Translating medication effects for alcohol use disorder across preclinical, human laboratory, and clinical trial outcomes using meta-analysis

Abstract

Animal models are used for preliminary testing of novel compounds for alcohol use disorder (AUD). However, it is unclear whether early efficacy in preclinical models reliably predicts efficacy in human laboratory and clinical trials. We searched the literature for medications tested for AUD in preclinical models (i.e., two-bottle choice [2-BC] and operant reinstatement), human laboratory cue-reactivity, and randomized clinical trials (RCTs). For preclinical models, we computed medication effects on 2-BC alcohol preference and consumption (k = 77 studies, 14 medications) as well as operant reinstatement (k = 18 studies, 8 medications). For human laboratory studies, we computed medication effects on alcohol cue-induced craving (k = 36 studies, 15 medications). For RCTs, we computed medication effects on RCT endpoints including return to any drinking and return to heavy drinking (k = 139 studies, 19 medications). We used medication as the unit of analysis and applied the Williamson-York bivariate weighted least squares estimation to preserve the errors in both the independent and dependent variables. Medication effects on 2-BC alcohol preference ( $\hat{\beta }$ = 0.04, p = 0.004) and reinstatement ( $\hat{\beta }$ = 0.20, p = 0.05) were positively associated with medication effects on return to any drinking in human clinical trials but no associations were found on other RCT outcomes tested. Preclinical medications findings were not associated with medication effects on cue-induced craving in the human laboratory. Medication findings on 2-BC alcohol preference and operant reinstatement track medication effects on select clinical trial outcomes, specifically return to any drinking. This study provides empirical support for the association between medication effects across species and experimental models, a critical, yet untested, premise of preclinical studies in medications development.

Fig. 1. PRISMA flow diagram for systematic review of alcohol use disorder medication studies across different experimental paradigms.

Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) flow diagram of preclinical (a), cue-reactivity in the human laboratory (b), and randomized clinical trial records (c).

Fig. 2. The linear relationship between medication effect sizes on preclinical endpoints and medication effect sizes on percent of participants who return to any drinking in randomized clinical trials.

Williamson-York regression models with preclinical (2-BC alcohol consumption [a], 2-BC alcohol preference [b], and reinstatement [c]; x-axis) medication effect sizes predicting medication effect on the return to any drinking RCT endpoint; y-axis). Each medication appears as a dot on the regression line, with smaller dots indicating more error variance and larger dots indicating less error variance around each estimate. Negative effect sizes represent a favorable medication effect relative to placebo/active control.

Fig. 3. The linear relationship between medication effect sizes on preclinical endpoints and medication effect sizes on percent of participants who return to heavy drinking in randomized clinical trials.

Williamson-York regression models with preclinical (2-BC alcohol consumption [a], 2-BC alcohol preference [b], and reinstatement [c]; x-axis) medication effect sizes predicting medication effect on the return to heavy drinking RCT endpoint; y-axis). Each medication appears as a dot on the regression line, with smaller dots indicating more error variance and larger dots indicating less error variance around each estimate. Negative effect sizes represent a favorable medication effect relative to placebo/active control.