An Evaluation of the Effect of the OxyContin Reformulation on Unintentional Fatal and Nonfatal Overdose (2024)

Study Population and Design

This study included the national Medicaid database and 2 national commercial claims databases [IBM MarketScan Commercial and Medicare Supplemental Claims and Encounters database (MarketScan) and the HealthCore Integrated Research Database (HIRD)]. The Medicaid database, Medicaid Analytic eXtract (MAX), covers all 50 states and Washington, DC through 2012. The Medicaid population was restricted to treatment episodes that were from fee-for-service (FFS) or comprehensive managed care (CMC) plans in which the combined state, year, and basis of eligibility group demonstrated research usability (ie, continuity and connectivity between data elements following criteria defined by Li et al).20 MarketScan collects data from employers and health plans. The HIRD includes health plan members insured through Anthem. Each database was restricted to populations that were linkable to the National Death Index (NDI), a central computerized index of death record information derived from state vital statistics data.21,22 For inclusion into this study, individuals were required to be aged 16 to 74 years (16 to 64 y in Medicaid) and have ≥3 months continuous health plan enrollment before eligible opioid dispensings.

The study design was a retrospective cohort study comparing rates of OD before and after the OxyContin reformulation in users of OxyContin and in contemporaneous users of comparator opioids. The study period ranged from July 1, 2008 to June 30, 2010 as the prereformulation period (preperiod), and January 1, 2011 to September 30, 2015 for the postreformulation period for the commercial databases and through December 31, 2012 for Medicaid (postperiod) (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CJP/A873). The main analyses excluded a transition period of July 1, 2010 to December 31, 2010 when both the original and reformulated versions of OxyContin were likely used. Analyses included all users (new and prevalent) and separately restricted to new users which allowed for the application of the new user design. We compared the 5-year postperiod to the 2-year preperiod for the main analyses in the commercial insurance databases, and a 2-year postperiod and 2-year preperiod in Medicaid because of the unavailability of more recent data (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CJP/A873).

Exposures

The primary exposure evaluated in this study was OxyContin. The primary comparator opioids were ER morphine, transdermal fentanyl, and methadone tablets/capsules, selected because of similarities with OxyContin in labeled indications, long marketing histories, and large, stable market shares throughout the study period. Secondary comparator opioids consisted of single-entity ER oxymorphone, incidence rates (IR) oxycodone tablets, and IR hydromorphone tablets. Evidence showed that these products were the most preferred alternatives for abuse among prescription opioids during the time of decreasing popularity of OxyContin after it was reformulated.12

The analytic datasets were constructed using treatment episodes, with time-dependent covariates evaluated at the beginning of each treatment episode. Individuals were considered exposed to OxyContin and comparator opioids beginning on the day of dispensing and continuing through the days’ supply of the drug plus an extension period of half of the days’ supply. A treatment episode ended if there was 1 or more days of discontinuity between dispensings, and was censored if an individual discontinued any of the opioids that defined the episode, initiated another study opioid, had an OD, died, terminated their health insurance coverage, or reached the end of the preperiod, transition period, or study period. Thus, if 2 study opioids were dispensed within days of each other, the treatment episode for the first study opioid was terminated on the day of the dispensing of the second study opioid, and a new treatment episode was created at the dispensing of the second study opioid. If an individual obtained a new dispensing of an opioid agent before exhausting the days’ supply of a prior dispensing of that agent, we assumed that medication was taken starting on the day of the pharmacy dispensing.

Exposure for each treatment episode was classified in subcategories including “any use” of the drug (ie, with or without concomitant opioid use) and “only use” of the drug (ie, without concomitant opioid use). An example of treatment episode construction and exposure groups is provided in Supplemental Figure 2 (Supplemental Digital Content 1, http://links.lww.com/CJP/A873). “Only use” treatment episodes represented approximately one-third of the “any use” person-time. For comparative analyses with the primary comparators, we excluded treatment episodes with concomitant use of 2 or more primary comparators, or OxyContin with a primary comparator.

New (ie, incident) use was assessed for each treatment episode and defined as having had no recorded dispensing of any opioid study drug in the 3 months before the start of the treatment episode. This new use treatment episode plus any adjacent, continuous treatment episodes for that study drug comprised an incident use episode. An individual could have multiple incident use treatment episodes in the study if more than 1 treatment episode met the incident use criteria.

Outcomes

The primary outcome was unintentional fatal or nonfatal unintentional OD. Opioid OD was defined using a previously validated algorithm that uses the diagnostic codes associated with services in US health insurance data23,24 for events resulting in health services (both fatal and nonfatal). Further identification of fatal OD used linkage to the US NDI. A patient was classified as having experienced an OD event if they had at least one OD related ICD-9 and ICD-10 code (provided in Supplemental Table 1, Supplemental Digital Content 1, http://links.lww.com/CJP/A873) in any position or setting. Events were restricted to those not classified as intentional (defined through codes for suicide and other correlated factors)23,24 and those that occurred during a treatment episode.

Validation of nonfatal (insurance claims-based) OD was conducted through medical record review for 159 cases in the HIRD. Standardized data collection techniques were used to abstract information from the medical records including medical history and treatment of the possible OD event. Results from this validation study were similar to the previous validation studies of Green et al23,24 Among the 159 code-defined ODs, 135 (positive predictive value=85%) were confirmed on chart review (47 intentional and 88 unintentional). The algorithm to detect intentionality had lower accuracy than the overall OD algorithm (positive predictive value=69%), but a high sensitivity (98%). This suggests that a sizeable subset of intentional ODs were classified as unintentional with the unintentional OD algorithm. Additional methods and results for the validation study are further described in our prior publication.25

Covariates

Demographic characteristics, clinical characteristics, and other comorbidities and conditions noted in Supplemental Table 2 (Supplemental Digital Content 1, http://links.lww.com/CJP/A873) were assessed in claims for each treatment episode during a 3-month or 6-month lookback period before each treatment episode, except certain demographic characteristics that were assessed on the index date. Evaluation at each treatment episode allowed for the ability to account for potential time-varying confounding.

Statistical Analyses

Poisson regression models, which model log-linear relationships for rates, were used to calculate IRs and all comparison measures for OxyContin and comparator opioid groups. Since only the numerator of a rate (the number of events) is random, the dependent variable for each treatment category was the logarithm of the number of events. The logarithm of the denominator (person-time) was introduced into the model as an offset. The regression models were carried out using repeated-measures Generalized Estimating Equations with sandwich variance estimators and with independent covariance matrices for repeated observations because of multiple treatment episodes per patient. Approximately 10% of patients with an OD had multiple, distinct ODs during follow-up of the study. A time-varying covariate term for “prior OD event” was introduced to account for the most substantial source of within-person correlation.

Incidence rate ratios (IRRs) compared IRs between OxyContin and comparator groups within the prereformulation and postreformulation time periods. For pre-post comparisons, analyses were conducted in 3 ways in each database: (1) unadjusted for covariates; (2) adjusted for possibly time-varying covariates; and (3) adjusted for possibly time-varying covariates as well as other baseline characteristics through propensity score weighting in a new user cohort.

Postreformulation versus prereformulation rate ratios (RRs) were calculated for OxyContin and each of the comparator opioid groups. To compare pre-post reformulation changes in OD rates (ie, the RRs) for OxyContin and for comparator opioids, this study utilized a difference-in-differences design [implemented here as ratios-of-rate-ratios (RoRs)]. RRs and RoRs were also adjusted for baseline demographic and clinical covariates provided in Supplemental Table 2 (Supplemental Digital Content 1, http://links.lww.com/CJP/A873) using the Poisson regression models with Generalized Estimating Equations.

Adjusted RRs and RoRs were determined in the new user cohort using propensity score weighting (through inverse probability of treatment weighting) to match the covariate distributions of the prereformulation and postreformulation cohorts. The aim was to weight the postreformulation group to match the covariate structure of the prereformulation group in regard to demographic and clinical characteristics, in addition to other comorbidities and conditions. For OxyContin and each of the comparator opioids propensity scores were derived using logistic regression to estimate the fitted probability that a given incident use treatment episode with a specified set of covariates was from the prereformulation period as opposed to the postreformulation period. Propensity scores outside the range of overlap between the 2 time periods (prereformulation and postreformulation) were trimmed.

Sensitivity analyses included examination of the main results stratified by incident versus prevalent opioid use, fatal versus nonfatal OD status, and intentional versus unintentional OD status. We also examined results restricting the postperiod to 3 years in the commercial databases, and stratifying results by CMC versus FFS status in the Medicaid database. Analyses were conducted by HealthCore, IBM, and STATinMED using SAS.

Results from the 2 commercial databases (HIRD and MarketScan) were additionally pooled using random-effects meta-analysis.26–28 The Medicaid population was not included in the pooling given a priori differences in this population (eg, lower social economic status).

Human Participants

Institutional Review Board (IRB) approval was obtained for each site before NDI linkage and for conducting the validation of the algorithm for OD in the HIRD. No individual-level data were shared with the study sponsor.

Study Population

This study included 94,445 Medicaid patients, 122,254 MarketScan patients, and 81,137 HIRD patients who were dispensed OxyContin, and 367,814 Medicaid patients, 181,240 MarketScan patients, and 110,619 HIRD patients dispensed at least 1 of the primary comparator opioids. Exposed person-time per patient in each of the databases ranged from 6 to 10 months (Table ​(Table1).1). The dose of the dispensed OxyContin was 40 mg or more in 54% of Medicaid patients, 20% of MarketScan patients, and 36% of HIRD patients. Much of the use of OxyContin and primary comparator was concomitant with other opioids such as IR oxycodone or other nonprimary or secondary opioids (Table ​(Table1).1). Patient characteristics were largely similar between OxyContin treatment episodes and primary comparator opioid treatment episodes, with a few notable differences including a higher proportion of females and patients with certain pain disorders (abdominal, chronic, and neuropathic) among primary comparator opioid treatment episodes compared with OxyContin treatment episodes. Each of these personal characteristics included in the analysis had modest or no association with subsequent OD risk (all IRRs <2 and >0.5). By contrast, having experienced a prior OD was strongly associated with a subsequent OD (IRRs>14 in each database; results available upon request).

Incidence of OD During OxyContin Use

The IR for OD during OxyContin treatment episodes varied by database between 0.8 (MarketScan) and 1.6 (Medicaid) per 1000 person-months and was lower than the IR during comparator opioid use, which varied between 0.9 and 2.9 per 1000 person-months (Table ​(Table22 and Fig. ​Fig.1).1). Over 80% of ODs were nonfatal for both OxyContin and comparator opioids. In each database, the IR for OD during any use OxyContin exposure time declined from prereformulation to postreformulation [Medicaid: postperiod vs. preperiod adjusted aRR=0.93, 95% confidence interval (CI)=0.83-1.04; MarketScan/HIRD: aRR=0.86, 95% CI=0.75-1.00; Supplemental Table 3, Supplemental Digital Content 1, http://links.lww.com/CJP/A873]. For the any use analysis, these comparisons permitted concurrent dispensing of Schedule II opioids (other than the comparators). The decline in the IR for OD over time was more pronounced during use of OxyContin when it was dispensed with no other concurrent Schedule II opioid (Medicaid: aRR=0.80, 95% CI=0.63-1.01; MarketScan/HIRD: aRR=0.57, 95% CI=0.42-0.77; Fig. ​Fig.22 and Supplemental Table 3, Supplemental Digital Content 1, http://links.lww.com/CJP/A873).

OxyContin Versus Primary Comparators

For “any use” exposures, the postreformulation versus prereformulation RRs of OD were similar for OxyContin and each of the primary comparators (Fig. ​(Fig.3).3). In Medicaid, the crude RoRs measuring the postreformulation versus prereformulation RRs for comparator versus OxyContin ranged from 0.89 to 1.04, with the adjusted ratio-of-rate-ratios (aRoRs) ranging from 0.90 to 1.05 (Fig. ​(Fig.3;3; aRoR>1 means that the pre-post decline was greater during use of OxyContin than during use of the comparator). In the pooled MarketScan/HIRD results, the RoRs ranged from 1.13 and 1.26 comparing the pre-post trend in each comparator to OxyContin, with the aRoRs ranging from 1.10 to 1.22 (Fig. ​(Fig.3).3). In the “only use” comparisons, which were restricted to person-time exposed only to OxyContin or only to a single primary comparator, there was evidence of trends in OD incidence in the prereformulation to postreformulation periods that were more favorable for OxyContin versus the primary comparators in the commercial databases (Medicaid: aRoRs ranged from 1.06 to 1.30 comparing primary comparators to OxyContin; MarketScan/HIRD: aRoRs ranged from 1.64 to 1.85 comparing the primary comparators to OxyContin; Fig. ​Fig.33).

OxyContin Versus Secondary Comparators

Results were largely similar to the primary comparator analysis for OxyContin versus the secondary comparators. There was a lack of an association when comparing any use, but there was evidence of lower risk of OD during OxyContin only use than comparator only use, particularly in the commercial databases (Table ​(Table3).3). There were differences between the 2 commercial databases. In the HIRD, there were relatively similar patterns in OD during any OxyContin use, while there was a stronger decline in the OD rate during OxyContin only use than during secondary comparator only use (Table ​(Table3).3). In MarketScan, there was evidence of a stronger decline for OxyContin in the OD rate over time than those observed in the HIRD for both analyses (any use and only use; Table ​Table3).3). The pooled aRoR results for any use of each of the 3 secondary comparators relative to any use of OxyContin ranged from 1.42 to 1.74. The pooled aRoRs for only use of each of the 3 secondary comparators relative to OxyContin only ranged from 1.43 to 2.91 (data available upon request).

Sensitivity Analyses

Results were largely similar in sensitivity analyses. Results were similar when stratified by prevalent versus incident use, by fatal versus nonfatal OD, by intentional versus unintentional OD, and when restricting the postperiod in commercial databases to 3 years (data available upon request). In the Medicaid database, results appeared modestly different by FFS/CMC status. A decline in the OD rate during any use of OxyContin in the preperiod to postperiod was observed in the FFS population (aRR=0.81, 95% CI=0.69-0.95), but no change was observed in the CMC population (aRR=1.10, 95% CI=0.92-1.31). The estimates for the aRoRs were also often slightly different by population, as the OD comparing any transdermal fentanyl to any OxyContin was 1.28 (95% CI=1.03-1.59) in the FFS population, and 0.84 (95% CI=0.66-1.07) in the CMC population.

SUPPLEMENTARY MATERIAL:

Study conception/design: D.C.B., K.H., L.B., G.B., R.G., H.Y., A.M.W. Analysis: L.L., R.G., G.B., H.Y. Interpretation of data: D.C.B., K.H., L.B., G.B., R.G., L.L., H.Y., A.M.W. Draft or revision of manuscript: D.C.B., K.H., L.B., G.B., R.G., L.L., H.Y., A.M.W.

This work was supported by Purdue Pharma L.P., Stamford, CT; HealthCore Inc., Wilmington, DE, STATinMED Research, Plano, TX, IBM Watson Health, Cambridge, MA, and World Health Information Science Consultants (WHISCON), Boston, MA received funding from Purdue Pharma L.P. to conduct this study and for the development of this manuscript. This underlying study was an FDA-mandated Post Market Requirement (PMR) study presented at the “Joint Meeting of the Drug Safety and Risk Management Advisory Committee (DSaRM) and the Anesthetic and Analgesic Drug Products Advisory Committee (AADPAC)” on September 10 to 11 2020. D.C.B. and K.H. are employees of HealthCore Inc. R.G., G.B., and L.L. were employees of HealthCore Inc. at the time of the study. L.B. is an employee of IBM Watson Health. H.Y. is an employee of New York City College of Technology-CUNY, Brooklyn, NY and collaborated as an external consultant to STATinMED Research for this study. A.M.W. is an employee of WHISCON. Purdue Pharma L.P. provided funding to HealthCore Inc., STATinMED Research, IBM Watson Health, and WHISCON to conduct this study and for the development of this manuscript. H.Y. declares no conflict of interest.

2. Butler SF, Black RA, Cassidy TA, et al.. Abuse risks and routes of administration of different prescription opioid compounds and formulations. Harm Reduct J. 2011;8:29. [PMC free article] [PubMed] [Google Scholar]

3. Hays LR. A profile of OxyContin addiction. J Addict Dis. 2004;23:1–9. [PubMed] [Google Scholar]

4. Katz N, Dart RC, Bailey E, et al.. Tampering with prescription opioids: nature and extent of the problem, health consequences, and solutions. Am J Drug Alcohol Abuse. 2011;37:205–217. [PubMed] [Google Scholar]

5. Cone EJ, Giordano J, Weingarten B. An iterative model for in vitro laboratory assessment of tamper deterrent formulations. Drug Alcohol Depend. 2013;131:100–105. [PubMed] [Google Scholar]

6. Harris SC, Perrino PJ, Smith I, et al.. Abuse potential, pharmaco*kinetics, pharmacodynamics, and safety of intranasally administered crushed oxycodone HCl abuse-deterrent controlled-release tablets in recreational opioid users. J Clin Pharmacol. 2014;54:468–477. [PMC free article] [PubMed] [Google Scholar]

8. FDA letter dated April 05, 2010, NDA 02272, NDA approval. Silver Springs, MD: Department of Health and Human Services, 2010.

9. Chilcoat HD, Coplan PM, Harikrishnan V, et al.. Decreased diversion by doctor-shopping for a reformulated extended release oxycodone product (OxyContin). Drug Alcohol Depend. 2016;165:221–228. [PubMed] [Google Scholar]

10. Cicero TJ, Ellis MS, Surratt HL. Effect of abuse-deterrent formulation of OxyContin. N Engl J Med. 2012;367:187–189. [PubMed] [Google Scholar]

11. Cicero TJ, Ellis MS, Kasper ZA. A tale of 2 ADFs: differences in the effectiveness of abuse-deterrent formulations of oxymorphone and oxycodone extended-release drugs. Pain. 2016;157:1232–1238. [PubMed] [Google Scholar]

12. Cicero TJ, Ellis MS. Abuse-deterrent formulations and the prescription opioid abuse epidemic in the United States: lessons learned from OxyContin. JAMA Psychiatry. 2015;72:424–430. [PubMed] [Google Scholar]

13. Coplan PM, Kale H, Sandstrom L, et al.. Changes in oxycodone and heroin exposures in the National Poison Data System after introduction of extended-release oxycodone with abuse-deterrent characteristics. Pharmacoepidemiol Drug Saf. 2013;22:1274–1282. [PMC free article] [PubMed] [Google Scholar]

14. Coplan PM, Chilcoat HD, Butler SF, et al.. The effect of an abuse-deterrent opioid formulation (OxyContin) on opioid abuse-related outcomes in the postmarketing setting. Clin Pharmacol Ther. 2016;100:275–286. [PMC free article] [PubMed] [Google Scholar]

15. Dart RC, Severtson SG, Bucher-Bartelson B. Trends in opioid analgesic abuse and mortality in the United States. N Engl J Med. 2015;372:1573–1574. [PubMed] [Google Scholar]

16. Havens JR, Leukefeld CG, DeVeaugh-Geiss AM, et al.. The impact of a reformulation of extended-release oxycodone designed to deter abuse in a sample of prescription opioid abusers. Drug Alcohol Depend. 2014;139:9–17. [PubMed] [Google Scholar]

17. Hwang CS, Chang HY, Alexander GC. Impact of abuse-deterrent OxyContin on prescription opioid utilization. Pharmacoepidemiol Drug Saf. 2015;24:197–204. [PubMed] [Google Scholar]

18. Larochelle MR, Zhang F, Ross-Degnan D, et al.. Rates of opioid dispensing and overdose after introduction of abuse-deterrent extended-release oxycodone and withdrawal of propoxyphene. JAMA Intern Med. 2015;175:978–987. [PubMed] [Google Scholar]

19. Sessler NE, Downing JM, Kale H, et al.. Reductions in reported deaths following the introduction of extended-release oxycodone (OxyContin) with an abuse-deterrent formulation. Pharmacoepidemiol Drug Saf. 2014;23:1238–1246. [PMC free article] [PubMed] [Google Scholar]

20. Li Y, Zhu Y, Chen C, et al.. Internal validation of Medicaid Analytic eXtract (MAX) data capture for comprehensive managed care plan enrollees from 2007 to 2010. Pharmacoepidemiol Drug Saf. 2017;27:1067–1076. [PubMed] [Google Scholar]

21. Skopp NA, Smolenski DJ, Schwesinger DA, et al.. Evaluation of a methodology to validate National Death Index retrieval results among a cohort of US service members. Ann Epidemiol. 2017;27:397–400. [PubMed] [Google Scholar]

22. Williams BC, Demitrack LB, Fries BE. The accuracy of the National Death Index when personal identifiers other than Social Security number are used. Am J Public Health. 1992;82:1145–1147. [PMC free article] [PubMed] [Google Scholar]

23. Green CA, Perrin NA, Janoff SL, et al.. Assessing the accuracy of opioid overdose and poisoning codes in diagnostic information from electronic health records, claims data, and death records. Pharmacoepidemiol Drug Saf. 2017;26:509–517. [PubMed] [Google Scholar]

24. Green CA, Perrin NA, Hazlehurst B, et al.. Identifying and classifying opioid-related overdoses: a validation study. Pharmacoepidemiol Drug Saf. 2019;8:1127–1137. [PMC free article] [PubMed] [Google Scholar]

25. Beachler DC, Zhou S, Gangemi K, et al.. Validation of an algorithm to identify opioid overdose among US commerically insured individuals. International Conference on Pharmcoepidemiology and Therapeutic risk Mangement Meeting (ICPE 2018). Prague, Czech Republic. 2018.

26. Borenstein M, Hedges LV, Higgins JP, et al.. A basic introduction to fixed-effect and random-effects models for meta-analysis. Res Synth Methods. 2010;1:97–111. [PubMed] [Google Scholar]

27. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7:177–188. [PubMed] [Google Scholar]

28. DerSimonian R, Laird N. Meta-analysis in clinical trials revisited. Contemp Clin Trials. 2015;45:139–145. [PMC free article] [PubMed] [Google Scholar]

29. Roland CL, Lake J, Oderda GM. Prevalence of prescription opioid misuse/abuse as determined by international classification of diseases codes: a systematic review. J Pain Palliat Care Pharmacother. 2016;30:258–268. [PubMed] [Google Scholar]

30. Compton WM, Jones CM, Baldwin GT. Relationship between nonmedical prescription-opioid use and heroin use. N Engl J Med. 2016;374:154–163. [PubMed] [Google Scholar]

31. Hser YI, Mooney LJ, Saxon AJ, et al.. High mortality among patients with opioid use disorder in a large healthcare system. J Addict Med. 2017;11:315–319. [PMC free article] [PubMed] [Google Scholar]

32. Mojtabai R, Amin-Esmaeili M, Nejat E, et al.. Misuse of prescribed opioids in the United States. Pharmacoepidemiol Drug Saf. 2019;28:345–353. [PubMed] [Google Scholar]

33. Hedegaard H, Minino AM, Warner M. Drug overdose deaths in the United States, 1999-2017. NCHS Data Brief. 2018;329:1–8. [PubMed] [Google Scholar]