Estimation of the graft failure by current value joint model, and extension to alternative parameterization...

Area of Science:

• Biostatistics
• Medical Statistics
• Clinical Data Analysis

Background:

• Clinical practice often involves tracking longitudinal measurements to predict patient outcomes.
• Traditional statistical approaches may analyze longitudinal data and survival events separately, potentially missing crucial associations.
• The relationship between longitudinal estimated glomerular filtration rate (eGFR) and kidney transplant graft failure warrants advanced modeling.

Purpose of the Study:

• To compare the classical approach, extended Cox model, and joint models (JMs) for analyzing time-varying longitudinal eGFR and kidney transplant survival.
• To evaluate different JM parameterizations, including current value and weighted cumulative effect models.
• To assess the utility of JMs for dynamic predictions in kidney transplant recipients.

Main Methods:

• Utilized a cohort dataset of 158 kidney transplant recipients with baseline and follow-up data.
• Applied the extended Cox model, current value JM, and weighted cumulative effect JM using R statistical software.
• Assessed model performance using parameter and standard error comparisons, and goodness-of-fit criteria.

Main Results:

• Hazard ratios for graft failure per unit decrease in log(eGFR) were 8.80 (extended Cox), 10.58 (current value JM), and 3.65 (weighted cumulative effect JM).
• The current value JM and weighted cumulative effect JM showed significant associations between coronary heart disease and log(eGFR).
• The current value JM outperformed the extended Cox model and weighted cumulative effect JM based on statistical criteria.

Conclusions:

• Joint models, particularly the current value JM, are preferable to traditional methods for analyzing longitudinal eGFR and predicting kidney transplant outcomes.
• JMs provide more accurate predictions by accounting for the dynamic nature of eGFR, facilitating improved clinical decision-making.
• Model selection should consider baseline biomarker levels, trends, distribution, and the number of longitudinal biomarkers.