ITT), but we also present as-treated analysis.

RESULTS: In our final study sample (N = 21,558), significant difference in survival among index AD medication groups were observed with donepezil being associated with better survival than memantine, and oral and transdermal forms of rivastigmine for both ITT and as-treated analysis. Difference in hazards of all-cause hospitalization among index AD medication groups was observed in ITT analysis but not in as-treated analysis.

DISCUSSION: Significant differences exist in terms of mortality and hospitalization risk with different AD medication initiation in real-world setting.

IMPORTANCE: Current approaches to identifying individuals at high risk for opioid overdose target many patients who are not truly at high risk.

OBJECTIVE: To develop and validate a machine-learning algorithm to predict opioid overdose risk among Medicare beneficiaries with at least 1 opioid prescription.

DESIGN, SETTING, AND PARTICIPANTS: A prognostic study was conducted between September 1, 2017, and December 31, 2018. Participants (n = 560 057) included fee-for-service Medicare beneficiaries without cancer who filled 1 or more opioid prescriptions from January 1, 2011, to December 31, 2015. Beneficiaries were randomly and equally divided into training, testing, and validation samples.

EXPOSURES: Potential predictors (n = 268), including sociodemographics, health status, patterns of opioid use, and practitioner-level and regional-level factors, were measured in 3-month windows, starting 3 months before initiating opioids until loss of follow-up or the end of observation.

MAIN OUTCOMES AND MEASURES: Opioid overdose episodes from inpatient and emergency department claims were identified. Multivariate logistic regression (MLR), least absolute shrinkage and selection operator-type regression (LASSO), random forest (RF), gradient boosting machine (GBM), and deep neural network (DNN) were applied to predict overdose risk in the subsequent 3 months after initiation of treatment with prescription opioids. Prediction performance was assessed using the C statistic and other metrics (eg, sensitivity, specificity, and number needed to evaluate [NNE] to identify one overdose). The Youden index was used to identify the optimized threshold of predicted score that balanced sensitivity and specificity.

RESULTS: Beneficiaries in the training (n = 186 686), testing (n = 186 685), and validation (n = 186 686) samples had similar characteristics (mean [SD] age of 68.0 [14.5] years, and approximately 63% were female, 82% were white, 35% had disabilities, 41% were dual eligible, and 0.60% had at least 1 overdose episode). In the validation sample, the DNN (C statistic = 0.91; 95% CI, 0.88-0.93) and GBM (C statistic = 0.90; 95% CI, 0.87-0.94) algorithms outperformed the LASSO (C statistic = 0.84; 95% CI, 0.80-0.89), RF (C statistic = 0.80; 95% CI, 0.75-0.84), and MLR (C statistic = 0.75; 95% CI, 0.69-0.80) methods for predicting opioid overdose. At the optimized sensitivity and specificity, DNN had a sensitivity of 92.3%, specificity of 75.7%, NNE of 542, positive predictive value of 0.18%, and negative predictive value of 99.9%. The DNN classified patients into low-risk (76.2% [142 180] of the cohort), medium-risk (18.6% [34 579] of the cohort), and high-risk (5.2% [9747] of the cohort) subgroups, with only 1 in 10 000 in the low-risk subgroup having an overdose episode. More than 90% of overdose episodes occurred in the high-risk and medium-risk subgroups, although positive predictive values were low, given the rare overdose outcome.

CONCLUSIONS AND RELEVANCE: Machine-learning algorithms appear to perform well for risk prediction and stratification of opioid overdose, especially in identifying low-risk subgroups that have minimal risk of overdose.

Increasing gabapentinoid use has raised concerns of misuse and abuse in the United States (US). Little is known about the characteristics of gabapentinoid use in general clinical practice over time. This cross-sectional study used data from the National Ambulatory Medical Care Survey. We examined the trends of patient and prescriber characteristics and the diagnoses associated with US ambulatory care visits involving gabapentinoids for adult visits from 2003 to 2016. Using multivariable logistic regression, we estimated the adjusted proportion of gabapentinoid-involved visits among all visits and tested for trend significance. Among the weighted estimate of 260.1 million gabapentinoid-involved visits (aged 18-64 years: 61.8%; female: 61.9%; white: 85.5%), the adjusted annual proportion of gabapentinoid-involved visits nearly quadrupled from 2003 to 2016 (9.1 to 34.9 per 1000 visits; Ptrend < 0.0001), driven mainly by gabapentin. Nearly half had concurrent use with opioids (32.9%) or benzodiazepines (15.3%). Primary care physicians (45.8%), neurologists (8.2%), surgeons (6.2%), and psychiatrists (4.8%) prescribed two-thirds of the gabapentinoids. Most (96.6%) of the gabapentinoid visits did not have an approved indication for gabapentinoids among the first three diagnoses. Among US ambulatory care visits from 2003 to 2016, gabapentinoid use increased substantially, commonly prescribed by primary care physicians.