5.2.1 Summary of Findings—Aim 1
Out of over 4,400 NC Medicaid beneficiaries who were enrolled in the NC MLIP during its first two years of operation, a total of 1,647 MLIP enrollees were included in Aim 1 analyses. This cohort was about 40 years of age, on average, and mostly female, White, and residents of metropolitan counties in NC. Nearly all subjects had a baseline diagnosis of chronic noncancer pain and there was high mental health and physical comorbid burden in the cohort. High opioid utilization was the primary cause of MLIP eligibility for these subjects. Nearly all subjects experienced an MLIP enrollment delay. On average, the Aim 1 cohort was not subject to the MLIP intervention until nearly six months after first demonstrating high risk controlled substance use, as defined by the NC MLIP eligibility criteria.
All findings from Aim 1 descriptive, bivariate, and GEE modeling analyses clearly indicated that enrollment in the NC MLIP induced subjects to engage in controlled substance circumvention behavior. The mean number of circumvented opioids and benzodiazepine prescriptions per-person-per-month tripled following MLIP enrollment. When adjusting for policy- and patient-level characteristics,
enrollment in the NC MLIP was associated with 3.6 times higher likelihood of engaging in circumvention in a given month. MLIP enrollment also had similar effects on increases in the extent of circumvention behavior within a given person-month.
The marked increase in circumvention behavior following MLIP enrollment substantially offset what appeared, from NC Medicaid’s perspective, to be a drastic reduction in controlled substance use attributed to the MLIP. Looking only at NC Medicaid claims data—which would be the only data source
available to NC MLIP administrators to evaluate program outcomes—Medicaid prescription claims for opioids and benzodiazepines dropped over 40% after subjects were enrolled in the MLIP, from 2.72 claims per-person-per-month to 1.39 claims. However, what NC Medicaid could not observe was that over half of this decrease in Medicaid claims for controlled substances was offset by an increase in controlled substance prescription fills obtained through circumvention.
Aside from MLIP enrollment, GEE models from Aim 1 revealed additional factors associated with increased controlled substance circumvention behavior. The likelihood and extent of circumvention increased roughly 25% during periods in which subjects had demonstrated high-risk controlled substance use qualifying them for the MLIP but had not yet been enrolled in the program. Additionally, subjects who qualified for the MLIP due to high utilization of benzodiazepines and pharmacies were more likely to circumvent. Numerous patient-level characteristics were also associated with circumvention. This behavior was more common among MLIP enrollees who were younger, living in areas with high supplies of dispensing pharmacies, had anxiety disorder diagnoses, and had a high physical comorbidity burden.
5.2.2 Summary of Findings—Aim 2
A total of 98,243 opioid analgesic prescription fills from 1,646 NC MLIP enrollees were included in Aim 2 analyses. Overall, 80% of the prescriptions were paid for using Medicaid coverage, while the remaining 20% were obtained through circumvention. Sixty-one percent of opioid prescriptions contained Schedule II products, 18% contained a long acting opioid formulation, and the mean prescribed daily opioid dose was 79 MME/day.
Short acting oral hydrocodone (Vicodin, Norco) and oxycodone (Percocet, Roxicodone) were by far the most common opioid products received by subjects through circumvention and Medicaid
coverage. There were few notable changes in opioid product prevalence before and after MLIP
enrollment to suggest obvious targeting of specific opioid products for circumvention. The prevalence of long acting oxycodone (Oxycontin) among circumvented prescriptions decreased by half after MLIP enrollment. However, this finding may have been an artifact of underlying changes in prescribing practices in which long acting oxymorphone (Opana ER) gained market share as an alternative to long
acting oxycodone.253 Of note, though, methadone was the most commonly circumvented long acting opioid product and in the top three of most commonly circumvented Schedule II opioid products. However, among Medicaid covered prescriptions, methadone was only the fifth most common long acting opioid and not even among the top five most common Schedule II products.
The risk profile of opioid prescriptions—characterized by the proportion of prescriptions written for Schedule II and long acting opioid products and the mean prescribed MME/day—was comparable between circumvented and Medicaid-paid opioid fills before MLIP enrollment. After MLIP enrollment, however, opioid prescriptions paid for using NC Medicaid coverage contained Schedule II opioids 30% more often, long acting opioids twice as often, and had nearly 50% higher average daily doses of opioids. Opioid prescriptions obtained through circumvention did not experience any increase in prescription-level risk measures. The proportion of long-acting opioid fills and average daily opioid dose actually decreased after MLIP enrollment in the context of circumvention behavior.
These trends were confirmed in GLM analysis adjusting for policy- and patient-level
characteristics. MLIP enrollment had no association with the likelihood of circumventing Schedule II opioid products or on the average MME/day of circumvented opioid prescriptions. The likelihood of circumventing a long acting opioid decreased 42% following MLIP enrollment. Opioid prescriptions paid by subjects using their Medicaid coverage were 30% more likely to contain a Schedule II opioid, twice as likely to be written for a long-acting opioid formulation, and had an additional 55 MME prescribed/day compared to Medicaid covered opioid prescriptions prior to MLIP enrollment.
Although MLIP enrollment was not associated with an increase in risk profile of opioid prescriptions obtained through circumvention, GLM models identified a handful of patient-level characteristics associated with increased likelihood of circumventing an opioid prescription with a high- risk attribute: male sex, residence in a border county, depression, and high physical comorbidity burden.
5.2.3 Summary of Findings—Aim 3
Nearly 140,000 adult NC Medicaid beneficiaries were included in Aim 3 analysis investigating the performance of potential MLIP eligibility criteria as they are currently applied in the NC MLIP. Half
of subjects were used to test opioid exposure measures, and the other half were used to validate optimal measures and thresholds identified in the testing phase. The primary outcome of interest was
unintentional opioid overdose, for which 376 subjects recorded the primary outcome. Secondary
outcomes were unintentional APAP overdose, an unintentional overdose of any drug, opioid use disorder diagnosis, and any substance use disorder diagnosis.
Based on survival ROC analysis, no single opioid exposure measure, or combination thereof, even reached “fair” discriminatory performance in predicting subjects who experienced a study outcome versus those that did not. Measures of number of unique pharmacies used to obtain opioids and average daily APAP dosage from opioid/APAP combination products failed as predictors and were excluded from further evaluation. Measures of number of opioid prescription fills and average MME/day over a 60-day period were on the high end of the “poor” performance with AUCs of 0.67 and 0.68 for the primary
outcome, respectively. These two measures were further validated using optimal thresholds of ≥5 opioid
prescription fills and ≥12 MME/day (with alternative high-performing cutpoints at ≥45 MME/day and
≥140 MME/day).
Overall, sensitivity was low for capturing subjects with the primary outcome event using the best
performing opioid exposure criteria identified in the testing phase. The ≥12 MME/day criterion was 59%
sensitive, while the alternative mean MME/day cutpoints and the ≥5 opioid fill measure were 16-33%
sensitive. Specificity of these measures was generally over 90%, except for the ≥12 MME/day criterion
which was 72% specific for the primary outcome. Despite low sensitivity, the opioid use criteria selected for validation were all highly predictive of the study outcomes in bivariate Cox proportional hazard modeling. Subjects flagged as high risk based on these measures had 4 to 7-times higher hazard of an unintentional opioid overdose compared to those classified as low-risk for that measure. Based on bivariate Kaplan-Meier survival models, an MLIP would need to enroll 124, 192, 129, and 63 subjects
based on thresholds of ≥5 opioid fills, ≥12 MME/day, ≥45 MME/day, and ≥140 MME/day over 60 days,
respectively, to have the opportunity to prevent one accidental opioid overdose. Improved criterion performance across all evaluative methods was largely dictated by a positive relationship with the number
of subjects flagged as high risk. The current NC MLIP eligibility criterion of ≥7 opioid fills performed demonstrably worse than the optimal opioid use measures selected in Aim 3.