Differential Misclassification of Overdose Safety of in a Prominent Comparative Effectiveness RCT Comparing Buprenorphine and Naltrexone for Opioid Dependence

Background: X:BOT is an influential RCT published in The Lancet comparing safety and efficacy of buprenorphine and naltrexone for opioid dependence. The original publication reported no statistically significant difference in safety endpoint of overdose (OD) using Fisher's exact test. ODs occur when patients try to overcome naltrexone's opioid receptor antagonism by taking higher doses of street drugs. Epidemiology studies find naltrexone less protective against OD than buprenorphine. Yet, evidentiary strength of X:BOT RCT has led to US drug policy giving full parity to the 2 treatments.

Objectives: Replicate findings while accounting for differential induction failures. Explore timing & coding of OD

Methods: Multi-site, 24-week, open-label comparative effectiveness RCT
Reanalysis using public dataset
Tx: sublingual buprenorphine (bup) vs. ER injectable naltrexone (ntx)
Outcome: time to overdose
Stats: Survival analysis, using per protocol and ITT approaches, with novel inverse probability of induction failure weighting (IPIW)

Results: We replicated most findings. However, we identified 6 *additional* OD events that had been missed: 5 on naltrexone, 1 on buprenorphine. With misclassification corrected, 19/283 participants had ODs in ntx arm, 9/287 for bup; HR 2·40 (95% CI 1·09, 5·30; Wald p=0.03). When presented with our analysis the original authors conceded overdoses were missed. The original study identified cases through verbatim match for "overdose" MedDRA PT. However, it did not tally the term "overdose" when it appeared in 2 associated columns describing AE etiology. Cases with clinically described cause of "Opioid Overdose" were miscoded as "Respiratory depression" and not counted; a case with etiology of "NARCO OVERDOSE" was miscoded to PT "Drug abuse" and similarly excluded.

Post-randomization induction failures were differentially distributed: 28% (ntx) vs 5·9% (bup). Focusing on the maintenance period (study days 22-168, after induction and before therapy discontinuation), when most overdoses occurred, we found ITT HR 3·77 (95% CI 1·23, 11·57; Wald p=0·02), and per protocol HR 3·81 (95% CI 1·01, 14·36; p=0·05). X:BOT shows higher risk of OD with naltrexone than buprenorphine even during therapy. IPIW method accounting for differential induction failure is complete, and will be presented once peer reviewed in time for ICPE; results were consistent.

Conclusions: Naltrexone should not be presented as the only option for opioid dependence treatment, as is common in prisons and drug courts. Hampered by outcome misclassification, induction failures, & a weak statistical test, the original study made 4 claims that OD "did not differ" between arms. A retraction/correction has been requested; this presentation will describe the response and follow-up.