Emil Coman, Ph.D. 1 & Peg Weeks, Ph.D. 2 Challenges in assessing impact of an intervention to reduce HIV when comparison data is not available. Latent Growth Mixture Modeling (GMM) to recover classes of HIV risk groups users with differential female condom adoption and use patterns Emil Coman, Ph.D. 1 & Peg Weeks, Ph.D. 2 1- Senior Research Scientist ; 2 - Senior Scientist & Executive Director; Institute for Community Research, Hartford, CT USA Statement of problems Female Condom Study Intervention conducted in Hartford, CT aimed at increasing awareness and use of the female condom (FC) as a women-initiated HIV and sexually transmitted infections (STI) prevention method. Prospective, longitudinal cohort study of 461 women (baseline, 1 & 10 months) Community-based interventions in non-experimental framework, when no control or matched comparison group is available by design: one cannot link changes to the intervention itself. ~ Causal inferences regarding changes can still be made however by ‘un-mixing’ the intervention group. Comparisons There seem to be three major latent groups of women that respond differently to potential community-based awareness and promotion FC campaigns.: One ‘fast FC adopters’ (N=29), a second made up of ‘interested/increasing users of FC‘ (N=125) and a third of ’non-adopters’/laggards (N=74). These 3 groups differ significantly in terms of : 1. FC stages of use (never tried, thought about using it but not used it, tried but decided not to use, use F occasionally, use FC regularly). 2. FC beliefs/attitudes (positive/negative attitudes towards FC) 3. Education (high school graduates are already interested, more than expected). 4. Marital status: more singles among laggards, separated/divorced are already interested. 5. Unprotected sex at 1 month and 10 months, but not at baseline. The groups did not differ on: Knowledge about FC Homelessness Ever tested positive for HIV CONCLUSIONS More research is needed to identify the potential FC adopter groups and specific characteristics (like barriers) before interventions are launched. Tailored interventions should address different potential responses to interventions by latent groups of community members. Mixture modeling can uncover differential growth processes and nuanced responses to community interventions. Count measures are can be analyzed as both dichotomies (absence/presence) and quantity (count once present); sometimes simple count would suffice. Female Condom (FC) use can be successfully promoted as alternative women-initiated HIV prevention method in community-based settings. Analyses and results 1. There is a decrease in ‘non-users’ and a marked increase in FC use 2. The one-group model shows that there is a clear increase in FC use over time. The model seems to fit in terms of reproducing the observed means HOWEVER, there is clear variation in trajectories of change between participants (a sample of 20 below): Latent Growth Model estimated and observed means of the zero inflated count outcome "Times used Female Condom in the past 30 days‘ (N=228) Conclusion: We need to explore differential trajectories of development. Comparing models We looked primarily at the ‘times used FC in the past 30 days’ outcome. Count data can be modeled as Zero Inflated Poisson (ZIP) growth curve or s a two-part Poisson model (‘hurdle’): a transition stage followed by an events stage. Once the hurdle is crossed, a nonzero count is observed. The causal structure of the “none versus some” part of the measure may be different than that of the “how much given some” part1. ‘interested/increasing users of FC‘ (N=125) ‘fast FC adopters’ (N=29), The 2 group model fits worse than the 1 group, and the 3 group model fits best (aBIC=2647.3). The 4 group model did not converge successfully. ’non-adopters’/ laggards (N=74) References 1. Liu, H., & Powers, D. A. (2007). Growth curve models for zero-inflated count data: An application to smoking behavior. Structural Equation Modeling, 14(2), 247. 2. Muthén, B. (1998). Second-generation structural equation modeling with a combination of categorical and continuous latent variables: New opportunities for latent class/latent growth modeling. In A. Sayer & L. Collins (Eds.), New methods for the analysis of change (pp. 291-322). Washington DC: American Psychological Association. Model comparisons: A two-part ZIP model fit worse than a one-part ZIP model: we do not need to model the transition from ‘no FC use’ separately from the ‘change in FC use’, once use is initiated. Acknowledgment We are indebted to Dave Kenny (Uconn) for introducing the first author to the world of causal modeling (SEM), for extensive generous discussions, and for reviewing many of the 1st author’s ideas. This research was funded by a grant from NIMH (1 R01 MH069088) awarded to the second author as Principal Investigator.