Joint longitudinal models for dealing with missing at random data in trial-based economic evaluations

TL;DR

This paper introduces joint longitudinal models to handle missing at random data in trial-based economic evaluations, improving inference accuracy by utilizing the data's longitudinal structure rather than aggregating and discarding information.

Contribution

It proposes a novel joint modeling approach that accounts for longitudinal data structure in economic evaluations, enhancing the handling of missing data under MAR assumptions.

Findings

Joint models outperform traditional methods in simulation studies.

Application to real case studies demonstrates improved estimation accuracy.

Method reduces bias caused by data missingness in economic evaluations.

Abstract

Health economic evaluations based on patient-level data collected alongside clinical trials~(e.g. health related quality of life and resource use measures) are an important component of the process which informs resource allocation decisions. Almost inevitably, the analysis is complicated by the fact that some individuals drop out from the study, which causes their data to be unobserved at some time point. Current practice performs the evaluation by handling the missing data at the level of aggregated variables (e.g. QALYs), which are obtained by combining the economic data over the duration of the study, and are often conducted under a missing at random (MAR) assumption. However, this approach may lead to incorrect inferences since it ignores the longitudinal nature of the data and may end up discarding a considerable amount of observations from the analysis. We propose the use of joint longitudinal models to extend standard cost-effectiveness analysis methods by taking into account the longitudinal structure and incorporate all available data to improve the estimation of the targeted quantities under MAR. Our approach is compared to popular missingness approaches in trial-based analyses, motivated by an exploratory simulation study, and applied to data from two real case studies.