Sex differences in factors predicting post-treatment opioid use

Abstract

Background and aims: Several reports have documented risk factors for opioid use following treatment discharge, yet few have assessed sex differences, and no study has assessed risk using contemporary machine learning approaches. The goal of the present paper was to inform treatments for opioid use disorder (OUD) by exploring individual factors for each sex that are most strongly associated with opioid use following treatment.

Design: Secondary analysis of Global Appraisal of Individual Needs (GAIN) database with follow-ups at 3, 6 and 12 months post-OUD treatment discharge, exploring demographic, psychological and behavioral variables that predict post-treatment opioid use.

Setting: One hundred and thity-seven treatment sites across the United States.

Participants: Adolescents (26.9%), young adults (40.8%) and adults (32.3%) in treatment for OUD. The sample (n = 1,126) was 54.9% male, 66.1% white, 20% Hispanic, 9.8% multi-race/ethnicity, 2.8% African American and 1.3% other.

Measurement: Primary outcome was latency to opioid use over 1 year following treatment admission.

Results: For women, regularized Cox regression indicated that greater withdrawal symptoms [hazard ratio (HR) = 1.31], younger age (HR = 0.88), prior substance use disorder (SUD) treatment (HR = 1.11) and treatment resistance (HR = 1.11) presented the largest hazard for post-treatment opioid use, while a random survival forest identified and ranked substance use problems [variable importance (VI) = 0.007], criminal justice involvement (VI = 0.006), younger age (VI = 0.005) and greater withdrawal symptoms (VI = 0.004) as the greatest risk factors. For men, Cox regression indicated greater conduct disorder symptoms (HR = 1.34), younger age (HR = 0.76) and multiple SUDs (HR = 1.27) were most strongly associated with post-treatment opioid use, while a random survival forests ranked younger age (VI = 0.023), greater conduct disorder symptoms (VI = 0.010), having multiple substance use disorders (VI = 0.010) and criminal justice involvement (VI = 0.006) as the greatest risk factors.

Conclusion: Risk factors for relapse to opioid use following opioid use disorder treatment appear to be, for women, greater substance use problems and withdrawal symptoms and, for men, younger age and histories of conduct disorder and multiple substance use disorder.

Keywords: Adolescents; machine learning; opioid use disorder; opioids; risk factors; trauma; treatment.

Figure 1

Overall return to opioid use suivival curves for men and women (solid lines), with confidence intervals (dotted lines)

Figure 2

Hazard ratios (LASSO) and rank order variable importance (random forest model) for men (a, top), and women (b, bottom). PTSD = post traumatic stress disorder; SUD = substance use disorder

Figure 3

Prototypical survival curves for combinations of the top three predictors of latency of return to opioid use from the random forest variable importance models for men (a, top) and women (b, bottom). Among women, those with the quickest return to opioid use are adolescent/young adult women, with low levels of criminal justice involvement (this was a protective factor for women in our models) and at least moderate levels of withdrawal symptoms. Those with the lowest risk of return to use are women aged 26+ with high levels of criminal justice involvement and no withdrawal symptoms. For men, those with the quickest return to use are adolescents/young adults who endorse high levels of conduct disorders with fewer substance use disorder (SUD) diagnoses. For men, the group with the lowest risk are adults aged 26+ with low conduct disorder symptoms and multiple substance use disorders. These are just two examples; survival plots can be created for any combination of factors from the random forest models, thus providing an individualized risk profile for each person