islet autoantibodies by age 6 and 7.6% risk of diabetes by age 10, compared with those with scores below this who had rates of 4.1% and 2.7% respectively [59]. ADVANCE ARTICLE: Endocrine Reviews Downloaded from https://academic.oup.com/edrv/advance-article-abstract/doi/10.1210/er.2019-00088/5535575 by 81225740 user on 24 July 2019 ADVANCE ARTICLE Endocrine Reviews; Copyright 2019 DOI: 10.1210/er.2019-00088 9 Leveraging advances in density of SNP arrays as well as larger reference panels, the most recently updated T1D rsPS includes 67 SNPs and accounts for interactions between 18 HLA DR-DQ combinations [60]. When applied to the UK Biobank, this enhanced T1D rsPS significantly outperformed previous scores, identifying individuals with T1D with AUROC of 0.92. These figures are close to the maximum performance figures predicted for T1D, based on the modelling analyses described earlier in the context of T2D [42]. 3.3 Refining the diagnosis of major diabetes subtypes As well as predicting future disease risk, polygenic scores are emerging as powerful tools to support diagnosis of major diabetes subtypes. Determining whether a particular patient has T1D, T2D, or one of the other specified forms of diabetes is not always straightforward. A clinical diagnosis of T1D can often, but not always, be substantiated by the presence of one or more islet autoantibodies (GAD, IA2, IAA, ZnT8), as these are found in >90% of newly diagnosed patients [61]. However, these antibodies are not always measured in clinical practice, and do not provide perfect determination of T1D diagnosis due to a combination of (i) background presence in some individuals without T1D, (ii) lower rates of positivity for T1D individuals diagnosed in adulthood, and (iii) waning titers over time from initial diagnosis [57]. The measurement of C-peptide levels in plasma or urine can also help distinguish T1D from other forms of diabetes, but use of this test is not routine, not least because it has reduced value at the time of diagnosis (where it can be suppressed even in T2D or monogenic diabetes) or during the “honeymoon period” of T1D, given residual beta-cell function in the early years following presentation [62]. The consequence is relatively high rates of both under- and over-diagnosis of T1D when trying to differentiate it from both T2D and less common forms of diabetes, such as MODY [63]. The stable nature of a polygenic score, unchanged throughout life, offers a useful tool to aid in diagnostic characterization of individuals with established diabetes. An early application of the initial T1D rsPS developed by Oram and colleagues was in discriminating between T1D and T2D. The authors applied both a 69-SNP T2D rsPS and a 30-SNP T1D rsPS to a sample of well-defined cases of T1D and T2D from the Wellcome Trust Case Control Consortium GWAS [64]. They found the T1D rsPS was highly discriminative (AUROC 0.88), whereas the T2D EPS was less so (AUROC 0.64), and that combining the two offered little improvement beyond the T1D score alone (AUROC 0.89) [57]. Application of the 30-SNP T1D rsPS alone to a cohort of 223 adults, aged between 20 and 40 diagnosed with diabetes at least 3 years previously, predicted progression to insulin deficiency (AUROC 0.87) and offered information additional to that provided by antibody status [57]. In 8,608 individuals with a clinical diagnosis of T2D after 35 years of age, treated without insulin for at least 6 months following diagnosis, the same T1D rsPS predicted progression to insulin use at five years, but only in the small subset of GAD antibody-positive participants: the probability of insulin use ranged from 17.6% in those in the lowest tertile of T1D-risk to 47.9% in the highest [65]. T1D polygenic scores have also provided a clearer sense of the extent of T1D prevalence across the age spectrum. Using a 29-SNP T1D rsPS, Thomas and colleagues demonstrated that, amongst individuals participating in the UK Biobank, 42% of genetically-defined T1D was observed in those diagnosed with diabetes between 31 and 60 years, pointing to a far higher proportion of overall T1D presenting in adulthood than is commonly appreciated [66]. It can be challenging to detect these individuals clinically since, in this age range, they represent only a small minority (~4%) of patients with any form of diabetes. Compared to those with T2D, individuals with T1D defined on the basis of a high T1D rsPS had lower BMI, were more likely to use insulin in the first year of diagnosis, and were at higher risk of diabetic ketoacidosis [66]. ADVANCE ARTICLE: Endocrine Reviews Downloaded from https://academic.oup.com/edrv/advance-article-abstract/doi/10.1210/er.2019-00088/5535575 by 81225740 user on 24 July 2019 ADVANCE ARTICLE Endocrine Reviews; Copyright 2019 DOI: 10.1210/er.2019-00088 10 T1D polygenic scores have also shown utility in discriminating early-onset T1D from monogenic forms of diabetes including MODY [67], neonatal diabetes [67], and monogenic autoimmune diabetes [68], that typically present during childhood. In these settings, a T1D score can prioritize patients who are most likely to benefit from sequence-based testing for rare causal variants, and support correct interpretation of novel variants of uncertain functional significance that emerge from such sequencing. Prioritization of patients in this