Association of Opioid Use Disorder Treatment With Alcohol-Related Acute Events

Key Points Question Are opioid use disorder (OUD) medications associated with alcohol-related poisonings, falls, and injuries in persons with OUD and co-occurring alcohol use? Findings This case-control cohort study of 13 335 individuals with OUD from a large, nationally representative data set found that patients prescribed OUD medication had significant reductions in alcohol-related acute events. Meaning The use of OUD medication in opioid-dependent patients with co-occurring alcohol use may hold promise in reducing alcohol-related acute events, but more investigation is needed.


Introduction
Over the last decade, an unprecedented epidemic of drug-related poisonings has affected countries spanning North America and Europe. Recent estimates of unintentional poisoning deaths (19.1 deaths per 100 000) have eclipsed those of falls (11.4 deaths per 100 000) and motor vehicle traffic deaths (11.6 deaths per 100 000). 1 This is evidenced by an all-time high of 70 980 overdose deaths in the US in 2019. 2 Although the medical community has largely focused on the central role of opioid use disorder (OUD), polysubstance use has been underrecognized, even though most patients with substance use disorder use multiple substances, 3 and the presence of multiple substance use disorders is associated with worse treatment outcomes. [4][5][6] Alcohol is the most commonly misused substance, both as a standalone drug and in the context of polysubstance use. 5 More than 25% of patients with OUD exhibit problematic alcohol use, and, therefore, mitigation of risk for incident or recurrent alcohol use disorder (AUD) presents a salient problem in the treatment of OUD. 5,7-9 As central nervous system depressants, 10 alcohol and opioids can be lethal in combination, leading to increased overdoses and mortality among patients with OUD, 11,12 in addition to increased health care utilization. [13][14][15] Because problematic alcohol and opioid use are traditionally studied as independent conditions, limited evidence-based strategies exist for mitigation of alcohol-related risk among patients with OUD.
Despite the therapeutic challenge of treating alcohol and opioid co-use, there is reason to believe that existing OUD medications may hold promise in treating comorbid AUD, thus potentially reducing alcohol-related risk among patients with AUD. For instance, some treatment options for dual-diagnosis patients overlap, such as naltrexone, which holds dual US Food and Drug Administration (FDA) indications for OUD and AUD treatment and has been found to be used more commonly in patients with both OUD and AUD than patients with OUD but without comorbid AUD although an open randomized study 16 suggested that methadone and buprenorphine suppressed alcohol intake in heroin users, other analyses 17 have not replicated these effects, with some studies 18 showing increased problematic use of alcohol after initiation of opioid agonist treatment. Existing studies 16,17,19 are also characterized by short follow-up periods, modest sample sizes, and selfreported data.

Study Overview
This case-control cohort study used a repeated-event, case-control design using the MarketScan pharmaceutical claims data. We studied whether alcohol-related acute medical events were associated with OUD pharmacologic treatment. This design used each subject as his or her own control by comparing OUD medication use (exposure) at the time of event with exposure during control periods. Thus, the exposure variable was various types of OUD pharmacologic treatments, and the outcome variable was alcohol-related acute events. We also examined whether patients with

Study Setting
The MarketScan databases represent claims from both employer-sponsored and Medicaid insurance holders in the US. 20  Because all data were deidentified, our study was exempt from human subjects review and the need for informed consent by the institutional review board at Washington University. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) and RECORD-PE reporting guidelines.

Participants
Our analytic sample was derived from a previously described cohort of 304 676 individuals in the US, aged 12 to 64 years, who had prescription drug coverage, a diagnosis of OUD, at least 1 OUD claim during enrollment, and continuous insurance enrollment. 20 Figure 1 depicts the derivation of our final analytical sample. Because we used a case-control design, we limited analyses to the 29 157 individuals with alcohol-related acute events (ie, 275 519 individuals were ineligible because they did not experience the outcome of interest). The first event for each person subsequent to initiating OUD treatment was defined as the index event. The index event date was used to generate a longitudinal data set at the day level containing all days in which a given individual was observed (ie, covered by insurance), up to 1 year before and after the index event; that is, the final data set contained 1 record per day per person, with the units of observation being person-days. Our decision to choose a 2-year maximum observation window was guided by the mean length of time individuals were observed in our data set, which was approximately 2 years. The selection of a standard, 2-year observation window also helps fulfill the case-control design's requirement of a stable within-person probability of exposure over a defined period. 21 Finally, we further limited our analyses to individuals who received OUD medication at least once, thus excluding 15 822 individuals who never received any pharmacotherapy for OUD (buprenorphine, naltrexone, and methadone). Participants were permitted to contribute multiple events as long as they fell within a maximum of 1 year before and after the index event.

Variables
Our primary outcome was emergency department visit or inpatient hospitalization for any alcoholrelated acute event during insurance enrollment. Such events were defined on the basis of established methods and ICD-9 and ICD-10 codes 22 as either (1) 100% alcohol-attributable acute events (eg, alcohol poisoning) or (2) health care encounters involving an alcohol-related diagnosis (eg, alcoholic liver disease or chronic alcohol dependence) in combination with acute events that are known to be associated with excess alcohol use (eg, falls, injuries, and poisonings). 22 Our primary exposure variable was OUD medication by days, operationalized as days characterized by the presence or absence of prescription coverage or procedure codes for buprenorphine, methadone, or naltrexone (oral or extended-release). Specific national drug codes (buprenorphine and oral naltrexone) and procedure codes (methadone and extended-release naltrexone) can be found elsewhere. 20 We assumed that the prescription was filled on the date after the prescription was received by the pharmacy and taken until the last day supplied. We assumed that extended-release naltrexone was active for 28 days from the date of injection.
Because the MarketScan data do not differentiate between patients who were receiving naltrexone for OUD as opposed to AUD, we further stratified between patients with existing AUD diagnoses at the time of OUD pharmacologic treatment initiation or during the 6 months preceding treatment initiation, and those without recent AUD diagnoses. AUD diagnoses were identified by insurance claims using ICD-9 and ICD-10 codes during any inpatient or outpatient encounter.
For the purposes of sample description, we also obtained data on time-invariant variables, such as age, sex, and insurance status, which were compared among the subsamples with and without recent AUD claims. Because the case-control design allows for individuals to serve as their own control, we did not adjust for these time-invariant variables in our models. However, we included time-varying covariates corresponding to prescriptions for medications commonly found in alcohol and drug poisonings 23 ; these included muscle relaxants, antispasmodics, opioids (not buprenorphine or methadone), antidepressants, mood stabilizers, antipsychotics, nonsteroidal anti-inflammatory drugs, and benzodiazepines (eTable 1 in the Supplement). The logic for including these was 2-fold.
First, any interaction between these medications and OUD medications may be associated with the point estimates for the OUD medications. Second, use of these medications may serve as proxy variables for comorbidities that can be measured with greater frequency than diagnoses.

Statistical Analysis
We calculated descriptive statistics for the primary analytical sample using χ 2 and Kruskal-Wallis tests, which were stratified by recent AUD claims. For our primary analysis, we used a case-control approach described by Allison et al 24 that includes all observed days limited to 365 days before and after the index alcohol-related acute event already described. Thus, case days included the index event and any subsequent event, whereas control days included all other days within the observation window. An overview of repeated-event, case-control design is illustrated in the eFigure in the Supplement. Models to estimate the odds ratio (OR) describing the association between events and OUD medication exposure days were estimated using conditional logistic regression, with each person serving as the stratification (conditioning) variable. Medication days were coded as 1, and nonmedication days were coded as 0 (reference group). This approach is analogous to a stratified Cox-regression with recurrent events. 24 As a recurrent-event approach, our analysis allowed for incorporation of time (days since index event) as a covariate, mitigating secular confounding that is often a concern with single-event, case-control studies. [24][25][26] We controlled for secular time trends using restricted cubic splines. 27 Unadjusted models included only OUD medications, whereas adjusted models controlled for exposure medications commonly associated with alcohol poisonings. 23 All P values were derived from 2-sided tests, with significance reached at P < .05.
Statistical analysis was performed using SAS statistical software version 9.4 (SAS Institute) from June 26 through September 28, 2020. A total of 55.9% of the sample (7462 participants) had a recent AUD claim before or at start of OUD treatment, defined by diagnoses within 6 months of the start of treatment. These patients were   Figure 2 shows the adjusted ORs of admissions for alcohol-related acute events associated with days using OUD medications vs nonmedication days. A full listing of point estimates and 95% CIs is enumerated in eTable 2 (unadjusted models) and eTable 3 (adjusted models) in the Supplement, with Abbreviations: AUD, alcohol use disorder; OUD, opioid use disorder.

Association of OUD Medication With Alcohol-Related Acute Events
a This table depicts the percentage of insurance coverage days entailing an alcohol-related acute event, which is compared among days when participants were receiving buprenorphine, methadone, naltrexone (extended release or oral), and nonmedication treatment days. These analyses were further stratified between persons who had recent AUD claims and those without recent AUD claims.
the association between OUD medication and alcohol-related acute events unchanged between unadjusted and adjusted analyses. Covariate adjustment had little impact on the other parameters in the model. Among agonist medications, buprenorphine was associated with a 43% reduction (OR, 0.57; 95% CI, 0.52-0.61) and methadone was associated a 66% reduction (OR, 0.34; 95% CI, 0.26-0.45) in adjusted odds of alcohol-related acute events. Among antagonist medication, oral naltrexone was associated with a 37% reduction (OR, 0.63; 95% CI, 0.52-0.76) and extended-release naltrexone was associated with a 16% reduction (OR, 0.84; 95% CI, 0.76-0.93) in the adjusted odds of such events.
In the stratified analysis, buprenorphine's protective associations against alcohol-related acute events were more pronounced among patients with OUD with recent AUD claims (OR, 0.41; 95% peers without claims; however, differences in adjusted ORs by AUD status were not significant because of wide 95% CIs among patients with AUD. Across all analyses, the 2 forms of naltrexone did not significantly differ from one another in their association with alcohol-related acute events.

Discussion
Our study shows that OUD treatment days were associated with reductions in overall odds of admissions for alcohol-related acute events (ie, falls, injuries, and poisonings). We observed risk reductions of 43% for buprenorphine and 66% for methadone treatment days. Extended-release and oral naltrexone treatment days were associated with 37% and 16% reductions, respectively.
Naltrexone use was more prevalent among patients with OUD with existing AUD claims than their peers without such claims, consistent with naltrexone's dual FDA indications of naltrexone for both AUD and OUD treatment. However, naltrexone and buprenorphine showed similar overall protective outcomes against alcohol-related acute events.
Interestingly, our findings illustrate that the greatest overall reduction in alcohol-related acute events were observed with methadone. Time-varying confounding cannot be ruled out because  these results can be partly explained by the structure of opioid treatment programs (ie, methadone clinics), which frequently monitor alcohol and illicit drug use with urine drug screens. The effectiveness of OUD medications in AUD may be more limited when it is not accompanied by psychosocial interventions, such as relapse-prevention therapy or counseling focused on medication adherence, 28 which are seen in opioid treatment programs that dispense methadone. The association between methadone treatment days and alcohol-related events warrants further investigation.
Of note, our results should be interpreted in conjunction with known differences in medication adherence. The true practical effectiveness of antagonist therapies may be hindered by lower medication adherence associated with naltrexone compared with agonist treatments in the OUD population. For instance, substantially greater medication adherence has been observed for buprenorphine treatment (median filled prescriptions of 19 months) compared with both extendedrelease (9 months) and oral (5 months) naltrexone, 29 with patients with OUD found to show similar adherence to methadone and buprenorphine. 30 As our study evaluates ORs that describe protective outcomes associated with medication at a given point in time, the protective outcomes observed in association with buprenorphine and methadone treatment days are likely to be amplified by their higher adherence rates compared with naltrexone. However, our analyses rely on the implicit assumption that risks associated with medication or lack of medication are constant over time. More investigation is needed to evaluate how these risks might vary before, during, and after OUD treatment, in addition to investigating the comparative effectiveness of buprenorphine, methadone, and naltrexone in suppressing alcohol use in patients with OUD.

Limitations and Strengths
There are several limitations to note. First, although the MarketScan data benefit from a large, national sample with ample longitudinal follow-up, measurement error cannot be ruled out, such that medication coverage does not always reflect actual consumption. Furthermore, patients with OUD without observed AUD claims may have a history of treated or untreated AUD not reflected in recent insurance claims. However, we observed a higher prevalence of naltrexone use in patients with AUD claims, and we also observed significant differences in OUD medication treatment outcomes between patients with OUD with recent AUD claims and peers without such claims, suggesting that the distinction between these 2 groups may be clinically meaningful and thus warrants more research. Second, unmeasured time-varying factors associated with alcohol-related admissions and OUD medication exposure can introduce confounding in case-control designs. We mitigated this, however, by including calendar time and time from event as covariates and restricting participants to 2-year observation periods surrounding the index event in order to reduce heterogeneity in observation time. We also used bidirectional sampling in order to reduce overlap bias resulting from control period selection as a function of event time. 25 Third, even though self-matching is used by the case-control design, residual confounding by unmeasured time-invariant variables cannot be ruled out. Fourth, the MarketScan data only encompass insured patients with observed alcoholrelated acute events, which limits our study's generalizability.
Despite these limitations, our study was strengthened by its repeated-event, case-control design, allowing for reduction of selection and sampling bias stemming from conventional observational study recruitment. In addition, we controlled for conditions leading to alcohol-related events via adjustment for medications commonly found in association with such events. The protective associations of buprenorphine, methadone, and naltrexone were not significantly changed in the adjusted analyses, providing evidence of robustness to measured confounders.

Conclusions
The findings of this study suggest that OUD medications are associated with decreased hospital admissions for alcohol-related acute events among persons with OUD. Although naltrexone use was