Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 172 - Prediction and Misclassification in Biomedical Research
Type: Contributed
Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #312293
Title: Linear Mixed Model Vs Two-Stage Methods for Developing Prognostic Models for Kidney Function Decline
Author(s): Brian Kwan* and Loki Natarajan
Companies: University of California, San Diego and University of California, San Diego
Keywords: biomarker; diabetes; kidney disease; mixed model; prediction; two-stage

Diabetic kidney disease (DKD) is a common co-morbidity of diabetes, and much research is underway to identify early markers of DKD progression, i.e. estimated glomerular filtration rate (eGFR) change. A goal in DKD research is to develop prognostic models to accurately predict eGFR change. The linear mixed model with serial eGFR measures (outcomes) is the optimal statistical approach for prognostic model development, namely by evaluating the coefficient of the time*prognostic factor interaction term. However, two-stage methods that first estimate subject-specific eGFR slopes, and then use these outcomes in a regression framework, offer advantages as they are easy to interpret and can be readily implemented by researchers. In this work, we compared the linear mixed effects and two-stage methods, in terms of bias, efficiency and coverage via simulations and analytic methods, allowing for irregularly spaced measures, and varying degrees of missingness. We determined study design settings where the two-stage models performed competitively, thus offering a simple modeling alternative to DKD researchers. Notably, our work applies to other longitudinal marker-disease studies.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program