Predictors of successful initiation of buprenorphine in enriched enrollment randomized withdrawal clinical trials in both opioid experienced and naïve participants: a participant-level meta-analysis.

We reviewed all studies of buprenorphine transdermal and buccal patches submitted for approval to the Food and Drug Administration in both opioid-naïve and opioid-experienced pain available in electronic format. Enriched enrollment randomized withdrawal studies were used to evaluate the efficacy and safety 10 involving 2 phases: (1) the titration of all eligible patients onto the study drug followed by (2) randomization of all participants not dropping out for side effects or lack of efficacy, to continue the drug or be withdrawn to placebo. The population for the titration phase was all successfully screened participants who started the transition phase lasting 1 to 8 weeks (average 2 weeks). For the maintenance phase, we used those successfully randomized and maintained on the study drug. Each of several patient-reported outcomes (PROs) had to be modeled separately as they were not collected consistently across all studies. Titration phase success was defined as meeting the study-defined randomization criteria of tolerable side effects and efficacy (pain ≤ 4/10-Numerical Rating Scale) within buprenorphine transdermal doses from 5 to 20 μgs/h or buccal doses from 75 μg to 900 μg. Maintenance phase success was defined in the original study protocols as continuing treatment postrandomization to the end of the 12-week study while maintaining an average pain score ≤4/10 and that did not increase more than 2/10 from baseline. In addition, the drug efficacy of maintaining buprenorphine vs switching to placebo was graphed as (1) a Kaplan–Meier curve for using rescue or dropping out and (2) as the difference in the group average level of pain over time (results in Supplemental Content, http://links.lww.com/PR9/A310 ). The prediction method used mixed-effects logistic regression models, incorporating fixed-effect study-level intercepts and random-effect slopes. To control for between-study differences, models were adjusted for baseline covariates, accounting for the direct effect of the covariates on the outcome and on the other covariates. 4 The predictive accuracy of models was assessed using the area under the receiver operating characteristic curve (ROC-AUC) with the commonly used threshold of clinical relevance (≥0.70). The Akaike Information Criterion (AIC) prediction modeling process was conducted stepwise to balance goodness of fit with model parsimony. 15 The variable entry and exclusion process targeted minimizing (improving) the AIC. Age and sex were included to adjust for common demographic factors. All available baseline variables (Table 1 ) were included as candidates for the titration phase, and the smaller number of studies for the maintenance phase. For the PRO variables, models were fit to the population subsets that included the PRO (Table 2 ). There was little overlap between the defined PROs, which were analyzed separately. The total or subscale scores, as appropriate, were included in the modeling, using published methods for each PRO (see Supplemental content—Variable and Subject Inclusion, http://links.lww.com/PR9/A310 ). For both titration and maintenance assessment, prestudy pain was the level of pain experienced before entering the titration phase. Titration phase average baseline values (standard deviations) of available predictor variables. History of or treatment for certain chronic diseases with regular treatment regimens—any nonpain diseases that require regular medication that might affect the consistency of drug use for pain symptoms (eg, diabetes, hypertension, hypothyroidism…). ADHD, attention deficit hyperactivity disorder; GERD, gastroesophageal reflux disease; MOS SS, medical outcomes study sleep scale; NSAID, nonsteroidal anti-inflammatory drug; Rx History, prescription history; SF36, short form 36; WOMAC, Western Ontario and McMaster Universities Arthritis Index. Maintenance phase average baseline values (standard deviations) of available variables. *History of or treatment for certain chronic diseases with regular treatment regimens—any nonpainful diseases that require regular medication that might affect the consistency of drug use for pain symptoms (eg, diabetes, hypertension, hypothyroidism…). ADHD, attention deficit hyperactivity disorder; GERD, gastroesophageal reflux disease; MOS SS, medical outcomes study sleep scale; NSAID, nonsteroidal anti-inflammatory drug; Rx History, prescription history; SF36, short form 36; WOMAC, Western Ontario and McMaster Universities Arthritis Index.

Of 122 available studies of all types, we selected only the 10 EERW efficacy studies. These included only partients with osteoarthritis or chronic low back pain. For the maintenance phase, we used only the 5 EERW studies with a 12-week randomization phase and dosing assigned from the titration phase (Fig. 1 —Study Flow Diagram). The screening phase for eligibility included 10,275 patients, with 6052 eligible for the titration phase with an average age of 54 years, 58% female, 79% White, and 2573 (42.5%) opioid experienced with a mean prestudy opioid dose of 23.6 morphine milligram equivalents (MME—Table 1 ). From this group, 3541 (58.8%) successfully titrated with an average pain level of 2.9/10 on a buprenorphine transdermal dose from 5 to 20 μg/h or buccal dose from 75 μg to 900 μg. There were 58.1% opioid-naïve and 59.1% opioid-experienced participants. The average pretitration pain level was 6.5/10 (SD 1.59, n = 6052). Flow diagram of selected studies. For the maintenance phase analysis, only the 5 studies that continued participants on the dose achieved in the titration phase were used (n = 1796), with 877 participants continued on buprenorphine. The participant demographics included an average age of 51 years, 53% females, 67% White, and 297 (33.9%) opioid experienced with a mean opioid dose at study entry of 32.4 MME (Table 2 ). Overall, the average pretitration pain level was 6.0 (SD 1.65, n = 877) on entry and 2.9 (SD 1.15, n = 877) at the end of titration. From this group, 614 (70.0%) successfully completed the maintenance phase on buprenorphine, with 68.4% opioid naïve and 73.1% opioid experienced. The prediction model for all 6052 participants in the titration phase had a ROC-AUC of 0.61 with only higher baseline pain decreasing the likelihood of success (OR = 0.86) and any prior medical history of obesity (OR = 1.40) increasing the likelihood of success both with only moderate effects (Table 3 ). There were no important differences between models of (1) opioid naïve and opioid experienced, (2) transdermal and buccal patches, or (3) osteoarthritis and chronic low back pain participants. For the PRO-based models (Table 3 ), a higher (worse) brief pain inventory (BPI) (OR = 0.94) or subject opioid withdrawal score (SOWS) (OR = 0.98) decreased the likelihood of success. Table of estimated mean odds ratios (with 95% confidence interval) of patient baseline characteristics in the titration phase. The SF-36 score is reversed compared to other PROs with lower being worse. A variable that was a candidate variable but was not selected. —, A variable that was not a candidate variable (eg, because it was not collected in all studies); AUC, area under the curve of the receiver operating characteristic (ROC); ALL, the dataset includes data from all studies; BPI, the dataset from all studies that collected the brief pain inventory; CONMEDS, the dataset including data from all studies with study-collected ATC codes; SOWS, the dataset including data from all studies that collected subjective opioid withdrawal scale. None of the following scales were retained in models created for the subset of data containing the measures: COWS, the dataset from all studies that collected clinical opioid withdrawal scale; RDQ, the dataset from all studies that collected the Roland–Meyer Disability Questionnaire; MOS_SS, the dataset from all studies that collected the methods of study sleep scale at or before the start of the titration phase; WOMAC, the dataset from all studies that collected the Western Ontario and McMaster Universities Arthritis Index at or before the start of the titration phase. From the 5 available trials, we modeled the 877 participants that completed titration and continued on buprenorphine in the 12-week maintenance phase, generating a ROC-AUC = 0.62 (Table 4 ). Only the higher screening baseline pain scores predicted less likelihood of success. No PROs were retained in the respective subsets. The Supplemental Content, http://links.lww.com/PR9/A310 provides additional details. Table of estimated mean odds ratio (and 95% confidence interval) of patient baseline characteristics in the maintenance phase. None of the following scales were retained in models created for the subset of data containing the measures: CONMEDS, the dataset including data from all studies with study-collected ATC codes; MOS_SS, the dataset including data from all studies that collected the methods of study sleep scale at or before the start of the maintenance phase; BPI, the dataset including data from all studies that collected the brief pain inventory at or before the start of the maintenance phase; POMS, the dataset including data from all studies that collected the profile of mood states at or before the start of the maintenance phase; RDQ_COWS, the dataset including data from all studies that collected the Roland–Meyer Disability Questionnaire at or before the start of the maintenance phase, which is also the same set of studies as collected the Clinical Opioid Withdrawal Scale; SOWS, the dataset including data from all studies that collected subjective opioid withdrawal scale at or before the start of the maintenance phase; PGIC, the dataset including data from all studies that collected patient global impression of change at or before the start of the maintenance phase.

While it would be useful for clinicians to be able to predict the likely benefit from the initiation or transition to buprenorphine for pain, our results produced a predictive model not accurate enough to be of clinical benefit, with about 60% of participants enrolled being successfully titrated in both the opioid naïve and experienced groups. Slightly more of the opioid experienced (73.1% vs 68.4%) participants successfully completed the maintenance phase, but it was not statistically significant. The models for the titration from full agonist opioids to buprenorphine did include obesity and the BPI and SOWS PROs in the appropriate subsets. Although known to have some level of relationship to pain and the use of opioids, their potential role in the benefits of switching to buprenorphine had not previously been explored. Our results are not definitive with all AUC-ROC ≤ 0.70. The potential role for higher baseline pain levels in our models is also likely related to a specified cut-off of ≤4/10, with participants with pain values closer to that level being statistically more likely to cross the threshold independent of other factors. Overall, conclusions about all these factors must wait for additional research specifically targeting these issues. In the current literature, there are few reports of successful prediction models of patients likely to benefit from the initiation or transition to any opioid using standard demographics or commonly used PROs, and none that evaluate buprenorphine. One potentially useful paradigm (the DIRE model) also used extensive inquiry into the patient's diagnosis, past performance, reliability, and social supports, 1 all of which are standard parts of a full clinical evaluation. Models predicting response in cancer patients also have variable results in predicting the use of opioids but almost always include cancer type and location of pain 6 , 13 with more studies recommended. 13 Substantial limitations of our analysis must be considered. Importantly, our results are derived from a patient population willing to volunteer for a study in which they might be titrated to a placebo, which limits generalizability. In addition, these clinical trials did not measure the duration of pain or catastrophizing, which should be explored in future studies. While buprenorphine is considered useful in the treatment of pain patients with an OUD, the exclusion of such patients in our population prevents us from commenting on this issue. Specialized studies in this important population will be needed. In conclusion, our study suggests that it is unlikely that any patient-oriented baseline measures strongly predict the successful initiation or transition of chronic pain patients to buprenorphine in patients without known OUD or serious psychological conditions. As such, patients otherwise appropriate for a trial of buprenorphine therapy could be considered with careful monitoring of their progress and a plan for handling problems that may occur. Further research may consider additional factors, but it seems that measures beyond standard demographics and common PROSs may be necessary to predict the likelihood of successful treatment with buprenorphine.

Opioids have both beneficial and detrimental effects on people in pain. During 2021, 20.9% of US adults (51.6 million persons) experienced chronic pain, with 6.9% (17.1 million persons) severe enough to restrict daily activities. 11 Of the 20.9% with chronic pain, 22% used opioids in the last week. 2 Long-term observational studies provide evidence that up to 44% maintain good pain control on a constant level of opioids for up to 1 year 5 , 8 although randomized trials show nonopioids may perform as well as opioid therapy in many conditions. 7 Chronic pain treatment should start with nonopioid approaches, but if ineffective, opioids may be considered. To potentially reduce the risk of opioid use disorder (OUD) and opioid overdose, there has been a recent interest in considering buprenorphine rather than full agonist opioids, given its unique partial agonist pharmacology and use to treat OUD. 3 Meta-analytic reviews demonstrate consistent moderate efficacy. 9 , 14 To maximize benefit and minimize harm, it would be useful to identify characteristics predictive of patients who may achieve and maintain adequate benefit from buprenorphine. 12 Using a harmonized dataset from both the titration and maintenance phases of 10 enriched enrollment randomized withdrawal (EERW) trials of buprenorphine for noncancer-related chronic pain, we explored available baseline factors as potential predictors of analgesic response.

Supplemental digital content associated with this article can be found online at http://links.lww.com/PR9/A310 .

J.T.F. reports that over the past 3 years, he has received funding from NIH-NCATS—UL1 Grant (Co-I), NIH-NIDDK—U01 Grant (Co-I), from NIH-NINDS—U24 Grant (PI), and 2 FDA-BAA Contracts; and compensation for serving on advisory boards or consulting on clinical trial methods from Vertex, EicOsis, 3Daughters, Scilex Holding Company, and Lilly. He is the past President of the United States Association for the Study of Pain. W.B.B. received compensation for serving on advisory boards for Genentech. A.C. reports no conflicts. C.J.M. was previously a full-time employee and has served as an independent consultant for 3D Communications, LLC, a communications firm with many clients in the pharmaceutical, biotechnology, and medical device industries, including those developing analgesic and anesthetic products. C.E.A. received compensation for serving on advisory boards or consulting on clinical trials from Collegium, Teva, Pfizer, Kaleo, Daiichi Sankyo, and Astra Zeneca. R.B. reports no conflicts. J.G. reports that in the past 36 months, she has received consulting income from Algo Therapeutix, Eikonizo Therapeutics, Eli Lilly, GW Pharma, Hoba Therapeutics, and Saluda Medical. She owns vesting shares in Eisana Corp. J.H. reports funding from the NIH-NIAMS UH3 grant (Co-I) and that she serves as the Managing Director of the United States Association for the Study of Pain. I.G. has received support from Vertex, Eli Lilly Combigene, GW Research, Eupraxia, and Novaremed. He has received grants from the Canadian Institutes of Health Research, Physicians' Services Incorporated Foundation, and Queen's University. K.N.T. has received an investigator-initiated research grant from Bayer Healthcare LLC.