Improved classification of alcohol intake groups in the Intermittent-Access Two-Bottle choice rat model using a latent class linear mixed model

Alcohol use disorder (AUD) is a major public health problem in which preclinical models allow the study of AUD development, phenotypes, and the exploration of potential new treatments. The intermittent access two-bottle choice (IA2BC) model is a validated preclinical model for studying alcohol intake patterns similar to human AUD clinical studies. Typically, the mean/median of overall alcohol intake or the last drinking sessions is used as a threshold to divide groups of animals into high or low alcohol consumers. Nevertheless, this approach has the potential for introducing bias due to the a priori selection of a threshold, as opposed to measuring the consumption drinking pattern along the protocol and subgrouping accordingly. This study aimed to assess the efficacy of utilizing longitudinal data of all drinking sessions to classify the population into high or low alcohol intake groups, employing a latent class linear mixed model (LCLMM). We compared LCLMM with traditional classification methods: (i) percentiles, (ii) K-means clustering, and (iii) hierarchical clustering. In addition, we used simulations to compare the accuracy, specificity, and sensitivity of these methods. By considering the entire trajectory of alcohol intake, LCLMM provides a more robust classification based on accuracy (0.94) between high and low alcohol classes. We recommend the use of longitudinal statistical models in research on substance use disorders in preclinical studies, since they could improve the classification of subpopulations.