650 – Estimation, Prediction, and Forecasting Through Data-Driven Models
Longitudinal Rasch Model in Harmonizing Adherence Recall Intervals
Yan Wang
UCLA Fielding School of Public Health
Honghu Liu
University of California at Los Angeles School of Dentistry
The data harmonization aim to create the common measure that can be used across studies. Harmonization is to combining data sets collected at different times into a single, consistent data series. Rasch model is not sensitive to the sample that generate the response conversion key. Recent literature conceptualize longitudinal Rasch model as the hierarchical generalized linear models by assuming the latent variable as random effect. In longitudinal data harmonization, the challenge part is to calibrate the latent trait over time, estimated the invariant parameters in the Rasch model and preserve the changing of the trait for modeling. Self-reported adherence is the most convenient way to obtain the adherence measure among HIV patients by directly asking the patients themselves. In this study, we use the longitudinal version of Rasch model to create the conversion key, which is used to calibrate the self-reported adherence of HIV medication over time. The results are compared with the device measured adherence measure to ensure the validity of calibration.