Predicting diabetes risk in diverse populations: what next?

Diabetes is a major health problem of the 21st century, affecting 463 million individuals globally.¹ Early identification of diabetes and therapy targeted to the individual can alter disease progression and reduce the risk of costly complications. The accuracy and utility of current prediction models that use clinical risk factors varies with age, race or ethnicity, and natural history of disease, including the characteristics and appearance of risk factors before disease onset. Incorporation of genetic risks, not affected by temporal changes, with clinical risk factors might improve outcomes and reveal the promise of precision medicine in diabetes.

Genome-wide association studies (GWASs) for diabetes subtypes have identified hundreds of risk-modifying alleles. These associations can be aggregated into a weighted sum of the number of risk alleles in each individual. This provides a single genetic score called a polygenic risk score (PRS), which estimates the probability of developing disease (eg, diabetes). Longitudinal studies have shown that PRSs can predict onset of complex diseases. In the FINRISK population-based study,² each standard deviation of the PRS for type 2 diabetes was associated with a hazard ratio of 1·74 for occurrence of disease. Furthermore, individuals at the top 2·5% of the PRS distribution had a 2·3 times increased lifetime risk of developing type 2 diabetes.² The PRS for type 2 diabetes improves prediction, regardless of age and before clinical risk factors become apparent, enabling early prediction and offering the opportunity of early lifestyle interventions.² In type 1 diabetes, the PRS facilitated the prioritisation of individuals for islet autoantibody screening.³ Patients with multiple islet autoantibodies are eligible for increased monitoring for prevention of diabetic ketoacidosis and prioritised for entry into immune intervention trials.³ In addition, the PRS can distinguish between both type 1 diabetes and type 2 diabetes with high accuracy and identify individuals with type 2 diabetes who will require insulin for optimal glucose control.⁴ The PRS can also be valuable to better inform the heterogenous course of diabetes. GWASs of type 2 diabetes and related traits have been clustered into physiologically relevant pathways to obtain process-specific PRSs. These PRSs aid in distinguishing subtypes of type 2 diabetes that are associated with distinct clinical outcomes in a precision medicine manner.⁵

Most GWASs for diabetes and other common diseases have been done in individuals with northern European ancestry; however, diabetes is a global disease. In fact, diabetes disproportionally affects individuals with non-European ancestry. In the USA, the prevalence of diabetes is highest in African American and Latinx communities, and these patient populations are more likely to develop diabetic complications. These health disparities could be exacerbated in the application of PRSs for risk prediction because PRS models derived from populations with European ancestry showed poor transferability to other ancestries.⁶ The inability to extrapolate PRS models is due to population differences in risk allele frequencies and linkage disequilibrium patterns between risk alleles, as well as the presence of ancestry-specific risk alleles.

Current PRSs for diabetes show poor prediction in populations with non-European ancestries. The DIAMANTE study consortium performed a meta-analysis on genome-wide association studies for type 2 diabetes, which included 1 463 694 individuals of diverse ancestries (49% non-Europeans), and tested the performance of transancestry and ancestry-specific PRS.⁷ The transancestry PRS performed better than did the ancestry-specific PRS in all populations, explaining the type 2 diabetes risk of up to 6% in populations with white European ancestry. The risks explained were lower for other ancestries, and the lowest variance explained was only around 2% in populations with African ancestry. In type 1 diabetes, use of a PRS specific to African ancestry had better prediction of the disease in those of African descent than did a PRS specific to northern European ancestry, even though the PRS for African ancestry was developed with a smaller sample size and fewer variants.⁸ A 2021 multi-ancestry meta-analysis has provided an opportunity for improved PRS in type 1 diabetes.⁹ Datasets for gestational diabetes and diabetes complications are becoming available, but the respective GWASs and ancestry-specific and transancestral PRSs are yet to be performed.

There are various resources available to improve PRS prediction in diverse ancestries. First, large-scale biobanks, including diverse ancestries with genetic information that is linked to electronic health records, allows for development of population-specific PRSs for diabetes, diabetic complications, and related risk factors. Second, sharing of genetic association results allows for aggregation and meta-analyses that provide precise variant effect size estimations to construct the PRSs. Third, ongoing development of novel PRS methods have shown improved transferability because they use GWAS data from diverse ancestries. For example, the PRS-CSx method, which integrates summary statistics and linkage disequilibrium patterns from multiple ancestries, has shown improved transferability to non-European populations for numerous quantitative traits and schizophrenia risk.¹⁰

The Polygenic Risk Methods in Diverse Populations (PRIMED) is a new consortium funded by the National Human Genome Research Institute and the National Cancer Institute that aims to improve the utility of PRSs in diverse populations. The PRIMED consortium comprises seven study sites, including the Diabetes Polygenic RIsk Scores in Multiple ancestries (D-PRISM). D-PRISM is an international consortium that focuses on improving PRS prediction of diabetes and progression across the lifecourse in diverse ancestries. One of the goals of this study is to apply methodologies that take advantage of multi-ancestry summary statistics from genome-wide association studies to create improved PRS for diverse ancestries, an approach that has been proven successful for other diseases. Although these and other consortia are using existing genetic data to develop PRSs that can augment clinical information, large-scale genetic data generated in populations with non-European ancestry are still needed. Not only will these data improve performance of PRSs for diabetes and its complications, but the availability and implementation of these PRSs should also reduce health disparities and disease burden globally, enabling the path to precision diabetes medicine for all individuals.