Joint Modeling of Multivariate Longitudinal Measurements and Survival Data: Application to Hemodialysis Data

In studies involving hemodialysis patients, repeated laboratory measurements (longitudinal data) and survival outcomes are often analyzed separately, which can lead to biased results due to ignoring measurement errors and the intrinsic dependency between the two analyses. Joint modeling has emerged as a powerful approach to handle such data. This study aims to investigate the impact of six time-varying biochemical markers, along with baseline covariates, on the survival of hemodialysis patients using a multivariate joint model.

A longitudinal cohort of 894 maintenance hemodialysis (MHD) patients, who had started dialysis between 2004 and 2023, were included. Baseline and follow-up clinical information and monthly laboratory measurements were analyzed. A multivariate linear mixed-effects model was jointly fitted with a Cox proportional hazards model to simultaneously assess the longitudinal biomarkers and time-to-event data. Analyses were performed using R software.

The model indicated that older age (Hazard Ratio, HR = 1.02, P < .001), male gender (HR = 1.72, P < .001), diabetes mellitus (HR = 1.61, P < .001), walking disability at admission (HR = 1.78, P < .001), and catheter-based vascular access (HR = 1.71, P < .001) were significantly associated with an increased risk of mortality. Higher square root of phosphate levels (HR = 13.97, P < .001) were linked to increased, and higher square root of creatinine (HR = 0.32, P < .001), hemoglobin (HR = 0.75, P = .009) and albumin (HR = 0.31, P < .001) levels were associated with decreased mortality.

Findings of the joint model confirm the importance of baseline clinical risk factors and modifiable biochemical markers on the survival outcomes of hemodialysis patients.