Matching study design to prescribing intention: The prevalent new‐user design for studying abuse‐deterrent formulations of opioids

In drug studies, research designs requiring no prior exposure to certain drug classes may restrict important populations. Since abuse‐deterrent formulations (ADF) of opioids are routinely prescribed after other opioids, choice of study design, identification of appropriate comparators, and addressing confounding by “indication” are important considerations in ADF post‐marketing studies.

• We evaluated the implications study design choice for estimating post-market effectiveness of ADFs and examined patterns of initiation.
• More than two-thirds of patients had claims for opioids before ADF or non-ADF ER/LA initiation, and it was common for patients that initiated ADF opioids to have previously or concurrently been prescribed non-ADF ER/LA or immediate-release opioids.
• Traditional new user designs may not adequately capture how ADFs are most commonly utilized in practice compared to the prevalent new user design.

Plain Language Summary
Post-marketing drug studies evaluate safety and effectiveness of medications in real-world patient populations, but study designs requiring no prior exposure to certain drug classes may exclude key patients.Using insurance claims data (2006-2018) from a North Carolina private insurer [NC claims] and Merative MarketScan [MarketScan], we evaluated the implications study design choice when studying post-market effectiveness of abuse-deterrent formulations (ADFs) of extended-release and long-acting (ER/LA) opioids, classifying patients as incident (no opioid claims six months before cohort entry) or prevalent users.We also examined prescribing patterns around the time of ADF or non-ADF ER/LA initiation, and described patient characteristics between the two exposure groups.We identified 8415 (NC claims) and 147 978 (MarketScan) ADF, and 10 114 (NC claims) and 232 028 (MarketScan) non-ADF ER/LA initiators.Most had prior opioid exposure (ranging 64%-74%), and prevalent users were more likely than incident users to have immediate-release opioids also prescribed at ADF or non-ADF ER/LA initiation.Patients starting ADFs were more likely to have recent acute pain, chronic pain, or surgery compared to patients starting non-ADF ER/LAs.Our findings suggest that important considerations must be made when selecting patients for inclusion based on prior drug use and identifying appropriate comparators post-marketing studies of ADFs.

| INTRODUCTION
Beginning in 1999, the United States (US) experienced rapid increases in opioid-related mortality and diagnosed opioid use disorders (OUD) related to prescription opioid use, 1,2 particularly extended-release and long-acting (ER/LA) formulations of opioids. 3Abuse-deterrent formulations (ADFs) of ER opioids were introduced as a "safer" alternative to ER/LA opioids, with the goal of curbing OUD and opioid overdose in patient populations treated for pain. 4The first ADF (August 2010) was a reformulation of oxycodone hydrochloride controlled-release tablets (reformulated OxyContin ® , Purdue Pharma), followed by several other ADFs introduced in subsequent years.By reformulating these medications to make it harder for individuals to obtain immediate release of the active ingredient by crushing or dissolving, the goal was to deter use through non-intended routes of administration, for example, injecting or snorting. 4,5[8] Post-marketing drug studies provide vital information about the safety and effectiveness of medications in real-world patient populations. 6However, important considerations must be made when selecting patients for inclusion based on prior drug use and identifying appropriate comparators. 7,8In drug studies, research designs requiring no prior exposure to specific drug classes, that is, the active comparator new-user (ACNU) design, 9,10 may restrict research on important populations.For example, currently marketed ADFs are routinely used in patients with prior prescription opioid exposure. 11Many implementations of the ACNU design in opioid research require no observed exposure to any or certain types of opioids in a pre-specified timeframe before cohort entry 12 ; hence, people with incident ADF use in this type of study design may not be representative of the overall ADF patient population.The prevalent new-user (PNU) design 13,14 expands upon the traditional ACNU study, allowing for inclusion of patients previously prescribed a comparator treatment before starting the new treatment, and may better represent how the treatment is used in real-world practice. 15Further, when designing pharmacoepidemiologic studies and selecting the study population, it is important to identify appropriate comparators and address confounding by "indication" 16 observable in claims data. 7,8,17 this study, we evaluated the implications of study design choice for estimating post-market effectiveness of ADFs and examined patterns of ADF and non-ADF ER/LA initiation.Further, we assessed demographic and clinical characteristics of patients initiating ADFs compared to patients initiating non-ADF ER/LA opioids to evaluate whether non-ADF ER/LA opioids represent an ideal comparator for ADF opioids.

| Cohort selection & opioid analgesic treatment history
Adult (18-64 years) patients were eligible for inclusion into analyses if they received an outpatient prescription for an ADF or non-ADF ER/LA opioid after August 1, 2010, when ADF opioids entered the market, following ≥six months of continuous enrollment before the prescription (Figure 1, eFigures 2-4).The filled date of the first prescription claim defining each cohort is hereafter referred to as the index date.Because treatment histories could span the period before August 1, 2010, we excluded patients who received their index ADF or non-ADF ER/LA prescription before August 1, 2010.Additionally, we excluded patients with evidence of overlapping ADF and non-ADF ER/LA prescriptions at the index date, defined as overlapping at ≥seven days or the duration of the ADF prescription.Patients were required to have ≥six months of continuous enrollment before their index date to characterize opioid treatment history before cohort entry.
We next categorized patients as traditional ACNU (incident users) or prevalent new users (prevalent users).Incident users of ADF and non-ADF ER/LAs were identified as those with no prescription opioid claims of any type in the six-month washout period before the index date (Figure 1, eFigures 2-4).The PNU design allowed for non-ADF ER/LA or immediate-release (IR) opioid claims during the six months before ADF initiation, provided the patient's claim history also satisfied a six-month washout period with no opioid claims before the first non-ADF ER/LA or IR opioid claim.Similarly, prevalent users of non-ADF ER/LA opioids had evidence of IR or ADF opioid claims before the index date.We required a six-month washout period before the first opioid claim (between January 1, 2006 and the index date) in order to characterize each patient's opioid treatment history preceding their index ADF or non-ADF ER/LA prescription.
We further examined whether a patient had concurrent IR use at entry, defined as an overlap of the index prescription and an IR opioid that spanned ≥seven days of the treatment episode with an ADF (ADF cohort) or non-ADF ER/LA (ER/LA cohort).Among prevalent users, we examined whether patients had a direct switch (≤7-day gap) between their previous opioid treatment and the index treatment versus a delayed switch (>7-day gap).We also identified the types of opioid analgesics (ADF cohort: IR and/or non-ADF ER/LA; non-ADF ER/LA cohort: IR and/or ADF) a patient had been exposed to between the most recent washout period and their index date, and time since the first opioid in the treatment episode (categorized: 1-3, 4-6, 7-9, 10-12, 13-17, or 18+ months).

| Patient characteristics
Demographic and clinical characteristics were examined to identify potential confounding by "indication."Demographic characteristics (age and sex), and year of the index prescription date were included.
We also identified outpatient pharmaceutical claims for benzodiazepines, gabapentin, and selective serotonin reuptake inhibitors (SSRIs) ≤six months before the index date as indications of physical and mental health status.Pain-related conditions (not mutually exclusive) were identified as invasive surgeries (using Current Procedural Terminology (CPT) codes 20 ) or diagnosed acute pain, chronic pain, arthritis (rheumatoid or osteoarthritis) pain, back/neck pain, and neuropathic pain (using International Classification of Disease, 9th revision, Clinical Modification (ICD-9-CM) and 10th revision (ICD-10-CM)) ≤30 days before the index date to capture potential clinical indications for pain management with ER/LA opioids.Finally, the Elixhauser comorbidity index 20,21 was used to identify comorbid conditions (Table 1 lists Elixhauser conditions) ≤six months before the index date based on ICD-9-CM and ICD-10-CM codes.

| Statistical analyses
We compared sample sizes by study design and described opioid treatment histories before ADF or non-ADF ER/LA initiation by cohort, using counts and frequencies.We created heat maps to visualize opioid dispensing patterns six months before and one year after treatment initiation.Further, we contrasted patient characteristics between cohorts, examining counts, frequencies, and absolute standardized mean differences (SMD) for evidence of differences between cohorts, using a cutoff of 0.1 for meaningful SMD. 22ta management was completed in SAS version 9.4 (Cary, NC, USA), and analyses were conducted in SAS 9.4 and R version 3.6.0. 23
Among ADF prevalent users, 17.2% and 8.  2).Patients initiating ADFs were less likely to be female (42.3% vs. 48.3%,respectively) and there was a calendar trend in prescribing, with ADFs prescribed more after 2012.A recent history of invasive surgery (53.9% vs. 18.2%), acute pain (50.6% vs. 28.2%),or chronic pain (78.8% vs. 65.5%)≤30 days before initiation were more prevalent     A recent study examining the risk of opioid-related harm associated with use of ADF oxycodone found an increased risk of opioidrelated harms among patients exposed to ADF oxycodone compared to patients on non-ADF oxycodone. 12However, this study restricted the study population to incident users without any prior exposure to oxycodone within 12 months before the start of follow-up.As we have shown, a substantial proportion of ADF users could be excluded as a result of the choice of study design, and including prevalent new users would increase sample size and capture a broader range of patients and treatment trajectories representing ADF initiation.
A PNU study, 13,14 wherein patients are prescribed similar treatments (or potential comparators) before starting the new treatment, likely better captures how ADFs are prescribed in real-world treatment settings.The traditional ACNU design can address the question of "among individuals without a history of prior exposure, what is the risk of opioid-related harms related to newly initiating ADFs versus newly initiating non-ADF ER/LA opioids?" 9,10,26 When implementing the PNU design, we can instead examine questions such as "what are the opioid related harms associated with initiating ADFs compared to initiating, re-initiating, or continuing non-ADF ER/LA opioids?" 13,14 Particularly when the research question is related to effectiveness and safety of treatment switching, discontinuing, and deprescribing, the PNU design should be considered to increase the available pool of patients, improve concordance with real-world prescribing practices, and allow for inferences beyond those limited to patients without a history of exposure to either treatment (or similar treatments). 14In opioid-related studies, the PNU design may be especially applicable to situations where prior exposure is a labeled prerequisite, such as higher-dose ER/LA opioids and second-line therapies.This design, when thoughtfully implemented, can mitigate potential bias from confounding by "indication" and healthy-user bias through careful selection of comparators. 14To account for exposure history, comparators can be chosen based on similar prior cumulative exposure history, considering factors such as (1) time since initial exposure; (2) number of prescriptions; and (3) type of initiator (direct or delayed switch). 14,15wever, trajectories of exposure history may be highly nuanced, as further demonstrated in this analysis.Researchers should carefully consider substantive and methodological implications in specifying the causal estimand when implementing the PNU design.Care must be taken in identifying types of initiators and capturing distinctions in treatment histories, 14 including gaps in treatment and whether switching occurred from the comparator (non-ADF ER/LAs) or less comparable treatment (IR opioids) in order to improve exchangeability between groups.
Additional methodological complexities should be taken into account when implementing the PNU versus ACNU design, including implications on person-time, censoring, and selection bias in when selecting and following comparators.As discussed by Suissa et al, selection bias due to exclusion of outcomes allocated to time during exposure history assessment is a threat when implementing the PNU design, and complexities due to treatment switching during comparator follow-up require thoughtful decisions around censoring strategies. 13In a recent analysis, we examined the association between ADF initiation and opioid use disorder and opioid overdose in a longitudinal claims-based cohort, including prevalent new users. 27counting for exposure history and treatment switches, we found that initiating ADFs compared to initiating, re-initiating, or continuing non-ADF ER/LA opioids was not notably associated with lower risk of opioid use disorder and opioid overdose.
Implementing the PNU design also introduces additional complexities with regard to the confounding structure compared to the ACNU design, as outlined by Webster-Clark et al 14

Available data spanned 13
years from 2006 through 2018 and was obtained from two sources: a large private health insurance provider in North Carolina (NC claims) and a commercially insured population from the Merative (formerly IBM ® ) MarketScan ® Research Databases (MarketScan).The NC claims data source contained longitudinal demographic, outpatient, inpatient, and prescription claims data from individuals who received health and pharmaceutical coverage from a single private insurance provider from January 1, 2006 through September 30, 2018 18,19 (>1.4 million average member months of data per year).The MarketScan data source included longitudinal inpatient, outpatient, and prescription claims data from commercially insured individuals and their dependents.This nationally representative database includes more than 43.6 million individuals per year of data from January 1, 2006 through December 31, 2018.

F
4% had prior exposure to non-ADF ER/LA opioid analgesics in NC claims and MarketScan, respectively.In both study populations, concurrent IR opioid use ≥7 days was more common among prevalent users (NC claims [ADF: 53.4%, non-ADF ER/LA: 52.7%]; MarketScan [ADF: 50.6%, non-ADF ER/LA: 52.4%]) than incident users (NC claims [ADF: 36.8%,non-ADF ER/LA: 25.0%]; MarketScan [ADF: 34.5%, non-ADF ER/LA: 45.0%]).The majority of prevalent users had a direct switch from IR opioids in both datasets (NC claims [ADF: 60.4%, non-ADF ER/LA: 64.0%]; MarketScan [ADF: 58.8%, non-ADF ER/LA: 61.8%]).Heat maps of opioid prescribing patterns (Figures2, 3) show that patients with prior exposure to opioids before starting an ADF or non-ADF ER/LA continued to have more claims for prescription opioids in the one year after the index ADF or non-ADF ER/LA opioid than patients without prior opioid exposure.3.1 | Patient characteristics: NC claims 3.1.1| Incident users Average [SD] age was similar between patients initiating ADFs (46.8 [12.7] years) and those initiating non-ADF ER/LA opioids (46.2 [12.0]I G U R E 1 Study timelines of eligibility and exposure ascertainment for (A) traditional new users, and (B) prevalent new users.ADF, Abusedeterrent formulation; ER/LA, extended-release/long-acting; IR, immediate-release.years; Table among ADF initiators.Patients initiating ADFs were less likely to have had recent claims for benzodiazepines or gabapentin.When examining specific pain indications (acute or chronic), patients starting ADFs were more likely to have rheumatoid or osteoarthritis pain, but less likely to have back/neck pain or neuropathic pain.When examining Elixhauser comorbid conditions, history of metastatic cancer (2.5% vs. 4.4%) and solid tumor without metastasis (5.1% vs. 7.4%) were less prevalent among patients initiating ADFs compared to those initiating non-ADF ER/LA opioids within six months before initiation.Likewise, a history of substance use disorders (SUD) and depression were less prevalent in patients initiating ADFs than those initiating non-ADF ER/LAs.

F I G U R E 2
Patterns of pharmaceutical claims for opioids among patients in the NC claims data source, 2010-2018.For each group, 150 patients were randomly selected.(A) Traditional new users initiating non-ADF ER/LA, (B) Prevalent new users initiating non-ADF ER/LA, (C) Traditional new users initiating ADFs, (D) Prevalent new users initiating ADFs.ADF, abuse-deterrent formulation; ER/LA, extended-release/ long-acting; IR, immediate-release.F I G U R E 3 Patterns of pharmaceutical claims for opioids among patients in the MarketScan data source, 2010-2018.For each group, 150 patients were randomly selected.(A) Traditional new users initiating non-ADF ER/LA, (B) Prevalent new users initiating non-ADF ER/LA, (C) Traditional new users initiating ADFs, (D) Prevalent new users initiating ADFs.ADF, abuse-deterrent formulation; ER/LA, extended-release/ long-acting; IR, Immediate-release.
and Suissa et al,13 particularly related to duration and trajectory of exposure history and ensuring temporality is maintained when measuring confounders in relation to timing of treatment initiation, re-initiation, or continuation.Hence, beyond comparator selection, researchers should consider additional adjustment related to treatment characteristics.Interestingly, in both patient populations in this study, patients initiating ADFs were more likely to have a history of recent surgery, acute pain, or chronic pain, regardless of study design type.However, when examining comorbid conditions, patients initiating non-ADF ER/LA opioids (incident or prevalent) were more likely to have a history of cancer, and among patients classified as incident users, individuals initiating non-ADF ER/LA opioids were more likely to have a recent history of SUD, depression, or prescriptions for benzodiazepines.When investigating the impact of ADFs on opioidrelated harms, non-ADF ER/LA opioids are likely a more suitable comparator choice than IR opioids due to opioid dosage, duration of action, and likely duration of treatment.However, observed differences in patient characteristics in this study indicate potential confounding by "indication" in analyses examining the relationship between ADF use and opioid-related harms that would need to be accounted for using weighting or other methods for confounding control.28This study examines two large populations of patients initiating ADF and non-ADF ER/LA opioids over nine years during the prescription opioid phase of the substance use epidemic using both a national and state data source.Notably, both data sources produced largely similar findings about opioid prescribing history and patient characteristics.However, there are limitations to consider when interpreting these findings.The NC claims data source only includes data from a single insurance provider on privately insured patients <65 years old and the MarketScan data includes claims information from individuals with employer-sponsored insurance.Therefore, these findings may not generalize to all privately insured individuals in the US or to individuals covered by Medicare or Medicaid or those without health insurance.Additionally, due to the nature of pharmaceutical claims data, we could only analyze prescriptions that were filled by a pharmacy and paid for by insurance, not what was paid for out-of-pocket or what was actually consumed.Finally, key confounders of the relationship between initiating ADFs and opioid-related harms (e.g., mental health, substance use disorders) are likely to be undermeasured or unmeasured in claims data.

Table 3 )
. Patients initiating ADFs were somewhat less likely to be female Opioid treatment histories for patients prescribed a non-ADF ER/LA or ADF opioid between August 1, 2010 and September 30, 2018 identified using insurance claims data from a private insurer in North Carolina (N = 17 140) and MarketScan (N = 380 006).
to have back/neck pain or neuropathic pain.Also, recent claims for gabapentin were less prevalent in patients initiating ADFs than those starting non-ADF ER/LA opioids.When examining Elixhauser comorbid conditions, recent history of metastatic cancer (13.2% vs. 17.2%) and solid tumor without metastasis (18.6% vs. 23.7%)werelessprevalent among patients initiating ADFs compared to those starting non-ADF ER/LA opioids.T A B L E 1Abbreviations: ADF, abuse-deterrent formulation; ER/LA, extended-release/long-acting; IR, immediate-release.DIPRETE ET AL.
T A B L E 2 Baseline characteristics of traditional new users.
Comparing baseline characteristics of patients newly initiating an ADF or non-ADF ER/LA opioid between August 1, 2010 and September 30, 2018 in insurance claims data from North Carolina (N = 17 140) and MarketScan (N = 380 006).All results presented are N (%) unless otherwise noted.Abbreviations: ADF, abuse-deterrent formulation; ER/LA, extended-release/long-acting; IQR, interquartile range; IR, immediate-release; SMD, standardized mean difference; SSRI, selective serotonin reuptake inhibitor.Baseline characteristics of prevalent new users.In this large study of individuals initiating ADF and non-ADF ER/LA opioids in the US, we found that most individuals had prior opioid exposure in the six months before initiation and would be excluded in post-marketing study designs that required no prior opioid exposure.There are important considerations that must be made regarding comparator selection and adjusting for confounding by "indication" in ADF post-marketing studies.We found that individuals initiating ADFs (regardless of prior opioid exposure) were more likely to have a recent acute or chronic pain diagnosis or surgery and were less likely to have a cancer diagnosis than those initiating non-ADF ER/LA opioids.Future work will further explore implementation of the PNU design and consider nuances in ADF initiation such as immediate versus delayed switching by incorporating opioid treatment history to address opioid tolerance.Comparing baseline characteristics of patients initiating an ADF or non-ADF ER/LA opioid between August 1, 2010 and September 30, 2018 in insurance claims data from North Carolina (N = 17 140) and MarketScan (N = 380 006).All results presented are N (%) unless otherwise noted.Abbreviations: ADF, abuse-deterrent formulation; ER/LA, extended-release/long-acting; IQR, interquartile range; IR, immediate-release; SMD, standardized mean difference; SSRI, selective serotonin reuptake inhibitor.
T A B L E 3