Meta-analysis of primary care delivered buprenorphine treatment retention outcomes

ABSTRACT Background: Currently, the capacity to provide buprenorphine treatment (BT) is not sufficient to treat the growing number of people in the United States with opioid use disorder (OUD). We sought to examine participant retention in care rates of primary care delivered BT programs and to describe factors associated with retention/attrition for participants receiving BT in this setting. Objectives: A PRISMA-guided search of various databases was performed to identify the articles focusing on efficacy of BT treatment and OUD. Method: A systematic literature search identified 15 studies examining retention in care in the primary care setting between 2002 and 2020. Random effects meta-regression were used to identify retention rates across studies. Results: Retention rates decreased across time with a mean 0.52 rate at one year. Several factors were found to be related to retention, including: race, use of other drugs, receipt of counseling, and previous treatment with buprenorphine. Conclusions: While we only investigate BT through primary care, our findings indicate retention rates are equivalent to the rates reported in the specialty care literature. More work is needed to examine factors that may impact primary care delivered BT specifically and differentiate participants that may benefit from care delivered in specialty over primary care as well as the converse.


Introduction
Opioid use disorder (OUD) is a public health crisis in the United States, associated with over 760,000 opioid overdose deaths in the US between 1999 and 2023 (1).Over 11 million people aged 12 or older in the US reported misusing opioids in the past year in 2017 (2).It is further estimated that 2.35 million people currently meet all criteria to be classified as having OUD (2).
Buprenorphine has been approved to treat OUD since 2002 (3).Buprenorphine can be prescribed for up to one month at a time, while each dose of methadone must be administered by a licensed opioid treatment program, meaning daily or near daily visits to the clinic, which may often be inconvenient for patients and not available in many parts of the country, especially rural areas.Approximately 82% of all opioid treatment programs are operating at 80% capacity, and even if all were operating at 100% it is estimated there would still be ~1 million people not being able to receive treatment (4).These alarming figures indicate a need for further expansion of buprenorphine treatment (BT).We use the term buprenorphine treatment (BT) throughout to indicate that we were looking at buprenorphine only (not methadone or naltrexone).Buprenorphine treatment is defined as medication for opioid use disorder which assists in soothing symptoms of withdrawal, decreasing cravings and the effects of opioids, and diminishing the risk of overdose due to opioids.
Between 2010 and 2018 BT availability in primary care facilities increased from 12.9 people per 10,000 to 29.7 people per 10,000 (5).This increase is a positive step, but the unmet needs of communities throughout the United States are still growing (5).BT through primary care facilities has been shown to be comparable to methadone in keeping patients retained for 2 years (6).Patients receiving primary care provision of buprenorphine have also been found to have similar retention rates as patients treated by specialty providers (7).Specialty providers include, MOUD in a behavioral health setting, SUD specialty care, as well as opioid treatment programs.
In addition to the expanded capacity offered by delivering BT in primary care settings, there are other health benefits.Preventive primary care screenings (HIV; HCV; A 1c ; hypertension) are higher among patients in BT (8).The receipt of BT is associated with lower HIV viral load and greater proportion of anti-retroviral uptake (9,10).Retention in BT has also been associated with an increase in HCV care and treatment (11).Glycemic control among diabetic patients receiving BT showed a significantly greater reduction in A 1c than those not treated (12).The addition of BT to primary care practices not only increases the number of patients that can be treated, but may also have an important impact on treating other conditions (13).Further, receiving BT in primary care can be less stigmatizing to patients, and potentially lead to better outcomes (14).
The extent of the opioid crisis, related consequences, frequent comorbid physical and mental health disorders, and the lack of treatment capacity for this disorder make primary care a logical venue for treatment.To examine the effectiveness of primary care delivered BT, we conducted a meta-analysis to examine retention rates at various time periods.The findings provide a benchmark for primary care programs that provide medication for opioid use disorder.We also conducted a systematic review of the literature and provided a summary of the findings related to factors that may affect retention.

Inclusion/exclusion criteria
We included articles that (1): assessed outcomes of BT in primary care clinics (2); included participants receiving care in the US (3); reported on buprenorphine/ naloxone or the mono product buprenorphine (4); were published between January 2002 and July 2020 (5); reported retention in proportions retained at standard time points (6); were published in English.Articles were excluded if (1): they included participants <18 years old (2); examined the impact of buprenorphine for pain management only or (3); included only outcomes of currently pregnant participants.

Search strategy
We searched electronic databases (PubMed, OVID, SCOPUS, and CINHAL) to identify articles using combined search terms (appendix 2).Citations were imported into Endnote X and duplicates removed.Each abstract was reviewed by two of the authors.A full article review was conducted if one or both authors considered it to be indicated.We have included a PRISMA diagram (Figure 1) detailing the review stages.

Data extraction
Of the 37 articles eligible for full review, 15 were included in full text analysis.Retention data was extracted for the time periods of interest (1, 3, 6, and 12 months).The weight of each study was calculated through the proportion retained in treatment and the sample size of the study as indicated in the formula below-W p = n/(p(1-p)), where n = the total sample size and p = the proportion retained at each time period of interest.
Weights were used due to the drastic difference in sample sizes between studies.Some studies had less than 50 patients, while others had over 1000.

Quality assessment
All studies were rated on a 1 (low quality) to 5 (high quality) scale by four authors (RLC, JW, RE, CB) utilizing a method more thoroughly described in Hammick et al. (15).Articles were rated using Likert scale for each of 5 domains of article quality including study design, ethical concerns, participant recruitment, data collection and analysis, and extent to which threats to validity were controlled.Each article was reviewed by two authors, and in the case of incongruent scoring, a third author assessed quality.A consensus building approach was used to arrive at a final rating.Only one article, Suzuki et al. (16), yielded an incongruent rating, and this was resolved via a conversation between the reviewers.Congruence of the study design to the research question, implementation fidelity, appropriateness of statistical analyses, and the degree of control of threats to validity were all considered in the ratings.

Meta-analytic approach
The outcome variable, retention in treatment, is measured as a proportion of patients still in treatment at 1, 3, 6, and 12 months.Four meta-analyses were performed to estimate the pooled retention rate at each time point.Random-effect meta-analysis models were estimated at a p-value less than 0.05 from the Cochran's Q.The alpha threshold for statistical significance was determined to be 0.05.I 2 statistics were used to examine the potential heterogeneities in retention rates at different time points.Tau-squared (τ 2 ) was employed to estimate the between-study variance in a random-effects metaanalysis.The explanatory variable in the metaregression was categorical ordinal time, with time point at month 12 as the reference group.Odds ratios for retention at each time point were calculated relative to month 12. Publication bias was assessed using the Egger's test, Begg's test and a funnel plot.All statistical analyses were conducted using version 3 of Comprehensive Meta-Analysis software.

Results
Including all 15 studies, a total of 3,461 participant observations were included in the review.When considering retention at each time point, the four studies that assessed 1-month retention included 680 participants; the seven assessing 3-month retention included 1,455 participants; the eight assessing six month retention included 1,132; and the three that assessed 12 month retention included 1,797 participants.Appendix 1 provides a summary of each study.

One month meta-analysis retention outcomes
The summary effect size (proportion), individual effect sizes, their 95% confidence intervals, Z-statistics, and p-values are shown in Table 1.The forest plot provides visual representation of individual and summary effect sizes.The summary proportion for retention random effects models was 0.811.The random effects model had a two-sided p-value of < .001.The Q statistic showed the observed dispersion as 25.285 indicating the absence of variation across effect sizes.The I 2 measure indicated that ~ 88% of the observed variance between the studies were due to real differences in the effect size.Based on a τ 2 of 0.406, the between studies variance was used to compute weights.It seemed that Stein et al. (17).was an Articles that were duplicates, book chapters, conference proceedings and opinion eds were excluded.Criteria such as research design, sample size, risk of bias, and confounders were incorporated in the screening.The quality assessment of screened articles was completed using a consensus process among the authors who were involved with the systematic review.
outlier study when compared to the rest of the studies for month 1 and was removed from the random effects model.Visually, the confidence interval of the Stein article did not overlap with any of the other studies, and as such we used the random effects model with an estimated retention rate of 0.811.

Three month meta analysis retention outcomes
The summary estimates for the seven studies reporting the 3-month retention rate, their corresponding 95% confidence intervals and p-values using random effects were (0.707, C.I:0.665-0.746,and < 0.001, respectively.The value of I 2 was 60.149, which indicated a substantial heterogeneity.The noncompatibility measure of studies Q = 15.056 was significant (p-value = .02).The τ 2 between studies variance was 0.038 with a standard error of 0.039.The random effects model, with an retention rate of 0.707, was the best fit for the 3-month retention period.Table 2 summarizes these findings.

Six month meta analysis retention outcomes
Nine studies are represented in Table 3, which describes the 6-month period.Estimates of retention, their corresponding 95% confidence intervals and p-values using random effects were 0.584, 0.536-0.629,and < 0.001, respectively.The value of I 2 was 75.128 which indicated a moderate heterogeneity.The non-compatibility measure of studies Q was equal to 32.165 which is significant (p-value < .001).The τ 2 between studies variance was 0.056 with a standard error of 0.045.The random effects model with an retention rate of 0.584 was used for the month 6.

Meta analysis for month 12
Table 4 summarizes the findings of the four studies reporting 12-month retention period.The summary estimates of proportion, their corresponding 95% confidence intervals and p-values using random effects are: proportion = 0.520; C.I:0.484-0.555;p-value = .27.The value of I 2 of 55.953 indicates moderate heterogeneity.The noncompatibility measure of studies, Q, is equal to 6.811 (p-value = .08).The τ 2 between studies variance was 0.011 with a standard error of 0.017.The random effects model, retention rate 0.520, is used for the month 12.

Random-effects meta-analysis for all time points
Table 5 provides the names of all studies, the time points the study ended, the sample size, the estimated event rates and their corresponding standard errors.Table 6 shows a random-effects-meta analysis of retention rate for all time points.The mean effect size for all study populations was 0.65 (95% CI 0.577,0.716)with a pvalue of < .001.The standard deviation of true effects τ was 0.562.τ 2 was 0.316.In our example, Q was 260.31 with 14 degrees of freedom and a p -value of < .001.Thus, we were able to reject the null hypothesis and conclude that there is evidence that the true effect size does vary across studies.I 2 had a value of 94.622 which indicated that approximately 95% of the variance in observed effects reflected variance in true effects rather than random error variance.

Random model logit meta regression analysis
Since the duration of time was not uniform across all studies, we performed a random model logit meta- regression with retention rate as our dependent variable and the moderator (Time point) as an ordinal categorical predictor of retention rate.Table 7 gives a summary output for the logit random effects meta regression for full model.As shown in Table 7, the statistic Q for our logit model with ordinal categorical time as its covariate was 37.67 with 3 degrees of freedom and a p -value of < .001.This Q statistic tells us that the model is able to explain at least some of the variance in our effect size (retention rate).The statistic τ 2 was 0.103, which represents the variance of true effects about the regression line.The statistic Tau is 0.321 which denotes the standard deviation of true effects about the regression line.The I 2 statistic tells us what proportion of the variance in observed effects about the regression line is due to variance in true effects rather than sampling error.In our model, I 2 was 82.19%, which means that about 82% of the observed variance about the regression line reflected variations in true effects rather than sampling error.The Q statistic is used to test the null hypothesis that all variation of observed effects about the regression lines is due to sampling error.Equivalently, the null hypothesis is that the true effect size for all studies falls directly on the regression line.Under the null hypothesis, Q would be distributed as chi-squared with 14 degrees of freedom.In this example, Q was 64.36 with 11 degrees of freedom and a p -value of < 0.001.Using type I error probability of 0.05, we rejected the null hypothesis.This is consistent with the fact that τ is not close to zero.The significant pvalue suggests that the true value of τ could not be zero.R 2 = 0.67 represents the proportion of variance explained by the variable time in our regression model.The total variance of all studies about the grand mean was 0.316 which is the variance of true effects.The full model showed that the variance of true effects about the regression line was 0.103.Thus, the proportion of variance explained by the covariate time is 0.213/0.316,which is approximately 0.67.Hence, 67% of variation in retention rates across all studies can be explained by the covariate time.Figure S1 (see supplemental material) illustrates the downward trend of logit event rate versus the moderator (Time point).The retention rate showed a downward trend starting with high retention rate for month 1 which steadily decreased subsequently.
Figure 2 shows the bar graph representing the odds ratios for retention using month 12 as a reference category.The odds of retention in month 1 were 4.47 times the odds in month 12 (95% CI: 2.660,7.504).The odds for retention in month 3 were 3.88 times the odds of  retention in month 12 (95% CI: 2.243,6.721).Finally, the odds ratio for month 6 compared to month 12 is 2.01 (95% CI: 1.068,3.751).

Publication bias
We used the non-parametric Begg and Mazumdar rank correlation test for testing the null hypothesis of the absence of publication bias.The Kendall's tau statistic using continuity correction factor was 0.229 with a z value of 1.188 and a two-sided p-value of .24which indicated a lack of publication bias.In addition, to validate the absence of publication bias, we used the parametric test of Egger's regression intercept.The intercept was 2.861 with a standard error of 1.956, and a 95% confidence interval estimate of (−1.364, 7.086) with a two-sided p-value of .17indicated the absence of publication bias.The results are illustrated in Figure S2 (see supplemental material).

Alcohol use and retention
Two studies considered the relationship between alcohol use and retention in treatment (18,23).One study found significantly increased retention of buprenorphine treatment while concurrently using alcohol at 12 months (18), while the second found decreased retention at 3 months (23).

Illicit drug use and retention
The participants in the reviewed articles were predominantly people using heroin, with 10 of the 11 studies reviewed reporting greater than 60% of the participants using heroin.The relationship between use of other drugs and retention was examined in four studies.One study reported no difference in sixmonth retention rates between cocaine users and nonusers (17), one reported higher retention in the cocaine use group at 12 months (18), and a third reported lower retention among the cocaine group at six-months (24).Two of the studies defined cocaine use as any use of cocaine at baseline or during treatment (8,18), while the third excluded patients with a diagnosis of cocaine dependence (25).Only one article examined concurrent cannabis use and its effect on treatment retention, but did not find any association between the two (17).Two studies examined the association between non-medical benzodiazepine use and treatment retention (17,18).Neither reported a significant relationship.

Comorbid physical and psychiatric disorders and retention
Among participants whose chart indicated the presence of another psychiatric disorder, 72.9% were retained at 1 year, while only 57.9% were retained among those who did not (20).The type of psychiatric disorder was not specified in the study, and none of the studies considered the relationship between comorbid psychiatric disorders and retention at shorter time periods.
One article compared outcomes of chronic pain patients using opioids non-medically and opioiddependent patients without chronic pain at 6 months (20).The chronic pain group was significantly less likely to be retained in treatment at 6 months when compared to those without.

Treatment factors and retention
Previous buprenorphine treatment was mentioned as a potential predictor in two studies and was found to be associated with short and midterm (3 and 6 months) (25), but not long-term retention.Overall, the findings were similar to those reported in prior research suggesting previous buprenorphine treatment was not an effective predictor for long-term retention (20).
One study in the review found that patients transferring from methadone to primary care delivered buprenorphine treatment had no association with treatment retention (17).Several of the studies reviewed excluded patients on higher doses of methadone, citing difficulty managing the transition from long acting opioid agonist treatment to buprenorphine in a primary care setting (17,22,23,26,27).
Analysis comparing retention by provider specialty was done in two studies.Retention rates were found to be equal for the comparison between primary care attending physicians and resident physicians (13).
While most studies in the review mentioned some form of counselling delivered alongside BT, only two examined the relationship between receipt and retention in care.Both found positive associations of retention and receipt of addiction counseling one at three and 6 months ( 8) the other at 6 months (2).

Socio-economic factors and retention
Three studies found employment positively associated with retention at three (25), six (17), and 12 months (20).The fourth study found that unemployment or receiving disability was associated with higher 12month retention (22).History in the criminal justice system was examined in one study and no difference was found in those involved compared to those that were not at 3-, 6-or 12-month points (21).One study analyzed the effect of housing on retention, finding no association between being unhoused and retention (28).

Discussion
We sought to identify the retention rates of participants in buprenorphine treatment delivered in primary care clinics to provide benchmark for other primary care delivered buprenorphine providers.This is of particular importance as the opioid crisis is expanding and additional treatment venues are needed.As these venues become more common, it is critical to examine the associated outcomes.Average retention rates across all four measurement points were: 1 month 76.45%; 3 months 70.9%; 6 months 57.76%; and 12 months 51.95%.These findings from the meta-analysis were similar to those found in specialty care settings (7).This is a critical finding, providing further support that primary care settings are a viable extension for opioid use disorder treatment.Further, the findings here provide a metric by which to assess the quality of care in this venue.These results illuminate a very important line of research that should be pursued in subsequent studies, which is the identification of patient profiles that are better served in primary versus specialty care could further improve outcomes.
Three studies examined associations between race/ ethnicity and retention and two of these found being Black or Latinx was associated with lower retention (10,20).Other studies have demonstrated that access to BT treatment is also lower for Black people, as overdose rates are rising in this population (29).Studies examining the cultural competence of prescribers and their impact on retention are needed to stem the rise of overdose in this population.
Heroin use was associated with lower rates of retention in most of the reviewed studies.As the drug market evolves to include more fentanyl, studies examining differences in retention that include fentanyl use are needed (29)(30)(31).Alcohol, cocaine, cannabis, and benzodiazepine use have all been examined in relation to MOUD retention with mixed findings (8,17,18,25).Additional studies examining co-morbid use of other substances are needed, particularly methamphetamine use, which is currently rising and has been associated with a higher rate of fatal overdoses (32).Further, no studies have examined the relationship of BT retention and a diagnosis of another drug use disorder.Similarly, the presence of additional psychiatric disorders has not been thoroughly studied as a factor for retention in primary care settings.One study in our analysis reported that "other psychiatric disorders" were associated with higher retention, however the type of disorder was not specified (20).The presence of other disorders may be a factor of particular relevance in primary care settings, as these practices may not have the resources to provide behavioral treatments nor have the psychiatric personnel required to manage mental illnesses.
Comorbid physical conditions might also be contributing factors associated with retention in the primary care setting.HCV is quite common among buprenorphine patients, and one study found no relation between HCV and retention at one year while another found lower retention at one and two years in another (15).HCV treatment can be easily managed in primary care settings, and studies examining the impact of successful HCV treatment on retention are needed.
A history of buprenorphine treatment was associated with 3-and 6-month retention (26) but not 12 (20), while previous treatment with methadone did not show any association with retention (17).Many studies excluded patients that were transferring from methadone treatment due to the required extended abstinence period and withdrawals associated with the transfer.As noted above, buprenorphine treatment is a less timeconsuming treatment for patients, making it more attractive particularly for patients that have reached stability.More research is needed to improve transfer protocols to facilitate patient movement to buprenorphine treatment.
Two studies reviewed here found that counseling was associated with increased retention (2,8).Neither study described the type of counseling provided.More research is needed to understand the impact of counselling on BT, whether it is needed for all patients, and what type of counseling approaches are most effective.It is vital that these treatments can easily be delivered in primary care, given the time limitations those providers have with patients.

Limitations
There are a few limitations to this study.First, there is no single definition of retention used in the studies examined.Each study reviewed defined retention but there were differences in each definition that potentially could bias the results.An agreed upon definition of retention in care would eliminate this limitation, and is needed to reliably compare findings.Another limitation is the lack of a comparative group of specialty delivered buprenorphine care.Further, we strongly suggest further research comparing these two approaches, as providing primary care delivered has the benefit of being lower cost, and creating a relationship with a provider to care for physical health disorders.

Figure 1 .
Figure 1.PRISMA flowchart showing the literature search strategy.The PubMed, OVID, SCOPUS and CINAHL databases were searched.Articles that were duplicates, book chapters, conference proceedings and opinion eds were excluded.Criteria such as research design, sample size, risk of bias, and confounders were incorporated in the screening.The quality assessment of screened articles was completed using a consensus process among the authors who were involved with the systematic review.

Figure 2 .
Figure 2. Odds ratios for adherence.This figure indicates the odds of adherence of the patients to buprenorphine treatment (retention) using month 12 as a reference category.The odds ratio for the different months ranged from 2.01 to 4.47 times compared to month 12.

Table 1 .
Pooled retention rates for buprenorphine treatment at month one: Random effects meta-analysis.

Table 2 .
Pooled retention rates for buprenorphine treatment at month three: Random effects meta-analysis.

Table 3 .
Pooled retention rates for buprenorphine treatment at month six: Random effects meta-analysis.

Table 5 .
Study name, time point, and retention rate statistics.

Table 6 .
Pooled retention rates for buprenorphine treatment for all time points: Random effects meta-analysis.

Table 4 .
Pooled retention rates for buprenorphine treatment at month twelve: Random effects meta-analysis.

Table 7 .
Logit random effects meta regression for full model.