A Method for Clustering Flexible Longitudinal Trajectories with An Application to An HIV Prevention Trial

Abstract

Dapivirine (25 mg) in a silicone elastomer Vaginal Ring is a safe and effective method to prevent HIV-1 infection in healthy, sexually active, HIV-negative women. Understanding women’s adherence to the dapivirine vaginal rings over time and exploring factors that might be associated with different patterns of adherence are helpful for supporting women who use the rings and achieving better levels of protection against HIV-1. In this analysis, we propose a trajectory clustering method that performs K-means clustering with generalized additive models implemented to produce cluster means. We identified six different adherence trajectories, using data from 1,356 HIV-uninfected women participated in the HIV open-label prevention extension (MTN-025/HOPE) trial. While most of the participants were clustered as high adherence (18% as high adherence 1; 18% as high adherence 2) and medium adherence (24%), we also identified declining adherence trajectories (15% as high – declining adherence; 13% as declining adherence) and a non-adherence trajectory (12%). Baseline characteristics including non-oral contraceptive use and lack of menstrual bleeding in the past 3 months were found to be associated with being clustered as high or medium adherence. Being clustered as declining adherence was found to be associated with having experienced menstrual bleeding in the past 3 months and non-use of injectable contraceptive. Among participants who showed high adherence, contraceptive choice of intrauterine device was found to be correlated with being clustered as high adherence 2.