In medical research, latent subgroups often emerge with characteristics or trends distinct from the general population, yet identifying them directly remain challenging. The latent variable mixture modeling, grounded in the idea that a population consists of a limited mixture of subgroups, assigns latent categories to individuals based on posterior probabilities. This model is suitable for both cross-sectional and longitudinal datasets. Approaching from a statistical perspective, this paper thoroughly explicates the foundational principles of four prevalent methods within the latent variable mixture modeling realm, outlining the essential modeling workflow. By integrating insights from previous cases and real-world data, we review the rational applications of these methods. The latent variable mixture modeling stands as a flexible classification tool for identifying and analyzing latent categories within research populations, further facilitating the in-depth exploration of predictors influencing these latent categories and their consequent effects on outcome variables.