Modeling multidimensional sexual risk behavior using latent class analysis

Understanding the intersection of various dimensions of sexual risk behavior in a population is critical for effective prevention. For example, frequency of sexual intercourse, the number of sexual partners, and inconsistent condom use are three dimensions of behavior that relate to the acquisition of sexually transmitted infections (STI's). Although estimating the differential risk posed by each dimension of behavior can be informative, taking a person-centered approach to modeling sexual risk behavior that incorporates multiple dimensions simultaneously can provide an intuitive and more complete picture of different profiles of behavior that are common in the population. Further, the identification of individual characteristics that predict membership in groups characterized by a profile of high-risk behavior can inform how to target intervention resources. Latent class analysis (LCA) is a latent variable model that can be used to identify risk profiles in empirical data. LCA measures an underlying, or latent, variable using a set of observed variables. The latent variable is made up of subgroups, or latent classes, that account for population heterogeneity. The variables measure dimensions of the latent classes. When applied to sexual risk behavior, each latent class represents a different multidimensional profile of risk. For example, Lanza and Collins (2008) used LCA to identify five latent classes of adolescents in the United States defined by common patterns of sexual risk behavior: Non-daters, Daters, Monogamous, Multipartner Safe, and Multipartner Exposed. A longitudinal extension of this approach, where transitions to more risky stages are estimated, can provide information on stability and change in sexual risk behavior profiles over time. Such a model would provide important insight about sexual risk behavior subgroups that are most at-risk of making a transition to high-risk behavior in the future. In addition, individual characteristics can be incorporated in the model to predict transitions to risky behavior. Identification of risk behavior profiles, modeling transitions between profiles over time, and predicting profile membership and transitions between profiles all have direct implications for prevention of HIV/AIDS and other STI's.