Model selection in linear mixed-effects analysis of craving outcomes in a randomized controlled trial of medications for opioid use disorders

Identifying an appropriate covariance structure is a critical step in modeling longitudinal changes in polysubstance cravings during medications for opioid use disorders (MOUD) treatment. This study analyzed data from the National Drug Abuse Treatment Clinical Trials Network trial CTN0051, in which participants with opioid use disorders (OUDs) were randomized to receive either buprenorphine-naloxone (BUP-NX; n = 287) or extended-release naltrexone (XR-NTX; n = 283). Cravings for opioids, alcohol, stimulants, and nicotine were assessed at baseline and every four weeks over 36 weeks.

Repeated-measures linear mixed models (LMMs) were used to examine longitudinal craving trajectories, with treatment, visit, and treatment-by-visit interaction specified as fixed effects and within-participant correlations among repeated observations modeled through the residual (R-side) covariance structure. Ten commonly used covariance structures were compared using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to determine optimal model fit. AIC-based comparisons indicated that the unstructured (UN) covariance structure provided the best fit for alcohol, opioid, and stimulant cravings and the second-best fit for nicotine craving.

Using the UN covariance structure, craving scores decreased significantly from baseline across all four craving domains during follow-up. Treatment-by-visit interactions were observed for stimulant craving at selected visits, indicating differences between treatment groups in changes from baseline over time. Baseline alcohol and amphetamine use disorders predicted higher alcohol cravings, while baseline depression, amphetamine, and cocaine use disorders predicted greater stimulant cravings; baseline cocaine use disorder also predicted elevated nicotine craving.

In conclusion, the UN covariance structure most effectively captured craving trajectories. These findings identify clinical characteristics associated with differences in reported craving; however, their utility for guiding treatment selection or predicting relapse requires prospective validation. First-order autoregressive moving average Centers for Disease Control and Prevention The information reported here results from secondary analyses of data from NIDA CTN-0051 (Extended-Release Naltrexone vs.

Buprenorphine for Opioid Treatment), a clinical trial conducted by the NIDA Clinical Trials Network. NIDA databases and information are available at https://datashare.nida.nih.gov. No funding source is given for the present paper.

Department of Biobehavioral Health & Nursing Science, College of Nursing, University of South Carolina, 1601 Greene St, Columbia, 29208, SC, USA Kesheng Wang, Donna M. Raynor Department of Epidemiology and Biostatistics, Arnold School of Public Health, University of South Carolina, 921 Assembly St, Columbia, 29208, SC, USA The authors declare no competing interests. The original CTN-0051 study (Extended-Release Naltrexone vs.