Using Latent Profile Analysis as a Unique Way to Visualize and Analyze Data with Illustration Using

Goal: It is important to identify subgroups at high risk of suicide as they are targets for intervention. Problem: Standard regression approaches evaluate individual covariates, but not combinations of covariates. For example, when using a standard regression model, individual predictors produce statistical estimates that are averaged across all study participants (e.g., “across all sexual minority youth in the sample, those who are bullied are twice as likely to exhibit suicide ideation”). While estimates like these are helpful in describing the overall landscape, in many instances this “one size fits all” assumption is not realistic, as there may be certain sub-groups of individuals where these estimates are stronger/weaker. Solution: Latent Profile Analysis (LPA) is one method in a family of statistical approaches called “mixture modeling.” These methods focus on uncovering how groups of individuals can be similarly “clustered” or “profiled” on a combination of risk factors. In this presentation we demonstrate the usefulness of LPA using data from the Youth Behavioral Risk-Factor Survey. Using LPA, sexual minority adolescents were categorized into “profiles” using 5 independent variables as risk factors (bullying, alcohol use, poor grades, electronics use, and sleep hours) which can then be associated with a single dependent variable: suicide risk. Interpretation and methodological considerations (particularly, the benefits of LPA beyond standard regression approaches) will also be discussed.