We focused our analyses on adult Medicaid enrollees diagnosed with OUD, and within that group, enrollees with at least one primary care visit, and enrollees who received MAT. We conducted a cross-sectional comparison between enrollees with OUD who had > 1 primary care visit versus enrollees with OUD and no primary care visits. We conducted logistic regression analyses to estimate the association between distance to MAT prescribers and the likelihood that enrollees with OUD initiated MAT and received continuity of pharmacotherapy.
We obtained Medicaid claims, encounter, and enrollment data for fee-for-service and managed care enrollees from 2014 to 2015 from the Pennsylvania Department of Human Services. Pennsylvania is the 7th largest Medicaid program by enrollment14 and the 4th largest by expenditure15 and mirrors national averages on health care utilization although it has higher overdose death rates. Pennsylvania has the 3rd largest rural population.16 Pennsylvania expanded Medicaid under the Affordable Care Act in 2015, which some studies have shown to improve access to OUD treatment.17
We included all full-benefit Medicaid-enrolled adults (age 18–64) who were not dually eligible for Medicare. We limited our sample to residents of 23 rural counties with OUD prevalence and/or opioid/heroin overdose rates above the national average. Enrollees were included if they had > 1 claim with a diagnosis of OUD (international classifications of diseases (ICD) version 9 and 10; see Appendix Table 4) in any diagnosis field during 2015. Diagnosis codes for OUD tend to have low sensitivity but high specificity,18,19 so we may underestimate the prevalence.
MAT Utilization and MAT Setting
As our focus was on MAT delivered by PCPs, our primary definition of MAT included buprenorphine and injectable or oral naltrexone but not methadone. We counted the number of fills or injections for buprenorphine (specifically formulations indicated for the treatment of OUD) or naltrexone per enrollee. We did, however, construct a variable for inclusion in descriptive analyses that measured the use of methadone, a potential substitute for buprenorphine or naltrexone, provided by methadone clinics based on procedure code. We categorized enrollees diagnosed with OUD in any of five service settings based on claim-level information: PCP visit, emergency department (ED) visit, acute care hospitalization, behavioral health visit (if the visit was covered by the enrollee’s behavioral health managed care organization), or other (if they did not fit into one of the previous categories).
Primary Care Utilization
We constructed three measures of contact between patients with OUD and primary care: the number of primary care visits per person-year; how frequently enrollees who were diagnosed with OUD during a PCP visit received any MAT, and how frequently these enrollees received MAT from a PCP. We also identified if the enrollee had any claims for behavioral health counseling using a set of procedure codes. See Appendix Tables 5 and 6.
Key Dependent Variables
In our multivariable analyses, we had two key outcomes of interest: the likelihood that enrollees with OUD had any use of MAT (either buprenorphine or naltrexone) and the continuity of pharmacotherapy, which was defined using National Quality Forum specifications. Specifically, we identified enrollees with OUD initiating treatment with buprenorphine, naltrexone, and methadone who received > 180 days of continuous pharmacotherapy with no more than a 7-day gap.20
Key Independent Variable—Distance to MAT Provider
To understand rural enrollees’ access to MAT, we measured two distances: That between enrollees with OUD and the nearest possible MAT prescriber, and the distance traveled by enrollees to their actual MAT prescribers. For nearest possible MAT prescribers to enrollees in our study sample, we included prescribers with > 1 Medicaid-paid prescription fill for buprenorphine or naltrexone. We then identified the prescriber nearest to the center of each zip code, the geographic identifier available to us for enrollees, and measured the driving distance to the provider’s office address using ArcGIS. We also examined distance to the nearest prescriber of buprenorphine or naltrexone to at least 10 Medicaid enrollees as some providers may prescribe to very few patients. To measure distance traveled to actual prescribers, we measured driving distance from the center of the enrollee’s zip code to the prescribing provider’s office address per prescription to account for prescriptions from different providers and providers that practice in multiple locations. We then took the mean of the distances traveled for the given enrollee. To be included in this measure, enrollees needed complete prescriber information (both on the claim and a valid address in the provider file) and continuous Medicaid enrollment without any gaps in coverage. We calculated the distribution of each distance measure by minimum, maximum, median, interquartile range, and 90th percentile.
We included demographic characteristics including age, gender, race, and type of Medicaid eligibility: disabled, newly eligible (i.e., eligible through Medicaid expansion), non-disabled, and other. We also included measures of relevant comorbidities including anxiety, mood disorders, and schizophrenia and other psychotic disorders.21 We grouped enrollees into one of three regions in Pennsylvania: Northeast, Northwest, and Southwest. We calculated the number of chronic conditions for each enrollee using a modified version of the Elixhauser Comorbidity Index which used both inpatient and outpatient claims to identify conditions.22 We included indicators for which type of provider prescribed the greatest number of days supply of MAT: a PCP or a non-PCP. We categorized enrollees with at least one fill of methadone into a separate category since these patients are likely systematically different and PCPs cannot prescribe methadone.
To test for differences between enrollees with OUD and > 1 primary care visit and enrollees with OUD and no primary care utilization, we used t tests and chi-square tests as appropriate. To understand the association between distance to MAT prescribers and MAT treatment patterns, we used two logistic regression models. First, we estimated the association between distance to the nearest MAT prescriber on the likelihood of enrollees with OUD receiving MAT, controlling for the covariates listed above (except PCP or non-PCP MAT prescriber type). In this model, distance was specified as a continuous variable in miles, and we only included providers that prescribed to 10 or more enrollees, as including all prescribers may have underestimated distances to an active prescriber. Second, we estimated the association between the mean distance traveled to actual MAT prescribers and the likelihood of continuity of pharmacotherapy. We tested the mean distance traveled as both a continuous and a categorical variable. As a categorical variable, we created two groups based on mean distances traveled greater or less than 45 miles, as Pennsylvania regulations for Medicaid managed care organizations require that 90% of enrollees outside of metropolitan areas be located within 45 miles of primary and specialty care providers.23 In this model, we included the same covariates used in the model described above as well as PCP vs. non-PCP prescriber type.
We conducted two sensitivity analyses. First, we restricted the sample to those enrollees whose mean distance traveled was less than 270 miles to determine the influence of outliers on our estimates. The odds ratios and statistical significance were comparable to the main model. Second, we specified the mean distance traveled as a continuous variable in the model. The direction of the association and statistical significance remained the same as the categorical specification. We present the latter for ease of interpretation and for policy relevance given the 45-mile managed care regulation.
All analyses were conducted using SAS version 9.4. This study was approved under an expedited review by the University of Pittsburgh Institutional Review Board.