Model Selection in Survival Analysis of EXT2 Gene Polymorphisms with Age at Onset of Type 2 Diabetes

Background: This study aimed to compare the Cox regression and parametric survival models in genetic association analysis of age at onset (AAO) of type 2 diabetes (T2D) and examine the effect of exostosin 2 (EXT2) gene on risk and AAO of T2D. Methods: We tested 22 single nucleotide polymorphisms (SNPs) within the EXT2 gene in the Marshfield sample among 878 T2D cases and 2,686 non-diabetes controls. Multiple logistic and linear regression models in PLINK software were used to examine the association of each SNP with the risk of T2D and AAO of T2D, respectively. Cox regression in PROC PHREG and parametric survival models (including exponential, Weibull, log-normal, log-logistic and gamma models) in PROC LIFEREG in SAS 9.4 were used to perform survival analysis of AAO. The Akaike information criterion (AIC) and Bayesian information criterion (BIC) were used to compare the competing models. Results: PLINK software initially identified 1 SNP associated with the risk of T2D (rs7111879 with p=6.33 x 10-3) and 3 SNPs associated with AAO (rs7111879, rs42376464 and rs4755230 with p=3.26 x 10-2, 5.79 x 10-5 and 2.76 x 10-5, respectively). AIC values showed that the gamma distribution is the best model for above 3 SNPs and followed by the Weibull distribution; whereas BIC criteria showed that the gamma distribution is similar to the Weibull distribution. Conclusion: This study reveals that the parametric gamma and Weibull models performed better than Cox regression in genetic association of AAO of T2D and provides evidence of several genetic variants within the EXT2 gene associated with the risk and AAO of T2D