Primary care models for treating opioid use disorders: What actually works? A systematic review

1Recent spikes in opioid-related overdoses have led experts to advocate for the creation of primary care-based treatment models to expand access to treatment for Opioid Use Disorders (OUD). OUD is categorized by individuals exhibiting signs and symptoms of compulsive behavior related to the self-administration of opioid substanceswithout a legitimate medical cause or in doses excess of what is clinically necessary. Internationally, of an estimated 48.9 million opioid/opiate users, 187,100 experience drug-related deaths annually. In the US alone, about 2.5 million citizens have OUD, with an estimated 60,000 deaths due to drug overdoses occurring annually. A paucity of specialized substance treatment facilities and rising demand for OUD treatment presents primary care-based models the opportunity to increase access to treatment.

2Over the past 15 years, health systems have developed and tested models to incorporate the use of medication-assisted treatment (MAT), also referred to as opioid-assisted treatment, into primary care settings. MAT uses pharmacological treatments such as buprenorphine and methadone coupled with psychosocial care to treat patients with OUD. Primary care-based models for MAT appear to have roughly equivalent efficacy and outcomes to specialty substance treatment facilities in certain populations with the added advantage of managing, and potentially improving, comorbidity outcomes. A recent scoping review has only looked at U.S. models, but no systematic, rigorous international comparisons with a focus on implementation structures and processes of OUD MAT in primary care settings exist to date. No studies have attempted to synthesize the core implementation structures of these interventions, and as a result, little is known about the components included in effective models in primary care settings. This gap in the literature demonstrates the need to identify which components of primary care models for OUD treatment have shown success in implementation and acceptance by patients.

3This systematic review aims to evaluate the literature on interventions for treating OUD in primary care settings using an established systems design framework: Systems Engineering Initiative for Patient Safety (SEIPS) 2.0. We use this framework to answer the questions: what structural characteristics and implementation components are described in existing primary care models for treating OUD, and how can we improve upon them in the future? Specifically, we aim to: (1) identify thematic components of primary care OUD MAT models that are accepted by patients and physicians and associated with improved health outcomes (2) use those findings to guide future policy and provide recommendations on design features of delivery models found to be effective in the primary care setting.

4We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) recommendations in conducting this systematic review (PROSPERO 2016: CRD42016033762). With the assistance of a medical research librarian (MC), we performed serial literature searches for English language articles. MEDLINE via PubMed, CINAHL, EMBASE, and PsychInfo were searched for studies published prior to August 1, 2016 using Medical Subject Headings (MeSH) and keywords based on primary care settings and treatment of OUDs (S1 Table). All human studies published in full-text were eligible for inclusion, and no publication date or status restrictions were placed. Additional studies of interest were identified by hand searches of bibliographies. Authors were contacted by email if further clarification was needed.

5Two authors (KK and CB) independently screened titles and abstracts for eligibility. Given the complexity of designing and evaluating care models, we included both experimental (RCTs) and observational studies (cohort, case-control, cross-sectional) if they met inclusion criteria. Articles were included if the intervention: (1) evaluated a primary care-based health delivery model where primary care was defined as care delivered in a general practice setting (i.e. private practice, academic primary care clinic) by a general medical internist and/or family medicine physician only, (2) targeted adults (18 years or older) with OUD defined as patients engaged in care to treat their opioid addiction, (3) evaluated patient-level outcomes (e.g. patient retention, urine toxicology screens, satisfaction, effect on health screening for comorbidities, etc.), and (4) evaluated the care model using qualitative or quantitative methods. Studies that did not include a description of the care delivery model evaluated (i.e. only discussed physician perceptions of OUD or drug dosage efficacy studies), focused exclusively on comparing intervention settings (e.g. specialty care versus primary care settings) without a detailed description of the primary care intervention/program design, and concentrated on specialty-based primary care (e.g. HIV care) outside of a primary care physician (PCP) led primary care practice were excluded (S2 Table). In the event of a disagreement in exclusion or inclusion between the two reviewers, a third reviewer (PL) resolved the discrepancy.

6Two authors (KK and CB) used a standardized form adapted from the Cochrane Collaboration to extract data from the included studies, independently and in duplicate. The following data was extracted for all studies: location, study design, intervention design and duration, care model structures and processes, classification, delivery staff, sample size, patient population, and primary/secondary outcomes as stated by the authors of each study.

7Two authors (KK and CB) independently assessed risk of bias via the validated Downs and Black tool which utilizes the following elements to assess risk of bias in both experimental and observational studies: quality of reporting, internal validity of the study and its power, and external validity and confounding. The tool evaluates each of these elements using 27 questions, allowing each article to receive a sum score of up to 32 points. For the purposes of this study, the last question assessing statistical power was interpreted as a dichotomous outcome: 0 for insufficient/no power calculation and 1 for studies that provided evidence of power calculation or reference to statistical power. From this alteration, 28 was the highest score possible. As previously reported the following were the final score ranges: excellent (26–28); good (20–25); fair (15–19); and poor (⩽ 14). Any discrepancies or disagreements in data review, extraction, or assessment of risk of bias were resolved by a third author (PL).

8Given substantial clinical heterogeneity in patient outcomes reported (i.e. retention, relapse rates, comorbidity management, satisfaction, etc.) and variability in the drug treatments and dosages used in the models (e.g. methadone versus buprenorphine) within the included studies, formal meta-analyses were not performed.

9The database search retrieved a total of 1,844 articles and 7 articles identified as related publications to those uncovered in the search through other sources. Initial screening eliminated 1,131 articles at the title and abstract level for not fitting the inclusion criteria. Following full review of each of the remaining 104 articles, 63 articles were eliminated because they did not meet inclusion criteria, leading to 41 included publications (Fig 1). Reasons for exclusion of full-text studies included not a delivery care system, not an intervention, not a primary care setting, opiate addiction not primary care diagnosis, and unreported patient-level outcomes. The qualitative synthesis included 41 publications that described a total of 35 unique interventions. Two included models each had >1 publications that reported different outcomes from the same study (implementation outcomes and patient outcomes), which required inclusion of multiple publications for the same study to best evaluate the model’s efficacy. Of these unique interventions (n = 35), there were 10 randomized controlled trials and 25 quasi-experimental or observational studies.

10We present the general study characteristics including settings and outcomes within the appropriate SEIPS domains to concisely summarize the findings without duplication of reporting. In addition, we present case-by-case examples of barriers and facilitators for the interventions using the SEIPS organizational framework (Table 1).

11The SEIPS 2.0 framework, previously used to evaluate system-level approaches to work systems, was used to evaluate the implementation of included study interventions. The SEIPS model is a widely used healthcare human factors framework adopted by patient safety leaders and applied to multiple health settings including primary care clinics. The three human factor principles this model embraces are 1) evaluating performance from a systems orientation, 2) supporting person centeredness through designing work systems that best fit peoples’ capabilities, limitations and performance needs, and 3) focusing on design-driven improvements to develop structures and processes that enhance patient, provider and organizational outcomes. This framework includes domains regarding the person, organization, technologies and tools, tasks, environment, process, patient outcomes, and employee and organizational outcomes of interventions (Table 2). This framework allowed us to categorize the components across the included studies in a systematic way for better comparisons among interventions considering the heterogeneity of study outcomes and processes. Furthermore, SEIPS guided our assessment of the various components that were included in each model to identify what specific processes were impacted and who was involved.

12Studies were conducted in the following countries: U.S. (24), U.K. (3), Australia (2), Canada (2), Austria (1), France (1), Ireland (1), and Italy (1). All studies occurred in primary care centers; however, ten studies compared specialty versus primary care settings. Some of the studies (n = 14) were conducted in academic primary care centers while others (n = 14) occurred in private practice settings. Five studies were in community health centers with limited resources.

13All of the studies meeting inclusion criteria studied buprenorphine (n = 25) and/or methadone (n = 12) treatment in primary care settings. Due to large variability in terminology and reporting, the interventions were grouped into at least one of five care models (e.g. collaborative care vs. integrated care). The most common type was a coordinated care model which had at least two different types of healthcare professionals actively communicating and working together to share care responsibilities (e.g. nurse case manager or pharmacist plus physician). Of the 32 “coordinated care” models, twelve relied upon a nurse case manager or other skilled nursing staff to lead and provide logistical support to the PCP. Often the nurse received training to provide some behavioral counseling. Pharmacists also provided assistance to PCPs by supervising medication dosages. Multi-disciplinary models consisted of two physician disciplines working closely together within the same clinic (e.g. addiction psychiatry and internal medicine). For example, one study specifically evaluated the benefit of adding in-clinic behavioral counseling to standard medical management provided by a general internist for patients receiving methadone for heroin use. Shared care models had specialty services lead the medication induction process (the first week of MAT where the physician determines the dosing, timing and treatment goals of the medication) and then later “handed-off” patients to general internal PCPs. In Cunningham et al (2008), patient induction was initiated by the pharmacist before transfer to the physician’s care. The chronic care model, utilized healthcare resources to increase patients’ self-efficacy in managing their chronic disease. Only two studies met this criterion. For example, one study used this framework to design a home induction protocol to empower patients to self-administer medications. Last, the physician-centric model had a single physician or group of physicians working together to provide patient-centered MAT without major structural support from other provider types or disciplines. In Doolittle & Becker (2011), the physician independently counseled and treated the patient.

14Studies were assigned to models via the criteria outlined above and could be categorized more than once to best capture treatment delivery. Most included studies (n = 32) had coordinated care models with 32 studies falling into two or more care models (e.g. coordinated care and multi-disciplinary care). Only three physician-centric studies existed, predominantly in community health centers, where physicians independently provided MAT.

15Thirty-one studies included non-physician providers (e.g. nurses, pharmacists, counselors) to carry out tasks. The level of training and specific tasks managed by each non-physician provider varied across interventions. First, multiple studies used nurses as liaisons in coordinating care between PCPs and behavioral specialists. This configuration improved performance processes and collaborative work. Both licensed practicing nurses (LPN) and advanced practicing nurses (NPs) were used as program coordinators to lead the intervention while supporting both patients and staff. For example, Lucas et al (2010) had LPNs oversee patient-physician scheduling and assist with induction. Other studies used nursing staff to not only provide care collaboration, but to lead patient visits. However, only physicians could prescribe medications. Second, pharmacist roles and tasks varied across interventions. Multiple studies had pharmacists supervise dispensing of buprenorphine or methadone. The majority of these were conducted in the Europe. One U.S. study used a clinical pharmacist to provide physician guidance regarding appropriate dosing/tapering strategies and to lead most patient follow-up appointments. Third, heterogeneity in formal training among clinicians for providing addiction counseling emerged. Behavioral counseling providers ranged from PhD-trained psychologists to certified addiction counselors to nurses with brief training in addiction counseling.

16Process focused on the flow of patient care within organizational models. Seven studies had a non-physician (e.g. nurse case manager) perform an initial detailed intake that often consisted of a physical and mental health history, allowing PCPs to focus their time on medication management. Following the intake visit, studies varied on how they handled medication induction. Twenty-nine studies supervised patient induction with frequent appointments and supervised medication dosing. In contrast, four studies evaluated “home” inductions, which increased patient autonomy via a specific plan for how to first begin self-treatment with the chosen medication. Following induction, frequency of appointments with staff ranged from daily to quarter-annually depending upon patient needs and intervention design. Behavioral counseling appointments were often coordinated with medical management appointments. In cases requiring more intense addiction counseling/treatment, nineteen studies had a plan for referral to specialty services.

17Only four studies formally tested if counseling modality or duration affected treatment outcomes. Of these four studies, only two demonstrated that their form of counseling was more efficacious (e.g. patients undergoing CBT had higher rates of negative urine toxicology screens or >80% Quality Health Indicator score) than medical management alone possibly because of their adaptive, stepped-care treatments or highly integrated care teams with numerous support staff.

18Technology/Tools focused on electronic and non-electronic tools that helped manage data and monitor patient outcomes like patient agreements, drug screenings, and electronic information technology systems. Ten studies had formal treatment agreements (“contract”) between patients and providers to outline consequences for continued drug misuse. Most interventions (n = 29) noted that they used urine drug screening as a tool to monitor adherence to medication and drug misuse; although, there was no standardization in what drugs were screened or how often.

19Regarding technologies, three studies noted that they had a panel management structure to monitor patient level data (i.e. urine toxicology screens, drug tests, etc.). Four studies noted using electronic medical records to facilitate treatment team member communication and to document patient updates. No studies utilized home-based or web-based counseling modalities.

20Reported patient outcomes varied. However, most studies (n = 25) reported patient-level retention within treatment. At 3 months, nineteen interventions achieved at least 60% retention. Some studies evaluated if patients had self-reported abstinence (n = 15) while others incorporated quantitative measures of abstinence (i.e. urine toxicology screens; n = 22).

21Few studies asked about patient perceptions of the care delivery models. One study evaluating a coordinated care model for patients receiving MAT obtained patient feedback regarding care with 90% of patients reporting overall satisfaction with the care model. Additionally, less than half of the interventions (n = 11) assessed management of other common primary-care comorbidities and age-appropriate screenings. One of these studies evaluated what percentage of patients were meeting nine quality health indicators (QHI) of an age appropriate health screening per CDC criteria via retrospective chart review. Greater than 3 months of treatment on buprenorphine was positively associated with achieving a recommended QHI screening score [(AOR) = 2.19; 95% confidence interval (CI) = 1.18–4.04)]. Similarly, Roll et al (2015) surveyed 28 patients receiving shared medical appointments for buprenorphine management therapy. They found that 60% of patients reported learning more about comorbidities like Hepatitis C and 43% reported receiving appropriate immunizations since starting the intervention. The BHIVES collaborative was a ten site intervention evaluating the use of buprenorphine/naloxone for patients with both HIV and OUD. The evaluation used mixed-methods and reported patient outcomes on OUD treatment, HIV treatment, HIV related quality of life, patient perspectives on the intervention, and overall quality of life..

22Provider level outcomes were only reported in ten of the included studies. Of these, six studies asked providers qualitatively about barriers and facilitators of the intervention’s success.

23Themes that emerged from provider outcome data included provider education, cost-related barriers, and benefits of a coordinated-care approach. Three studies noted that providers felt under-trained or that some training in providing the chosen medication (i.e. buprenorphine or methadone) or substance abuse treatment was important. O’Connor et al (1998) found that the intervention itself increased provider confidence in treating patients with OUD.. However, Fiellin et al (2004) reported that physicians felt “adequately prepared for much of the care they provided,” but requested that further training be offered with respect to medication tapering, billing, and additional training for support staff. Nine studies reported that providers felt that there were benefits of coordinated care. In Drainoni et al (2014), providers noted that the RN/Counselor taking ownership of the program was pivotal to program success. Likewise, in Weiss et al (2011), providers noted that coordinated care is crucial in a busy academic setting where physicians had limited availability during clinic hours.

24There was heterogeneity in study caliber, medications and dosages, reported patient outcomes, and duration of study making it difficult to perform any meta-analysis with the 35 included studies. However, in order to describe which SEIPS domains may be associated with successful treatment and establish a level of standardization, we defined successful studies as interventions that achieved 60% retention rates at 3 months and received a good score with our validated risk assessment tool. Seven studies met this metric for success (Fig 2).

25All seven successful studies used coordinated care models with multidisciplinary teams. Six of the studies used buprenorphine as the medication, had a modality for delivering behavioral counseling (although not necessarily through trained behavioral health specialists), used nurses as part of the care team, and monitored treatment outcomes with urine toxicology screens. With respect to technology/tools none of the included studies used patient and provider contracts or panel management structures. In addition, only 4/7 included studies provided additional educational training for the clinical staff.

26No study scored an excellent quality/risk of bias score: sixteen interventions scored good, seventeen scored fair, and two were poor for their reporting of patient outcomes significance (i.e. power within the study to detect differences in outcomes) and their lack of both external and internal validity.

27Few comparisons of primary care models for OUD exist. Our study addresses this gap in knowledge by using the SEIPS domains to categorize features of MAT interventions to better understand specific structural, process and outcome elements of primary care models. The range of studies spanned small physician-led interventions within single clinics to large RCTs with multi-disciplinary teams in academic settings. Based on our synthesis of peer-reviewed literature, we report the current structures and processes of primary care-based OUD MAT models and present a proposed research agenda for future studies within each SEIPS domain.

28Coordinated care models (with non-physician team members such as RNs helping manage patient appointments and lab results) are by far the most common delivery structures studied. There was some indication that physicians felt this program model allowed for improved team communication and higher quality of care delivery. Similarly, multidisciplinary teams can promote comprehensive behavioral health counseling in addition to standard PCP-led counseling during routine primary care appointments. However, future studies will need to further delineate the cost and feasibility of these resources in settings where multidisciplinary care is inaccessible or not viable. Ideally, studies would evaluate a physician-centered model against models with varying degrees of care coordination with randomized controlled trials but such studies are costly. With the need to rapidly disseminate primary care based models to provide MAT, this study highlights that policy makers and health care professionals should strive to provide and pragmatically evaluate at the very least, the provision of some coordinated care. These models may include smaller teams or clinical partnerships (i.e. physician-RN teams, physician-pharmacist teams) that are more common across resource settings.

29The effective use of clinicians’ skillsets can improve overall care delivery. For example, clinical pharmacists can provide medication dosing and management rather than only supervised medication-taking. In terms of behavioral counseling, more research is needed to identify the optimal level of training necessary for OUD care delivery. Studies have suggested that additional counseling beyond the scope of the physician is ineffectual for certain outcomes. However, future research will need to understand whether certain patient populations benefit more or less from additional counseling to help allocate limited behavioral health resources to those patients that would derive the most benefit.

30Home inductions proved successful (≥ 60% retention) for select patients, but future research is still needed to recommend the routine use of home inductions and to identify the patient characteristics that are associated with successful and non-successful home inductions. Furthermore, the use of RNs or other support staff to conduct patient intakes (i.e. physicals, mental health screenings) helped disencumber physician responsibilities. Providing patients with “after hours” support was another component of care noted in numerous studies. More research is needed to assess the effects of augmenting such support with the use of mobile technology (i.e. telehealth, text messages, emails) to improve the process of providing 24-hour support. Only three studies examined the influence of addiction specialists transferring stabilized patients to primary care. This approach may appeal to primary care workforces wanting to expand access to MAT through a stepped-care approach (i.e. providing stabilized patients maintenance dosing and managing comorbidities), but who are less comfortable providing initial MAT induction.

31Wide variation in the use of toxicology screens, patient contracts, and data management structures existed and were largely underdeveloped. Our conclusions correlated with findings from a systematic review examining the use of urine drug screens and treatment agreements in patients with chronic pain, and found that more research is needed to standardize these tools to not only monitor patient level outcomes, but provide population-level feedback to care teams. Additionally, much of the technological aids discussed were relatively nascent with Mullen et al (2012) noting that a more sophisticated data management structure would be helpful. There was substantial heterogeneity with respect to patient and provider outcomes measured. Patient retention was the most common outcome reported, yet there was no uniform definition given dissimilar program lengths between studies (i.e. 1 month to 2 years). Additionally, there was a lack of consideration given to the remitting nature of OUD. Many of the study samples excluded co-dependence on other illicit substances like benzodiazepines, leading to possible confounders that could affect the analyses.

32Few studies evaluated provider outcomes such as quality of medical care, physician perceptions, and factors related to care delivery in relation to a primary care-based MAT intervention. While other studies in the literature measure provider barriers to delivering OUD treatment in primary care settings, these studies do not measure provider outcomes concurrently to testing patient level outcomes within specified care model structures. This gap indicates key areas for future research such as determining the appropriate level of provider training, testing provider training and mentoring support mechanisms such as the Providers' Clinical Support System for Medication Assisted Treatment and Project ECHO (Extension for Community Health Outcomes) a video-based distance education program, and determining how to prevent physician burnout.

33Our study could not make any definitive statements regarding whether particular care models or treatment elements are strongly associated with favorable patient outcomes compared to alternatives. However, when we did look across studies with good quality and high patient retention there was a pattern suggesting that successful studies used coordinated, multidisciplinary models to support physicians in delivering MAT. The majority of the seven studies did not use tools such as patient/provider contracts nor did they provide additional clinical staff training suggesting that they may not be necessary to successfully carry out primary care based MAT.

34Our study has limitations. First, because of both the heterogeneity in reporting of outcomes and variability in medications used and dosages, we were unable to perform a pooled meta-analysis to clearly link structural domains identified via SEIPS to health outcomes. Second, not all of the studies included had randomized designs. Potential for bias and confounding should be considered, though it was formally assessed within our methods. Third, we were unable to quantitatively evaluate which particular SEIPS domains contributed to intervention success or failure. Fourth, we only included studies that were published in peer-reviewed literature. Therefore, we did not capture interventions that may be in the pilot phase or have outcomes presented via other “grey” literature such as websites/forums.

35Our study has several strengths. To our knowledge, this is the first systematic review that describes both patient outcomes and structural organization of interventions to treat OUD in primary care settings both domestically and internationally. Second, by using the SEIPS framework, we described systems design elements within each intervention rather than focusing only on the broad organizational framework of the intervention. Not all primary care settings have access to the same resources with some unable to structurally accommodate a multi-person coordinated care team. However, we provide details on how to improve systems (e.g. person, tasks, and process) depending on the resources that are available in various settings.

36There is variability in regulations for treatment programs, payment models, and provider training structures which may limit or enhance the ability of health systems to provide high quality multidisciplinary, coordinated care with the most cost-effective, efficacious OUD interventions. As the US allocates funding towards expanding MAT access, programs receiving such funding would benefit from considering the intervention models found to support MAT implementation in prior studies. By evaluating not only patient efficacy, but also structural characteristics of primary care models for delivering MAT, this review provides key insights for PCPs and researchers about ways to build upon existing resources and personnel to more effectively deliver OUD treatment. Specifically, this study identified key components of primary care OUD delivery models. As we continue to grapple with the global rise of opioid-related morbidity and mortality, this review can help enhance the rapid dissemination of effective OUD treatment programs across diverse settings.