Authors:
Guozhi Jiang
;
Eric S. Lau
;
Ying Wang
;
Andrea O. Luk
;
Claudia H. Tam
;
Janice S. Ho
;
Vincent K. Lam
;
Heung M. Lee
;
Xiaodan Fan
;
Wing-Yee So
;
Juliana C. Chan
and
Ronald C. Ma
Affiliation:
The Chinese University of Hong Kong, China
Keyword(s):
Multiple imputation, Bootstrap, Prediction model, Coronary heart disease, Type 2 diabetes.
Related
Ontology
Subjects/Areas/Topics:
Bioinformatics
;
Biomedical Engineering
;
Biostatistics and Stochastic Models
;
Model Design and Evaluation
Abstract:
The objectives of this study were to develop and compare the prediction models based on imputed data sets with that based on complete-case (C-C) data set for coronary heart disease (CHD) in type 2 diabetes mellitus (T2DM) and to identify novel genes associated with CHD from T2DM related genes. A prospective cohort of 5526 patients with T2DM and without known CHD and heart failure at baseline was used in this analysis. During a median follow-up time of 8.8 years, 406 (7.3%) patients developed incident CHD. Multiple imputation (MI) was performed to tackle missing values for 26 clinical variables and 40 genetic variables, while Cox proportional hazards regression with backward variable selection was applied to bootstrap samples. Five different MI or C-C models were compared and the performance based on C-index, 5 years AUC and the slope of prognostic index were similar, three SNPs located at NEGR1, CDKAL1 and ADAMTS9 were found to be significant after adjusting for clinical variables. I
n conclusion, multiple imputation and bootstrap can be benefit to the development of prediction model, and a stable risk factor set for CHD was successfully identified from our dataset containing clinical and genetic variables.
(More)