Advertisement

Using Latent Profile Analysis and Related Approaches in Adolescent Health Research

  • Devon J. Hensel
    Affiliations
    Division of Adolescent Medicine, Department of Pediatrics, Indiana University/Purdue University Indianapolis, Indianapolis, Indiana

    Department of Sociology, Indiana University/Purdue University Indianapolis, Indianapolis, Indiana
    Search for articles by this author
      See Related Article on p.194
      In this issue of Journal of Adolescent Health, Giano et al. [
      • Giano Z.
      • Currin J.
      • Deboy K.
      • Hubach R.
      Identifying distinct subgroups of lesbian, gay, and bisexual youth for suicide risk: A latent profile Analysis.
      ] use latent profile analysis (LPA) to examine how risk factors differentially combine to impact suicide risk among lesbian, gay, and bisexual youth. This article joins a fast-growing body of literature using statistical approaches to understand variability in adolescent health experiences. This editorial will focus on providing a brief review of how and when health professions researchers might use LPA or related methods in their own work [
      • Hensel D.J.
      Supporting adolescent sexual and reproductive health rights through innovative research approaches.
      ,
      • Hensel D.J.
      Concurrently advancing sexual rights and next-generation sexually transmitted infection prevention through innovative analytical methods.
      ].
      LPA is one method in a family of statistical approaches called mixture modeling, which has been described in excellent lay-focused reviews by Lanza and Cooper [
      • Lanza S.T.
      • Cooper B.R.
      Latent class analysis for developmental research.
      ] Oberski, [
      • Oberski D.
      Mixture models: Latent profile and latent class analysis. Modern statistical methods for HCI.
      ] and Williams and Kibowski [
      • Williams G.A.
      • Kibowski F.
      Latent class analysis and latent profile analysis.
      ]. These methods focus on uncovering how adolescents and young adults can be similarly “grouped” or “profiled” on a combination of different variables. The idea of looking for “sets” of related factors is not a new idea in adolescent health. Researchers often want to know how the intersection of social, interpersonal, and environmental/structural influences may have impact on health outcomes. However, the method that we choose to investigate these questions will influence our understanding of how these variable effects are interpreted in the population.
      Many people are familiar with variable-centered statistical approaches (e.g., multiple regression, factor analysis, and structural equation modeling) that focus on analyzing the relationship between a set of variables at the same time (e.g., if we entered multiple predictors into a regression model) [
      • Morin A.J.
      • Bujacz A.
      • Gagné M.
      Person-centered methodologies in the organizational sciences: Introduction to the feature topic.
      ]. An implication of this orientation is that we interpret a variable's average effect (e.g., using a regression coefficient's slope direction and magnitude) as being the same for all members of the population [
      • Collins L.M.
      • Lanza S.T.
      ,
      • Howard M.C.
      • Hoffman M.E.
      Variable-centered, person-centered, and person-specific approaches: Where theory meets the method.
      ]. Any programming effort that results from variable-centered approaches must also assume this similarity. For example, within the framework of the present study, if researchers had used a variable-centered approach, the results would likely be translated into recommendations that all lesbian, gay, bisexual, and transgender (LGBT) youth should receive the same suicide prevention programming content because the whole population would be assumed to have relatively similar levels of suicide risk and should therefore all respond in the same way to content.
      In many instances, this “one size fits all” assumption is not realistic. There are instances in which an aspect of a young person's experience (e.g., being a gender/sexual minority, being a person of color, and being incarcerated or being residentially unstable) and/or an intersection of these characteristics (e.g., being both a sexual minority youth and residentially unstable) can uniquely heighten an adolescent's vulnerability to (or protection from) an outcome [
      • Ghavami N.
      • Katsiaficas D.
      • Rogers L.O.
      Toward an intersectional approach in developmental science: The role of race, gender, sexual orientation, and immigrant status.
      ,
      • Jackson J.W.
      • Williams D.R.
      • VanderWeele T.J.
      Disparities at the intersection of marginalized groups.
      ]. This information allows for much more adaptable or tailored program delivery, as program design can focus on specific factors that are important for that group [
      • Lanza S.T.
      • Rhoades B.L.
      Latent class analysis: An alternative perspective on subgroup analysis in prevention and treatment.
      ]. Person-centered approaches such as LPA address these needs by identifying subgroups or subpopulations—here called “profiles”—that classify participants into discrete groups who show similarity on chosen measures [
      • Oberski D.
      Mixture models: Latent profile and latent class analysis. Modern statistical methods for HCI.
      ,
      • Williams G.A.
      • Kibowski F.
      Latent class analysis and latent profile analysis.
      ,
      • Howard M.C.
      • Hoffman M.E.
      Variable-centered, person-centered, and person-specific approaches: Where theory meets the method.
      ].
      LPA provides two key pieces of information that allow us to better understand unique health experiences within a person-centered approach, which are (1) profile characteristics and (2) profile link to outcome [
      • Hensel D.J.
      Supporting adolescent sexual and reproductive health rights through innovative research approaches.
      ,
      • Hensel D.J.
      Concurrently advancing sexual rights and next-generation sexually transmitted infection prevention through innovative analytical methods.
      ]. Profile characteristics identify both the optimal number of profiles in a population, which tell us the extent to which risk is heterogeneous (more classes) or homogeneous (fewer classes) in a population, and the likelihood of each variable being associated with a given profile, which helps us derive the substantive “meaning” of each class. For example, Giano et al. [
      • Giano Z.
      • Currin J.
      • Deboy K.
      • Hubach R.
      Identifying distinct subgroups of lesbian, gay, and bisexual youth for suicide risk: A latent profile Analysis.
      ] identified six unique profiles of risk factors for LGBT youth based on Youth Risk Behavior Survey measures of sleep, being bullied, alcohol use, poor grades, and electronics use. Class 6 was associated with the “highest risk” because of its association with the highest scores in all risk factors, whereas four other classes had mixed levels of risk based on their association of high- and low-risk score factors. Profile link to outcome provides information about how profile membership links to a higher or lower likelihood of experiencing the outcome variable of interest. Here, Giano et al. [
      • Giano Z.
      • Currin J.
      • Deboy K.
      • Hubach R.
      Identifying distinct subgroups of lesbian, gay, and bisexual youth for suicide risk: A latent profile Analysis.
      ] establish that both the highest risk class (Class 6) and one of the mixed risk classes (Class 3) had a higher risk of suicide compared with the lowest risk class (Class 2). Such analyses permit a more focused youth-centered means of identifying what types of adversity to target to reduce suicide risk in this population. By using a person-centered (rather than variable-centered) analytic approach, the results can inform suicide prevention programs in ways that allow the development of more nuanced strategies for addressing the needs of LGBT youth based on profiles of varying risk.
      Although an exhaustive review of LPA mechanics is beyond the scope of this work (earlier recommended citations provide excellent overviews), it is useful to briefly compare related methods and discuss when each application is used. The mixture modeling family include four primary “flavors,” including LPA, latent class analysis (LCA), growth mixture modeling (GMM), and latent transition analysis (LTA). Both LPA and LCA seek discovery of unobserved “profiles” or “classes” (as described previously) when data are cross-sectional, although LPA is used when variables are continuous and LCA is used when variables are categorical. GMM and LTA are used when data are longitudinal; GMM is used with continuous data and when research questions focus on identifying group-based differences in change over time, whereas LTA uses categorical data to examine how profile or class membership shift over time [
      • Lanza S.T.
      • Cooper B.R.
      Latent class analysis for developmental research.
      ]. All applications are flexible to the addition of proximal and distal variable additions and use a series of model fit criterion to choose the “best” option when multiple models are being compared [
      • Williams G.A.
      • Kibowski F.
      Latent class analysis and latent profile analysis.
      ,
      • Nylund K.L.
      • Asparouhov T.
      • Muthén B.O.
      Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study.
      ,
      • Zhou M.
      • Thayer W.M.
      • Bridges J.F.P.
      Using latent class Analysis to model preference heterogeneity in health: A systematic review.
      ]. As reviewed here, and in other publications [
      • Lanza S.T.
      • Cooper B.R.
      Latent class analysis for developmental research.
      ,
      • Oberski D.
      Mixture models: Latent profile and latent class analysis. Modern statistical methods for HCI.
      ,
      • Williams G.A.
      • Kibowski F.
      Latent class analysis and latent profile analysis.
      ], with careful attention to data structure and research interests, these four methods provide a powerful tool to better understand nuances in adolescent and young adult health.

      References

        • Giano Z.
        • Currin J.
        • Deboy K.
        • Hubach R.
        Identifying distinct subgroups of lesbian, gay, and bisexual youth for suicide risk: A latent profile Analysis.
        J Adolesc Health. 2020; 67: 196-200
        • Hensel D.J.
        Supporting adolescent sexual and reproductive health rights through innovative research approaches.
        J Adolesc Health. 2019; 64: 288-289
        • Hensel D.J.
        Concurrently advancing sexual rights and next-generation sexually transmitted infection prevention through innovative analytical methods.
        Sex Transm Dis. 2020; 47: 177-178
        • Lanza S.T.
        • Cooper B.R.
        Latent class analysis for developmental research.
        Child Dev Perspect. 2016; 10: 59-64
        • Oberski D.
        Mixture models: Latent profile and latent class analysis. Modern statistical methods for HCI.
        Springer, Basel, Switzerland2016: 275-287
        • Williams G.A.
        • Kibowski F.
        Latent class analysis and latent profile analysis.
        Handbook of methodological approaches to community-based research: Qualitative, quantitative, and mixed methods. Oxford University Press, Oxford, United Kingdom2016: 143-151
        • Morin A.J.
        • Bujacz A.
        • Gagné M.
        Person-centered methodologies in the organizational sciences: Introduction to the feature topic.
        SAGE Publications Sage CA, Los Angeles, CA2018
        • Collins L.M.
        • Lanza S.T.
        Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. Vol. 718. John Wiley & Sons, Hoboken, NJ2010
        • Howard M.C.
        • Hoffman M.E.
        Variable-centered, person-centered, and person-specific approaches: Where theory meets the method.
        Org Res Method. 2018; 21: 846-876
        • Ghavami N.
        • Katsiaficas D.
        • Rogers L.O.
        Toward an intersectional approach in developmental science: The role of race, gender, sexual orientation, and immigrant status.
        in: Advances in Child Development and Behavior. Vol. 50. Elsevier Academic Press, Cambridge, MA2016: 31-73
        • Jackson J.W.
        • Williams D.R.
        • VanderWeele T.J.
        Disparities at the intersection of marginalized groups.
        Soc Psychiatry Psychiatr Epidemiol. 2016; 51: 1349-1359
        • Lanza S.T.
        • Rhoades B.L.
        Latent class analysis: An alternative perspective on subgroup analysis in prevention and treatment.
        Prev Sci. 2013; 14: 157-168
        • Nylund K.L.
        • Asparouhov T.
        • Muthén B.O.
        Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study.
        Struct equation Model A Multidisciplinary J. 2007; 14: 535-569
        • Zhou M.
        • Thayer W.M.
        • Bridges J.F.P.
        Using latent class Analysis to model preference heterogeneity in health: A systematic review.
        Pharmacoeconomics. 2018; 36: 175-187

      Linked Article