Improving personalized prediction of type 2 diabetes

Type 2 diabetes (T2D) is a complex adult-onset disease resulting from multiple clinical, environmental and genetic factors. It is a major public health concern, and innovative methodological approaches are urgently needed to better predict it and to slow down its epidemic. We have developed a new algorithm (1) to calculate polygenic risk scores (PRS) for type 2 diabetes (T2D), based on a large number of genetic variants (SNPs) and their estimated effect sizes from large-scale meta-analyses of genome-wide association studies (2). We have seen in the Estonian Biobank (EstBB) cohort that such scores are strongly associated with prevalent disease status and also have a good predictive power for incident disease. These results suggest that implementation of such risk scores in personalized risk prediction may lead to better-targeted preventive measures as well as earlier and more accurate diagnosis and treatment. However, it is important to see whether the effect is similar across different cohorts or should the scores be computed differently for different populations. Therefore, it would be important to analyze the Lifelines cohort (Lifelines) in parallel with the EstBB cohort and study the differences and similarities. The first step has been done, and the methods developed in EstBB, have been validated in Lifelines (manuscript prepared for submission). The following steps would be to apply the new PRS in the UGLI data set and to improve its predictability even further. For example, in the original study of the new PRS method (1), the SNP selection was based on the Cardio-MetaboChip, containing SNPs highly associated with T2D only from specific genes related to cardio-metabolic traits, but not including genome-wide hits from across the entire genome. However, the availability of genome-wide UGLI data in Lifelines will provide a great opportunity to include more relevant SNPs in the new PRS associated with increased risk of T2D, likely leading to a more accurate prediction of T2D. Besides, the GRS computation in the original study for the new PRS method was based on meta-results from 2012 (2), but new meta-GWAS data for T2D have recently become available with an approximately seven-fold increase in the sample size (3), providing much more accurate SNP effects, which will likely result in a substantially improved PRS prediction ability. 

While focusing on predictors and complex interplay between predictors of T2D, the complicated nature of T2D should also be considered for its diagnosis. According to current guidelines T2D is typically diagnosed if one of the following criteria is met: fasting plasma glucose (FPG) > 7.0 mmol/L and/or glycated haemoglobin (HbA1c) ≥ 6.5% (4). Nevertheless, existing T2D treatment directed at lowering the blood glucose levels together with recommended lifestyle changes have not been able to stop the epidemic of T2D or solve the problem of severe comorbidities (5). It shows that considering only glucose levels for T2D diagnosis and its management for treatment might not be sufficient since the heterogeneous character of T2D involves many more clinical components. A novel approach to refine the classification of T2D was recently introduced, where T2D was divided into five subgroups based on six variables highly related to the development of diabetes (i.e., glutamate decarboxylase antibodies, age at diagnosis, BMI, HbA1c, β-cell functioning and insulin resistance), not only based on glucose levels (5). There were significant differences in disease progression and complication severities between T2D subgroups even supported by genetic differences referring to the importance of capturing the heterogeneous nature of T2D in its diagnosis. Refined T2D classification may improve personalized prediction of T2D, besides it is also promising for more individualized treatment and for identifying individuals with higher risk of complications already at diagnosis. Besides, the two different approaches (new whole-genome PRS and refined disease classification), access to two large European Biobanks (Lifelines and EstBB) would provide a unique opportunity to test the transferability of the PRS calculated based on one population to another population while accounting for target population structure. Ultimately, implementing such evidence-based knowledge in a clinical context could result in the improvement of personalized risk prediction.