Optimally Choosing Medication Type for Patients With Opioid Use Disorder

Kara E. Rudolph; Nicholas T. Williams; Iván Díaz; Sean X. Luo; John Rotrosen; Edward V. Nunes


Am J Epidemiol. 2023;192(5):748-756. 

In This Article

Abstract and Introduction


Patients with opioid use disorder (OUD) tend to get assigned to one of 3 medications based on the treatment program to which the patient presents (e.g., opioid treatment programs tend to treat patients with methadone, while office-based practices tend to prescribe buprenorphine). It is possible that optimally matching patients with treatment type would reduce the risk of return to regular opioid use (RROU). We analyzed data from 3 comparative effectiveness trials from the US National Institute on Drug Abuse Clinical Trials Network (CTN0027, 2006–2010; CTN0030, 2006–2009; and CTN0051 2014–2017), in which patients with OUD (n = 1,459) were assigned to treatment with either injection extended-release naltrexone (XR-NTX), sublingual buprenorphine-naloxone (BUP-NX), or oral methadone. We learned an individualized rule by which to assign medication type such that risk of RROU during 12 weeks of treatment would be minimized, and then estimated the amount by which RROU risk could be reduced if the rule were applied. Applying our estimated treatment rule would reduce risk of RROU compared with treating everyone with methadone (relative risk (RR) = 0.79, 95% confidence interval (CI): 0.60, 0.97) or treating everyone with XR-NTX (RR = 0.71, 95% CI: 0.47, 0.96). Applying the estimated treatment rule would have resulted in a similar risk of RROU to that of with treating everyone with BUP-NX (RR = 0.92, 95% CI: 0.73, 1.11).


Standard of care for opioid use disorder (OUD) is long-term treatment with one of 3 evidence-based medications—buprenorphine-naloxone (BUP-NX), methadone, or extended-release injection naltrexone (XR-NTX).[1,2] Available clinical trials show that outcomes of OUD are relatively similar for the 3 medications, while rates of return to regular opioid use (RROU) and dropout from treatment are unacceptably high across treatments.[3–7]

Choice of medication for OUD tends to be based on where the patient seeks treatment and not based on a patient's clinical characteristics. Patients presenting to traditional opioid treatment programs (i.e., methadone clinics) are likely to receive methadone, while those presenting to office-based practices are likely to receive buprenorphine. Whether patients present at traditional opioid treatment programs or office-based settings tends to be influenced by the demographic characteristics of where they live.[8–10] Facilities in lower-income communities with predominately Black and Hispanic/Latinx residents are more likely to provide methadone, while facilities in higher-income communities with predominately White residents are more likely to provide buprenorphine, even in areas where local and state policies have been adopted to increase buprenorphine prescribing in underresourced communities.[8–10] In addition, urban-rural disparities in buprenorphine providers also exist and vary by state.[11]

That the type of medication is largely determined by where a patient seeks treatment, which itself is influenced by structural factors related to sociodemographic and racial segregation, is likely not ideal. Clinical experience suggests that individual patients respond better to one medication than another. If clinicians had a systematic way of choosing the best treatment for each patient, based on their individual clinical characteristics at baseline instead of based only on where they present for treatment, this would have the potential to improve outcomes and move the field toward rational differential therapeutics.

The population of individuals with OUD is heterogeneous, with a range of characteristics that may relate to optimal choice of treatment (e.g., demographic characteristics, co-occurring substance use and other psychiatric disorders, and substance use patterns).[12,13] In previous work, we and others found evidence of heterogeneity in the relative effectiveness of medications across individuals.[14,15] For example, in a comparative effectiveness trial of XR-NTX vs. BUP-NX, homeless individuals were found to have lower RROU rates when treated with XR-NTX, while those with stable housing had lower RROU rates when treated with BUP-NX.[14,15] Such findings are fodder for calls to personalize OUD treatment,[2] but there remains no quantitative evidence about how to choose among the 3 Food and Drug Administration (FDA)-approved medications for a given individual.

Making treatment decisions personalized for an individual's characteristics and clinical history, such that their chance of a successful outcome is maximized, is the goal of learning optimal individualized treatment rules.[16] Learning such rules, which can take multiple characteristics into account simultaneously, is recommended over fitting a parametric regression model that identifies single individual characteristics that modify treatment effects.[17] One reason for this recommendation is that individuals are, of course, multidimensional—no person's race/ethnicity or comorbidities exist in isolation. For example, it may be that 26-year-old Latino men with a particular clinical and substance use history fare better on one medication versus another. Thus, learning optimal individualized treatment rules allows for the identification of subgroups characterized by a particular combination of characteristics that have qualitatively different treatment success across medication types.

Our goal was to learn a rule by which to assign individuals with OUD to treatment with BUP-NX, methadone, or XR-NTX based on their demographic and clinical characteristics such that their risk of RROU would be minimized. We were also interested in estimating the extent to which implementation of the optimal treatment rule would reduce risk of RROU compared with: 1) treating everyone with BUP-NX, 2) treating everyone with methadone, or 3) treating everyone with XR-NTX. To do so, we used harmonized data across 3 randomized comparative effectiveness trials for OUD treatment. Our statistical approach was doubly robust and incorporated machine learning for flexible, data-adaptive model fitting.