An alternative analysis of illicit opioid use during treatment in a randomized trial of extended-release naltrexone versus buprenorphine-naloxone: A per-protocol and completers analysis

Abstract

Background:

The distinct pharmacological properties and clinical uses of extended-release naltrexone (XR-NTX) and sublingual buprenorphine-naloxone (BUP-NX) present challenges in analyzing patient outcomes.

Methods:

We conducted a secondary analysis of a multi-site randomized trial comparing XR-NTX with sublingual BUP-NX treatment for opioid use disorder initiated during inpatient detoxification and continued in outpatient treatment. Urine testing data for non-study opioids from the last 22 weeks of the 24-week trial were analyzed in both a per-protocol sample (n = 474 participants who received at least one dose of medication) and a completers sample (n = 211 participants who received all XR-NTX doses or all BUP-NX prescriptions). The present analyses sought to identify differences in the weekly percentages of opioid-positive urine tests between participants treated with the two medications.

Results:

The proportion of opioid-positive tests in both conditions was less than 20 % for 21 of the 22 weeks in the per-protocol sample and all 22 weeks in the completers sample. Generalized linear mixed model analyses revealed a significant treatment (XR-NTX vs. BUP-NX) X week (weeks 3–24) interaction in the per-protocol sample but not the completers sample. In the per-protocol analysis, the BUP-NX, compared to XR-NTX, had significantly greater proportions of opioid-positive tests in 14 out of the 22 weeks.

Conclusions:

Longitudinal modeling approaches that utilize flexible procedures for handling missing data can offer a different perspective on study findings. Results from the present analyses suggest that XR-NTX appeared to be somewhat more effective than BUP-NX in reducing illicit opioid use in the per-protocol sample.

1. Introduction

Opioid agonists and antagonists are the two broad pharmacological classes of medications used in the United States to treat opioid use disorder (OUD). The partial opioid agonist buprenorphine is typically started when patients are experiencing opioid withdrawal symptoms. In contrast, the opioid antagonist, extended-release naltrexone (XR-NTX), cannot be started until after patients have been opioid-abstinent for an adequate period (typically a week or longer). Both have high affinity for the mu opioid receptor, which should (as long as doses and blood levels are adequate) block the positive reinforcing effects of other opioids. These contrasts in pharmacology and induction approach present a challenge to clinicians and patients, as well as to researchers wishing to characterize and compare patients’ responses to these treatments.

The X:BOT trial afforded the opportunity to determine whether participants randomly assigned to extended-release naltrexone (XR-NTX) or buprenorphine-naloxone (BUP-NX) differed in their patterns of opioid use over time. X:BOT was an open-label, randomized comparative effectiveness trial of XR-NTX versus BUP-NX initiated on eight inpatient substance use treatment services and continued for 24 weeks in community-based outpatient treatment (Lee et al., 2016). On an intent-to-treat basis, X:BOT found significantly lower rates of opioid relapse events and higher rates of opioid negative urine tests for participants who were assigned to receive BUP-NX, due to more participants failing to initiate XR-NTX. However, among participants who successfully initiated their assigned medication (“per protocol” basis), these two outcomes did not significantly differ across study medications.

This paper examines an alternative approach to analyzing the urine opioid testing data as reported in the X:BOT primary outcome paper (Lee et al., 2016). In that report, the total number of urine opioid-negative tests were analyzed as a secondary outcome with missing data considered as positive and without regard to the timing of urine sample collection. The present paper compares trends over time in opioid-positive testing results, using all available data examined weekly rather than as a summative measure at the end of the trial (as did Lee and colleagues) of opioid-negative tests for the BUP-NX and XR-NTX treatment conditions. We had four aims: (1) to determine if the two conditions differed on the likelihood of urine-positive testing results over the course of the trial for those participants who were able to start their assigned medication (i.e., per-protocol sample); (2) to examine these between-condition differences separately by week; (3) to investigate at what point differences occurred within the study period in the per protocol sample; and, (4) to examine these issues in a completers sample, defined as those participants who received all planned XR-NTX injections or all planned buprenorphine prescriptions. This fourth aim would permit an assessment of the medications’ comparative effectiveness under a condition of adherence to XR-NTX injections and receiving BUP-NX prescriptions. Results from this study can be used to understand changes in the proportions of opioid-positive urine screening test results between the two conditions over time in a per-protocol sample that began taking their medication at study outset but that included participants who stopped their medication during the trial, and in a distinct sample within the per-protocol sample that began taking their medication and continued receiving their injections or BUP-NX prescriptions throughout the trial.

2. Methods

2.1. Parent study

Detailed methods (Lee et al., 2016; Nunes et al., 2016) and primary outcome findings (Lee et al., 2018) have been described previously. In summary, participants (N=570) were recruited from eight inpatient detoxification treatment facilities around the US. Inclusion criteria were: (1) ages 18 years old or above; (2) meeting DSM-5 criteria for OUD; (3) seeking treatment for OUD; (4) willing to accept either XR-NTX or BUP-NX, and (5) absence of serious medical, psychiatric, or other substance use disorders, among other criteria. Participants were stratified based on the severity of opioid use and by treatment site and randomly assigned to XR-NTX or BUP-NX. All sites received approval from local Institutional Review Boards.

2.2. Current study

2.2.1. Samples

We conducted our analyses in two defined samples: a per-protocol sample (n = 474), which included study participants who successfully initiated their assigned medication and was the sample in the Lee and colleagues’ analysis (2018), and a completers sample (n = 211) included participants who received all of their planned XR-NTX injections or all of their planned buprenorphine prescriptions over the 22-week period (i.e., weeks 3–24, inclusive). The per-protocol participants were those who enrolled in the study and succeeded in starting their assigned study medication, while the completers sample was a subset of the per-protocol sample that either received all expected buprenorphine prescriptions or received all six scheduled naltrexone injections. Because the purpose of the present analyses was to compare longitudinal trends in opioid-positive urine tests between participants treated with BUP-NX and participants treated with XR-NTX, we did not include in these analyses, the 96 participants who were randomized but who never received an initial dose of medication (“induction failures”).

Out of 570 randomly-assigned participants in the parent study, 283 were assigned to the XR-NTX treatment arm. From the latter, 204 (73 %) comprised the per-protocol sample and 96 (34 %) comprised the completers sample. Of the 287 randomly assigned to the BUP-NX treatment arm, 270 (94 %) were in the per-protocol sample and of these, 115 (40 %) were in the completers sample. The demographic characteristics of the per-protocol and completers samples are shown in Table 1.

Table 1.

Demographic characteristics of participants in the per-protocol and completers samples within each of the treatment conditions.

	Per-Protocol (n = 474)				Completers (n = 211)
	BUP-NX (n = 270)		XR-NTX (n = 204)		BUP-NX (n = 115)		XR-NTX (n = 96)
Demographic Variables
Age at Randomization [Mean (SD)]	33.7 (9.8)		33.7 (9.3)		35.2 (10.3)		34.4 (10.4)
	n	%	n	%	n	%	n	%
Gender
Male	193	71.5%	138	67.6%	84	73.0%	64	66.7%
Female	77	28.5%	66	32.4%	31	27.0%	32	33.3%
Ethnicity
Not Hispanic	217	80.4%	177	86.8%	91	79.1%	85	88.5%
Hispanic	53	19.6 %	27	13.2%	24	20.9%	11	11.5%
Marital Status
Never Married	89	33.0%	69	33.8%	45	39.1%	31	32.3%
Ever Married	180	66.7%	134	65.7%	69	60.0%	65	67.7%
Unknown	1	0.4%	1	0.5%	1	0.9%	0	0.0%

2.2.2. Outcome measure

Urine specimens were collected weekly during the trial and tested for non-study opioids (buprenorphine, methadone, morphine [heroin, co-deine, morphine], or oxycodone). As in the parent study, the urine tests during the first two weeks of the study were not included in the current analyses because during that time participants may have had opioid-positive urine tests from buprenorphine or methadone used in their inpatient medically managed withdrawal (Lee et al., 2018).

2.3. Statistical analysis

Urine opioid testing results were analyzed as a longitudinal (by week) binary outcome. Weekly urine opioid testing results for BUP-NX participants were considered positive if any opioids, oxycodone, or methadone were present in the urine sample for the given week. Opioid tests for XR-NTX participants were considered positive if any opioids, oxycodone, methadone, or buprenorphine were present in the urine sample in a given week.

A generalized linear mixed model (GLiMM), with a logit link function to account for the binary outcome, was used to analyze longitudinal trends in opioid-positive urine test results over the final 22 weeks of the X:BOT trial. GLiMM uses full-information maximum likelihood (ML) based on all available data, to calculate parameter estimates rather than using multiple imputation (MI). ML was used in the GLiMM analyses rather than MI because ML is more efficient, involves fewer analysis decisions, and only one model is used rather than using an imputation model and an analysis model (Allison, 2012).

In both samples, we tested main effects for treatment condition (XR-NTX vs. BUP-NX) and week (week 3 thru week 24), treated as a categorical variable, and a treatment X week interaction effect, all considered fixed effects in the model. Site and participant were also included in the model as random effects, and an AR(1) correlation structure was estimated for within-subject observations. In the event of a significant overall treatment X week effect, all pairwise comparisons within-week were examined. There were no opioid-positive urine tests in the XR-NTX group at week 21 in the completers sample, producing an empty cell in the statistical model that resulted in a failure to converge. As such, week 21 observations for both groups were excluded from the model for the completers analysis.

Subsequently, the observed percentages of weekly opioid positive urine test results by study condition were plotted for the per-protocol and the completers samples. Additionally, repeated-measures logistic regression models, with an AR(1) correlation structure for within-subject observations, were also fit to test for treatment differences in missingness of urine opioid tests over the course of the study, with treatment and week as predictors. A treatment X week interaction was also tested, but was removed from the model, as it did not significantly predict missingness. All analyses were performed using SAS version 9.4. The significance level was set at .05.

3. Results

3.1. Medication adherence in the per-protocol sample

For the per-protocol sample, 42.6 % (n = 115) of the participants assigned to the BUP-NX condition (n = 270) received all of their planned prescriptions and 47.1 % (n = 96) of the participants assigned to the XR-NTX condition (n = 204) received all of their planned injections for the entire 24 weeks of the study. The mean number of buprenorphine prescriptions for the non-completers was 9.7 (standard deviation [SD] = 5.9), compared to 24.1 (SD = 2.3) in the completers sample. The mean number of XR-NTX injections among the non-completers was 2.2 (SD = 1.4), compared to 5.9 (SD = 0.7) in the completers sample.

3.2. Analysis of the opioid urine test results

Percentages of missing data in the per-protocol sample ranged from 11.9 % in week 3–63.2% in week 21 for the BUP-NX condition and from 19.6 % in week 3–59.8% in week 23 for the XR-NTX condition. The overall percentages of missing data for the BUP-NX condition (40.1 %) and the XR-NTX condition (40.2 %) were similar. Missingness was not significantly different between treatment groups (p = .95).

In both the per-protocol and the completers samples, the percentages of opioid-positive urine test results among the collected and analyzed specimens, were relatively low. The proportion of opioid-positive urine test results for both groups was less than 20 % for 21 of the 22 weeks that were analyzed, with the sole exception of week 24 (Fig. 1).

Fig. 1.

Observed percentages of opioid positive urine tests and model-estimated odds ratios, with 95 % CIs, for Buprenorphine-Naloxone (BUP-NX) vs. Extended-Release Naltrexone (XR-NTX) treatment conditions, among the per-protocol sample (N = 474).

*Indicates significantly different percentages of opioid positive urine tests at p < .05.

Note: The number of urine samples analyzed per week from weeks 3–24 for the XR-NTX participants were: 155, 156, 142, 140, 128, 128, 125, 116, 113, 105, 104, 103, 101, 98, 97, 93, 89, 91, 87, 85, 78 and 131. The number of urine samples analyzed per week from weeks 3–24 for the BUP-NX participants were: 237, 230, 198, 210, 189, 191, 159, 166, 141, 152, 135, 143, 127, 139, 117, 114, 115, 123, 99, 102, 103 and 183.

The treatment X week interaction was significant for the per-protocol sample. Contrasts at each week revealed significantly higher odds of opioid-positive urine tests in the BUP-NX sample compared to the XR-NTX sample in 14 out of the 22 weeks of the study (ps < .05; Fig. 1). The treatment X week interaction was not significant in the completers sample (p = .30, see Fig. 2).

Fig. 2.

Observed percentages of opioid positive urine tests and model-estimated odds ratios, with 95 % CIs, for Buprenorphine-Naloxone (BUP-NX) vs. Extended-Release Naltrexone (XR-NTX) treatment conditions, among the completers sample (N = 201).

*Indicates significantly different percentages of opioid positive urine tests at p < .05.

Note: The confidence interval for the observed proportion of positive opioid tests in the XR-NTX group were not estimated at week 21, as all urine opioid tests in XR-NTX group were negative.

Note: The number of urine samples analyzed per week from weeks 3–24 for the XR-NTX participants were: 86, 91, 85, 89, 83, 87, 89, 87, 87, 84, 86, 82, 83, 86, 84, 82, 79, 83, 81, 79, 72 and 89. The number of urine samples analyzed per week from weeks 3–24 for the BUP-NX participants were: 114, 113, 102, 112, 107, 113, 100, 109, 98, 113, 103, 110, 100, 112, 99, 98, 99, 111, 89, 94, 95 and 111.

4. Discussion

This secondary analysis compared detailed weekly proportions of urine opioid positive screens among participants in a multi-site, 24-week randomized clinical trial of XR-NTX vs. sublingual BUP-NX treatment initiated in inpatient drug treatment settings and continued in outpatient treatment (Lee et al., 2018). Data were reported for per-protocol (i. e., started on medication) and completer (i.e., received all BUP-NX prescriptions or all XR-NTX injections) samples.

The per-protocol sample provides a useful perspective because it represents a clinically relevant subgroup of patients who initiated medication. In contrast to the findings reported in the primary outcome paper for the per-protocol analysis, in which the number of opioid negative urine samples was not significantly different between the two medication conditions (Lee et al., 2018), in our alternative analysis using all available data examined weekly, per-protocol participants assigned to BUP-NX had significantly greater odds of having opioid positive samples compared to participants assigned to XR-NTX throughout the majority of the weeks of treatment. This finding may reflect differences in analysis strategies and handling of missing data. For example, in the previous paper (Lee et al., 2018), the urine testing data were analyzed as an aggregate count of negative tests across 22 weeks, rather than examining each week longitudinally in the current analyses. Also, in the previous paper (Lee et al., 2018), missing urine specimens were treated as positive, while in the current analyses no assumptions were made about missing data with no missing urine tests imputed as positive. Therefore, the differing conclusions are likely based on the handling of missing data and perhaps greater power associated with longitudinal modeling.

Although XR-NTX had lower weekly likelihoods of opioid positive urine tests in the per-protocol sample, no such differences occurred in the completers sample. This suggests that adherence to medication is an important predictor of achieving opioid abstinence, regardless of whether it is XR-NTX or BUP-NX. The relatively high proportion of missing urine testing data in the per-protocol sample may have contributed to the finding that XR-NTX was more effective in reducing illicit opioid use than BUP-NX. Therefore, these results should be considered tentative.

Compared to the proportions of opioid-positive urine testing results in other randomized trials of BUP-NX or XR-NTX (Johnson et al., 2000; Lee et al., 2016; Ling et al., 1998, 1996) in the US, the proportions observed in the current study were lower, with the majority of weekly positives at 20 % or below for the per-protocol sample and 10 % or below for the completers sample. These differences among the studies could be attributed in part to the differing analytic methods (e.g., maximum likelihood estimation) and accompanying assumptions (e.g., not treating missing urine tests as positive). Treating missing opioid urine test results as positive would raise the proportion of positive results, particularly in studies with considerable amounts of missing urine specimens. Importantly, in contrast to the prior research, the present study was initiated on inpatient detoxification treatment units rather than outpatient services. Thus, caution is required in comparing outcomes of those studies conducted in different treatment settings. It is also possible that there were differences in participant samples across these studies as well as differences in the illicit drug market among these studies conducted over the arc of the past 20 years.

Although urine drug testing results are typically used to examine participant outcomes, there are several approaches to analyzing such results. In early trials of buprenorphine treatment, urine opioid testing results were measured using the raw number of negative drug tests during the trial, an approach termed, “Treatment Effectiveness Score” (Ling et al., 1998, 1996). A second approach used by Ling and colleagues (Ling et al., 1998) reports a percent of negative urine tests that omits missing data from the numerator and denominator. A third approach (Ling et al., 1998) used in a trial comparing methadone versus buprenorphine, measured the percentage of opioid-negative tests by dividing the number of negative urine tests by the total number of urine specimens that could have been collected over the course of the study; thereby treating missing samples the same as positive. A fourth approach, used by Johnson and colleagues (Johnson et al., 2000) in a trial comparing methadone, buprenorphine, and LAAM, analyzed the weekly percentage of urine-positive tests with missing data considered positive. Lee and colleagues (Lee et al., 2016), also used this approach in a randomized clinical trial comparing XR-NTX to treatment-as-usual among individuals involved in the criminal justice system. Considering all of the different ways urine testing results are presented in the literature, direct comparisons among studies can be challenging. Therefore, future research should focus on harmonizing these measures and creating a standard for conducting sensitivity analyses to report on missing data.

The present analyses are complementary to the initial X:BOT study outcomes paper (Lee et al., 2018) because they provide an in-depth examination of the serial nature of illicit opioid use as measured by weekly urine testing data both in a sample that was able to begin medication including those who subsequently dropped out (per-protocol analysis) and in an adherent sample that completed treatment (completers analysis). While the previous paper analyzed time to relapse as the primary outcome (defined as seven consecutive days of self-reported opioid use or four consecutive weeks of self-report and/or positive urine testing results) and the total number of negative urine tests as a secondary outcome, the current paper examined weekly opioid test results, which lends a different perspective. This strategy enables the analysis of more data points and to examine differential change over time. This approach allows for the potential for intermittent (on again, off again) opioid use and is a strength of the current GLiMM analytic strategy. GLiMM analyses also allow for clustering with random variables, such as site with the eight sites theoretically having been selected from a larger population of sites to which the findings could generalize. This clustering takes into account that error terms may be correlated given that groups of participants were associated with particular sites.

The seemingly contradictory opioid test findings from the X-BOT trial can also be explained in part by the pharmacologic properties of XR-NTX and BUP-NX. For example, the intent-to-treat analysis conducted by Lee and colleagues (2018) found that participants assigned to BUP-NX compared with XR-NTX had lower rates of opioid-positive urine tests. This finding was attributed to the “detox hurdle” associated with XR-NTX, for which opioid antagonist properties necessitate about one week of opioid abstinence prior to initiation. This hurdle led to considerable drop-out prior to initiating XR-NTX compared to BUP-NX. In contrast, the present per protocol analysis found significantly higher rates of opioid-positive tests in 14 of the 22 weeks assessed in the BUP-NX compared to XR-NTX groups. This finding may be attributed to the complete blockade of illicit opioids by XR-NTX, which can lead to “testing” of the blockade with eventual suppression of illicit opioid use, in contrast to incomplete attenuation of the reinforcing effects of illicit opioids associated with buprenorphine treatment (particularly if buprenorphine is not taken consistently).

Despite the strengths of the GLiMM method, there are limitations associated with the present analyses. First, the missing at random assumption of mixed models is untestable, and, conceptually, there may be reason to assume that the missing urine tests between the two medications would not follow a similar distribution as observed. The requirement of a delay in starting XR-NTX following an inpatient treatment admission, compared to the ability to initiate BUP-NX immediately during an inpatient or outpatient admission, makes direct comparisons between these two medications challenging to interpret. In the per protocol analysis, the participant samples were not created randomly but based on whether participants initiated treatment (94 % of the BUP-NX participants vs. 73 % of the XR-NTX participants). Thus, it is possible that the participants who successfully initiated XR-NTX can be characterized, at least in part, by greater motivation. Additionally, for the purposes of the completers’ analysis, participants who received all of their buprenorphine prescriptions were considered completers – although they may not have adhered reliably to taking their medication. In contrast, receiving an injection of XR-NTX guarantees exposure to the medication for a month. There was a relatively high rate of missing urine testing data toward the end of the trial, with the exception of the final week in which participants received a higher payment for attending the last follow-up visit. During that final week, there were no significant differences in opioid-positive urine tests, a finding which could be explained by the more diverse sample of specimens collected. Obtaining weekly urine drug specimens over a 24-week period is quite challenging. It is clear that follow-up rates are strongly influenced by the level of reimbursement for the visit, a factor that should be considered in future trials. Another limitation is that we did not control for the severity of opioid use disorder at baseline. Finally, results may not be generalizable to individuals with OUD who start treatment on an outpatient basis.

5. Conclusion

These analyses provide a complementary approach to the use of summary statistics such as the total number of negative urine tests over the course of a trial by showing trends in opioid use over time. Given that the per-protocol analysis found that participants who started on XR-NTX injections had significantly more weeks of opioid-negative tests than those who started on buprenorphine treatment, it suggests that difficulties with adherence to transmucosal buprenorphine might be addressed by long-acting medication buprenorphine formulations. One such extended-release injected buprenorphine formulation (Haight et al., 2019) has recently been FDA approved and marketed, and another (Lofwall et al., 2018) has been provisionally approved, but not yet marketed. These medications might be particularly useful if tailored to poorly adherent patients by assuring delivery of buprenorphine at a stable concentration for one month at a time. Alternatively, differences observed in the current study between XR-NTX and BUP-NX may be due to pharmacodynamic differences between agonist and antagonist interventions, rather than adherence to daily vs. monthly dosing. In regard to future clinical trials of treatments for opioid use disorder, these findings suggest analyzing patterns of opioid use over time in addition to summary measures like relapse, or overall proportion of drug positive weeks. Results showed that both treatments were effective. Differences in opioid positive tests were small and of uncertain clinical significance due to the relatively high proportion of missing data.