Association of a Risk Evaluation and Mitigation Strategy Program With Transmucosal Fentanyl Prescribing

Key Points Question Was the implementation of the Transmucosal Immediate-Release Fentanyl (TIRF)–Risk Evaluation and Mitigation Strategy (REMS) associated with changes in prescribing of TIRF medications? Findings In this cohort study using interrupted times series analysis, implementation of TIRF-REMS was associated with a temporary reduction in the rate of overall TIRF prescribing to Medicare Part D beneficiaries and with a sustained decrease in the percentage of TIRF prescribed to patients without known opioid tolerance. The TIRF-REMS program may have also been associated with a temporary decrease in the percentage of TIRF prescribed to patients without cancer. Meaning Mandatory, restrictive drug distribution programs, such as the TIRF-REMS, may be associated with changes in opioid prescribing, although the changes may be temporary.


eAppendix 1. Source and processing of prescribing data
We obtained line-item prescription claims data for all drugs classified as opioids from the Chronic Condition Warehouse using the group "65" identifier in the Master Drug Database classification system, which is one of the classification systems the Warehouse uses to categorize drug claims. The data included anonymized beneficiary identifiers, date of prescription dispensation, brand and generic drug names, dose, quantity dispensed, and prescriber NPI. We linked the pharmacy claims to limited beneficiary-level demographic and health data using these anonymized beneficiary identifiers.

eAppendix 2. Exclusions
For all analyses, we excluded 229,705 prescription claims (0.06% of all prescriptions) for injectable opioids, and 98 prescriptions claims that were missing the route.
For the time series analyses of overall TIRF prescribing and prescribing to patients without cancer, we also excluded prescriptions for sublingual fentanyl tablets marketed as Abstral, fentanyl nasal spray marketed as Lazanda, and fentanyl buccal film marketed as Onsolis (0.96%, 0.49%, and 0.24% of all TIRF prescriptions, respectively), as these drugs had separate REMS implemented prior to the class-wide TIRF implementation. We did, however, include Abstral and Onsolis prescriptions in the analyses of prescribing to patients without known opioid tolerance since these prescriptions could have affected patients' tolerance status, and also included them in the descriptive analyses to provide a complete overview of Part D TIRF prescribing.
For the by-brand analysis for each of the primary outcomes, the only brands that had prescriptions during each month of the study period to allow for brand-level analysis were Actiq and Fentora, and only Fentora had enough prescriptions to enable brand-level analysis on prescribing to patients without known opioid tolerance. Additionally, for the time series analysis of percentage of prescribing to patients without known opioid tolerance, we excluded the first 3 months of 2010, since we used a 90 day look-back period to establish patients' prescribing history and opioid tolerance.

eAppendix 3. Interrupted Time Series Models
We performed interrupted time series analyses using segmented ordinary least squares regression 1 with robust Newey-West errors to account for autocorrelation and heteroskedasticity. 2 This method of analysis calculates independent tests of the level (the intercept) and trend (slope) before and after an interruption, or intervention, and then evaluates for differences between the levels and slopes.
For single group analyses, the model is represented by the following figure and equation: Y t =β 0 +β 1 T +β 2 X t +β 3 X tt +ε t Y t is the aggregated outcome variable measured at each time-point t; T t is the time since the start of the study; X t is a dummy variable representing the intervention; X t T t is an interaction term; β 0 represents the intercept, or starting level of the outcome variable; β 1 is the slope of the outcome variable until the introduction of the intervention; β 2 represents the change in the level of the outcome that occurs in the period immediately following the introduction of the intervention; β 3 represents the difference between pre and post-intervention slopes of the outcome.
In this model, Z is a dummy variable denoting the treatment/control groups; ZT t , ZX t and ZX t T t are all interaction terms. In the figure below: β 0 to β 3 represent the control group; β 4 to β 7 represent the treatment group. Specifically: β 4 represents the difference in the level or intercept of the outcome variable between treatment and controls prior to the intervention, while β 5 represents the difference in the slope or trend of the outcome variable between treatment and controls prior to the intervention. β 6 indicates the difference between treatment and control groups in the level of the outcome variable immediately following introduction of the intervention, and β 7 represents the difference between treatment and control groups in the slope of the outcome variable after initiation of the intervention compared to pre-intervention (akin to a difference-in-differences of slopes). 2,4 eAppendix 4. Sensitivity analysis excluding buprenorphine As noted in the manuscript, we performed sensitivity analyses to evaluate for potential confounding by factors that may have affected all opioid prescribing, such as increasing national attention to prescription opioid harms. To do this, we performed multiple group time series analyses, using all-opioid prescriptions as a control group. We first included all opioids, including opioid drugs commonly used for medication assisted therapy (MAT) for opioid use disorders such as methadone and buprenorphine, as well as medications generally marketed for cough and cold treatment that contain opioids. Notably, Medicare Part D does not cover methadone for MAT, but does provide coverage for methadone when prescribed for pain, and there were more than 7 million filled prescriptions for methadone during our study period (eTable 2). Part D does cover buprenorphine-containing medications. Since it is possible that an increase in buprenorphine prescribing for MAT could have masked a decrease in all-opioid prescribing (thus confounding our use of all opioid prescriptions as a control), we repeated the above analyses while excluding buprenorphine-containing drugs. Results are as noted in eFigure 2a.

eAppendix 5. Model adjustment
We tested for autocorrelation up to 6 "lags", or time periods, in the error distribution using the Cumby-Huizinga test, and specified the models to ensure correct autocorrelation structures. Initial examination of the data showed seasonal variation, with increases in prescribing around the beginning of the year. We therefore included a variable denoting the months of December and January. We also included a variable denoting the number of days in each month of the study.

eAppendix 6. Interpretation of coefficients
The study results included both absolute number changes (e.g. monthly TIRF prescriptions per 100,000 Part D participants), as well as percentage point changes (e.g. monthly percentage of TIRF prescriptions for patients without cancer). For ease of interpretation, we calculated and report relative percent changes for all outcomes. We calculated relative percent changes by log transforming the dependent variable for all analyses, and then exponentiating the resulting coefficients.     . When compared to generic prescriptions, there were no significant differences in the pre-TIRF-REMS trend among Actiq prescriptions (0.49%, 95% CI, -1.14, 0.16; p=0.14), nor in the level change (7.92%, 95% CI, -19.9, 5.81; p=0.24) or trend change (0.38%, 95% CI, -1.2, 0.46; p=0.38) after TIRF-REMS implementation. For Fentora, the pre-TIRF-REMS trend did differ from the trend for generics (1.2%, 95% CI, 0.6, 1.8; p<0.001), but there were no significant differences from generics in the level change (3.4%,95% CI,17.8;p=0.61)  Fentora: Observed Fitted Generic: Observed Fitted eFigure 5. Post-hoc analysis of the three primary outcomes with and without Subsys-brand transmucosal immediate-release fentanyl (TIRF) prescriptions: overall rate of TIRF prescribing, the percentage of TIRF prescriptions for patients without cancer, and the percentage of TIRF prescriptions for patients without known tolerance. This analysis is intended for hypothesis generation only. Interpretation is limited by the fact that there is no way to separate prescriptions that were 'converted' from generic to Subsys (following the introduction of Subsys to the market) vs those that were newly 'induced' by the manufacturer of Subsys through marketing and promotion.