MODELLING LONGITUDINAL TRAJECTORIES OF PATIENT-REPORTED OUTCOMES TO EVALUATE TREATMENT EFFECT
Author(s)
Jen M
Eli Lilly and Company, Surrey,, UK
Presentation Documents
OBJECTIVES: To evaluate treatment effect on longitudinal patient-reported outcomes using appropriate analytical strategy. METHODS: This was an ad-hoc analysis of longitudinal patient-reported outcomes using a two stages simulated data in which the true model is known, to explore and to evaluate the capability of the group-based trajectory method to identify the distinctive features of a highly irregular but still continuous population distribution of trajectories. Firstly, we created six different types of underlying trajectory in which the true model is known and added in level-one between occasion random noises. Then we added a level-two, between- patients variation (of the random intercept form) with differing variability to each of the six distinctive trends. This simulation allows us to examine how the software implementation identifies different group trajectories as well as their level-one and -two variances. It was recognized that a priori assignment of distinct longitudinal trajectories may not be appropriate and that no ability to calibrate the precision of individual classifications exists if ex-ante rules are used. Thus, latent group-based trajectory model, a method to map the developmental course of symptoms and assess heterogeneity in response to clinical interventions, was used to identify patient groups with varied response. RESULTS: The fitted trajectories closely approximate the true shapes and there is also a close correspondence for the percentage of places attributed to each group. Even the size of the level 1 random term is correctly estimated. The semi-parametric group-based trajectory method has demonstrated unequivocally its capability to capture the unobserved subgroups in the presence of considerable level-1 random variation. CONCLUSIONS: Patients in many disease areas experience changes in QoL in different ways. Identification of those groups is essential for appropriate evaluation of therapy treatment effects and identification of factors contributing to those groupings.
Conference/Value in Health Info
2014-11, ISPOR Europe 2014, Amsterdam, The Netherlands
Value in Health, Vol. 17, No. 7 (November 2014)
Code
PRM210
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Multiple Diseases