Trajectory analysis with attrition weights using a finite mixture model and Bayesian framework

Music City Center 

Numerous approaches for trajectory analyses are currently available. None, to our knowledge, are flexible enough or able to incorporate weights that often accompany survey data. Thus, we introduce a novel approach for trajectory analysis for longitudinal survey data with a continuous outcome while incorporating attrition weights using a finite mixture model. Under the Bayesian framework, we use the birth-death process for clustering and allow trajectories to follow linear, quadratic, and cubic patterns. Simulation studies were used to evaluate the model's performance, and the results show high sensitivity, specificity, and accuracy with respect to clustering assignment. We applied the approach to the Together 5,000 (T5K) study, a U.S. nationwide, internet-based cohort with a goal to identify modifiable factors associated with HIV acquisition and PrEP uptake among HIV-vulnerable populations. Participants completed surveys and HIV tests at baseline and following four years. Attrition weights for each follow-up have previously been calculated. We identified different patterns of drug use measured by ASSIST scores over 4 years while accounting for participant attrition.

Weighted trajectory analysis

Attrition weights

Finite mixture model

Bayesian 

Survey Research Methods Section