Despite the rise in type 2 diabetes prevalence worldwide, we do not have a method for early risk prediction. The predictive ability of genetic models has been found to be little or negligible so far. In this study, we aimed to develop a better early risk prediction method for type 2 diabetes. Materials and methods: We used phenotypic and genotypic data from the Nurses' Health Study and Health Professionals' Follow-up Study cohorts and analyzed them by using binary logistic regression. Results: Phenotypic variables yielded 70.7% overall correctness and an area under the curve (AUC) of 0.77. With regard to genotype, 798 single nucleotide polymorphisms with P-values of lower than 1.0E-3 yielded 90.0% correctness and an AUC of 0.965. This is the highest score in the literature, even including the scores obtained with phenotypic variables. The additive contributions of phenotype and genotype increased the overall correctness to 92.9% and the AUC to 0.980. Conclusion: Our results showed that genotype could be used to obtain a higher score, which could enable early risk prediction. These findings present new possibilities for genome-wide association study analysis in terms of discovering missing heritability. These results should be confirmed by follow-up studies.

Despite the rise in type 2 diabetes prevalence worldwide, we do not have a method for early risk prediction. The predictive ability of genetic models has been found to be little or negligible so far. In this study, we aimed to develop a better early risk prediction method for type 2 diabetes. Materials and methods: We used phenotypic and genotypic data from the Nurses' Health Study and Health Professionals' Follow-up Study cohorts and analyzed them by using binary logistic regression. Results: Phenotypic variables yielded 70.7% overall correctness and an area under the curve (AUC) of 0.77. With regard to genotype, 798 single nucleotide polymorphisms with P-values of lower than 1.0E-3 yielded 90.0% correctness and an AUC of 0.965. This is the highest score in the literature, even including the scores obtained with phenotypic variables. The additive contributions of phenotype and genotype increased the overall correctness to 92.9% and the AUC to 0.980. Conclusion: Our results showed that genotype could be used to obtain a higher score, which could enable early risk prediction. These findings present new possibilities for genome-wide association study analysis in terms of discovering missing heritability. These results should be confirmed by follow-up studies.