Harmonizing Longitudinal and Survival Data Using a Joint-Modeling Framework: An Efficient Approach to Assessing Social Interventions

Abstract

This article is an exposition of the joint-modeling approach to testing intervention effects through the harmonization of longitudinal and timeto-event data. We demonstrate the advantages of the joint-modeling approach over the classical approach of separately analyzing these types of outcome data. Method: We used a subset of 150 participants from the Illinois Birth through Three Title IV-E Waiver intervention study, which collected longitudinal Devereux Early Childhood Assessment for Infants and Toddlers (DECA-I/T) scores and time-to-permanence data for up to 3 years. We ran and contrasted three competing models: Cox proportional hazard, linear mixed-effects, and joint modeling. Results: If analyzed separately, the DECA-I/T scores are highly nonsignificantly related to time to permanence ( p 5 :929). However, when analyzed jointly, the significance level drops 88 percentage points, from .929 to .105. Because of its efficiency in addressing information loss when longitudinal and survival data are incorporated together, the joint model properly accounts for outcome-dependent missingness. Conclusion: This article highlights the utility of joint modeling in randomized longitudinal intervention studies by demonstrating its ability to preserve information from both longitudinal and time-to-event data, produce unbiased estimates, and retain higher statistical power than the traditional approach.