Use of a type 1 genetic risk score for classification of diabetes type in young Australian adults: the Fremantle Diabetes Study Phase II

The applicability of a UK‐validated genetic risk score (GRS) was assessed in 158 participants in the Fremantle Diabetes Study Phase II diagnosed between 20 and <40 years of age with type 1 or type 2 diabetes or latent autoimmune diabetes of adults (LADA). For type 1 versus type 2/LADA, the area under the receiver operating characteristic curve (AUC) was highest for serum C‐peptide (0.93) and lowest for the GRS (0.66). Adding age at diagnosis and body mass index to C‐peptide increased the AUC minimally (0.96). The GRS appears of modest diabetes diagnostic value in young Australians.

When relatively young people are diagnosed with diabetes, there can be challenges associated with determining which type is present. 1 An incorrect categorisation can have clinical implications, such as delayed insulin initiation with the risk of ketoacidosis in a person with type 1 diabetes (T1D) who is considered to have type 2 diabetes (T2D) and inappropriate use of insulin and consequent weight gain and hypoglycaemia potential in someone thought to have T1D but who is not truly insulin requiring. 2Clinical diagnostic uncertainty is increasing as rates of obesity increase in younger age groups. 3In addition, relevant laboratory tests can contribute to misclassification since islet autoantibodies can be present in people who do not have T1D, while the absence of islet autoantibodies does not rule out T1D, and serum insulin/C-peptide at diagnosis may not be significantly abnormal during the 'honeymoon period'. 4,5n light of these considerations, the genetic risk score (GRS) has been developed as a way of improving the accuracy of categorisation of diabetes type. 6Several are available and comprise groups of common single nucleotide polymorphisms (SNPs) that help discriminate between T1D and T2D independently of other relevant variables, including body mass index (BMI), serum C-peptide concentrations and islet autoantibody positivity.One GRS has been developed and validated in young Europeans and a 9-SNP version of its T1D GRS had an area under the receiver operating characteristic (ROC) curve (AUC) of 0.87 for both discrimination from T2D and progression to insulin deficiency (requirement for continuous insulin therapy within 3 years of diagnosis). 7he clinical utility of this GRS with an additional SNP was subsequently found to have significant but lower predictive ability to identify T1D in young South Asians, 8 but its performance was not as good in African Americans. 9ince Australia is experiencing increasing rates of obesity 10 and diabetes, 11 including T2D in young people, 12 we sought to determine whether the availability of the T1D GRS developed in Europid populations 7 could improve the categorisation of diabetes type in a representative, community-based cohort of Australians diagnosed in early adulthood.Although Australia is an increasingly multi-ethnic society, 13 we restricted participants to those of Caucasian ethnicity given that the test was developed, and appears most reliable, 8,9 in this group.
Adult participants in the community-based longitudinal Fremantle Diabetes Study Phase II (FDS2) who were aged 20 to <40 years at diagnosis and who were of Anglo-Celt, Southern European or Other European ethnicity 14 were included.The FDS2 cohort is drawn from a sociodemographically representative urban Australian setting, 11 and the basic demographic and diabetesspecific characteristics of recruited and non-recruited participants were similar. 14Type of diabetes was ascertained from the study investigator's review of FDS2 baseline data, including participant self-report, documentation by health professionals, treatment history, anthropometric measures, including BMI, and the results of any relevant prior laboratory tests.Participants with T1D were typically those started on permanent insulin therapy at or soon after diagnosis and who were nonobese (BMI <30 kg/m 2 ) and who had prerecruitment positive islet autoantibodies if available.Although the categorisation process was based on data obtained in a usual care setting, we also screened all FDS2 participants with a clinical diagnosis of T2D for glutamic acid decarboxylase antibodies (GADA) and categorised those who were positive as having latent autoimmune diabetes of adults (LADA).In addition, those with genotypically confirmed maturity-onset diabetes of the young and secondary diabetes were excluded from the present analyses. 11The FDS2 was approved by the South Metropolitan Area Health Service Human Research Ethics Committee (reference 07/397), and all participants gave written informed consent.
][9] Genotyping was performed using the KASP genotyping assay by LGC Genomics (Hoddesdon, UK).Each variant is weighted by its effect size on T1D risk based on published data, with weights for DR3/DR4-DQ8 assigned from imputed haplotypes.The combined numerical score represents genetic susceptibility to T1D in an individual.An abbreviated 10-SNP version was also calculated (see Supporting Information).The computer package IBM SPSS Statistics 28 (IBM Corporation, Armonk, NY, USA) was used for statistical analysis.To compare the AUCs for two different models, logistic regression with T1D as the dependent variable was used.Regression coefficients from each model were used to derive two new linear predictors which combined all the independent variables with the regression coefficients as weights.The two linear predictors were then compared using ROC analysis to do a test of the difference in AUCs.A P-value <0.05 was considered statistically significant.
The characteristics of the 158 participants categorised by diabetes type are summarised in Table 1.Those with T2D or LADA had an older age at diagnosis and a higher fasting serum C-peptide concentrations compared to those with T1D.Given these data, the T2D and LADA groups (the latter of which was small, with n = 10) were amalgamated into a single group of 99 participants.The AUC values for the main plausible demographic, clinical and laboratory variables differentiating T1D from T2D/LADA are shown in Table 2.The highest ROC AUC (0.93) was for serum C-peptide, and the lowest was for the 10-SNP GRS (0.65) which was marginally below that for the full 30-SNP GRS (0.66).Adding age at diagnosis to the serum C-peptide increased the AUC (95% confidence interval (CI)) to 0.95 (0.92-0.99), and the further addition of BMI increased the AUC to 0.96 (0.93-0.99), but only the latter model was significantly better than serum C-peptide alone (P = 0.113 and 0.044 respectively).For the 'high risk' 30-SNP GRS cut-off >0.280, 7 the sensitivity and specificity for identifying clinical T1D versus T2D/LADA in people from FDS2 who were diagnosed at a relatively young age were 28.0% and 87.9%, with a positive predictive value (PPV) of 56.0% and negative predictive value (NPV) of 69.0%.

Discussion
In community-based Australians from the FDS2 cohort who were diagnosed with non-monogenic, non-secondary diabetes before the age of 40 years, the GRS, regardless of whether it was 30-SNP or 10-SNP, had modest value in differentiating those with T1D from those with T2D/LADA.The 30-SNP score had low specificity, albeit high sensitivity, with moderate PPV and NPV.Other predictive variables, namely age at diagnosis, BMI and especially serum C-peptide, were more informative.The already very high AUC for serum C-peptide was increased only minimally by the addition of other variables in a regression model.These findings confirm the utility of serum C-peptide as part of the identification of young people with T1D and, thus, at greatest risk of metabolic decompensation if the diagnosis were missed and thus insulin therapy was not instituted promptly.The FDS2 participants in this substudy were diagnosed with a mean of approximately 20 years before recruitment.There is evidence from longitudinal studies in T1D that the serum C-peptide typically declines exponentially from diagnosis to low or undetectable concentrations at 15 years and beyond, 15 although there may be subgroups of patients in which there is a much slower, prolonged reduction. 16,17Nevertheless, the very low fasting serum C-peptide concentrations in our participants with T1D are likely to have generated a higher AUC differentiating them from those with LADA/T2D than if the same analysis had been performed close to the time of diagnosis when the serum C-peptide concentrations were closer to those of the other diabetes types.This consideration would, however, have no influence on the modest GRS AUCs, which were assessed independently.
The reason for the lower AUC in this study (0.65) compared to that in the initial development of the test (0.88) 7 likely relates to differences in participant characteristics and study design.Compared to participants in the initial validation study, 7 those in the present study tended to be older at recruitment (mean age 53 vs 49 years).This may mean that the original validation cohort was relatively enriched with individuals with 'true' T1D given that the prevalence of T1D is greater in people aged <30 years than in older age groups. 18We considered the small group of LADA participants together with those with T2D because of phenotypic similarities and issues with the use of a laboratory diagnostic test (GADA) with imperfect sensitivity. 19In the original validation study, the GRS was used to predict insulin requirement within 3 years of diagnosis rather than to identify T1D, an end-point that likely captured individuals with true LADA.Nevertheless, the AUC value for the comparison of T1D/LADA with T2D in the present study (AUC (95% CI) 0.68 (0.59-0.77)) was similar to that for T1D versus LADA/T2D (0.66 (0.57-0.76)).The original validation cohort was likely mainly Anglo-Celt given that it was from South West England, 7 whereas we included participants from all European countries.Given recognised ethnic/racial differences in the performance of the GRS, 8,9 the inclusion of non-Anglo-Celt individuals may have diluted the presence of key SNPs in our sample.
The limitations of the present study are the small sample size, but it was more than two thirds the size of the original GRS validation sample. 7The value of the GRS in differentiating T1D from T2D in younger individuals may be greater than we found, especially (as acknowledged) in comparison with serum C-peptide, when performed early after presentation compared to the relatively long diabetes duration (mean 18 years) in our FDS2 participants.The strengths of this study are the FDS2 comprehensive clinical assessments and the ability to select the present subgroup from a large community-based, representative sample.Notwithstanding the need for further studies in population-based samples of young Australian adults presenting with diabetes, as well as the development and validation of new GRSs that expand the detection of SNPs associated with T1D, the present data do not support a current role for the GRS in routine clinical management as an aid to determining the type of diabetes in this situation.Within the limitations of the present serum C-peptide dataset in the context of relatively long T1D duration, laboratory tests and simple clinical variables, including BMI, should guide initial and subsequent management, as supported by current primary care guidelines. 20

Table 1
Baseline characteristics of Fremantle Diabetes Study Phase II participants included in the present study † † †P < 0.001 versus LADA, Bonferroni-corrected for multiple comparisons.Data are percentages, mean ± SD or median [interquartile range]; two-sample comparisons were by Fisher's exact test for proportions, Student's t-test for normally distributed variables and Mann-Whitney U test for other variables.BMI, body mass index; GADA, glutamic acid decarboxylase antibody; HbA1c, glycated haemoglobin; LADA, Latent Autoimmune Diabetes of Adults; SNP, single nucleotide polymorphism; T1D, type 1 diabetes; T2D, type 2 diabetes.Genetic risk score in diabetes diagnosis

Table 2
Receiver operating characteristic area under the curve (AUC) for variables differentiating type 1 diabetes (T1D) from latent autoimmune diabetes of adults and type 2 diabetes