Enabling reproducible type 1 diabetes polygenic risk scoring for clinical and translational applications

Type 1 diabetes polygenic risk scores (PRS) offer a promising tool for identifying diabetes subtypes in adults with new-onset disease. We aimed to develop a pipeline for the clinical translation of type 1 diabetes PRS to support clinical decision-making within a large health system and to provide publicly available code for applying these methods to future PRS models.

We adapted two established type 1 diabetes PRS models: a 67-SNP (GRS2) and a 7-SNP (AA7) score for a clinical genotyping platform and applied them to 73,346 participants in the biobank at the Colorado Center for Personalized Medicine (CCPM). We evaluated the scores' performance differentiating between type 1 and type 2 diabetes in adults using a clinician-curated diabetes phenotyping algorithm and examined associations with diabetes-related clinical data extracted from patients' health records. The impact of technical genotyping missingness on score accuracy and ancestry calibration were assessed independently.

Both scores effectively distinguished type 1 from type 2 diabetes across genetically defined ancestry groups (all AUC > 0.80) and demonstrated consistent performance in the UK Biobank (all AUC > 0.75). Individuals in the top quintile of each PRS were enriched for diabetic ketoacidosis (DKA) cases, accounting for nearly half of all DKA cases in the cohort. Additionally, the top quintile showed nearly threefold increased odds of GAD autoantibody positivity (OR = 2.94 [95% CI 2.08-4.17]).

Our evaluations demonstrated the potential utility of PRS for diabetes subtyping in a clinical setting. We present a framework of critical steps toward a standardized system for future translation of diabetes PRS to equitable clinical use, along with software to make it possible for others.