A combination of 15 SNPs and clinical features in relatives with islet autoantibodies improves prediction of Type 1 Diabetes: implications for prevention trials. — ASN Events

A combination of 15 SNPs and clinical features in relatives with islet autoantibodies improves prediction of Type 1 Diabetes: implications for prevention trials. (#43)

John M Wentworth 1 , Munish Mehta 2 , Grant Morahan 2 , Leonard C Harrison 1
  1. WEHI, Parkville, VIC, Australia
  2. University of Western Australia, Perth

Better methods to predict progression to type 1 diabetes (T1D) should improve the efficiency of prevention trials. Although the vast majority of individuals with two or more islet antibodies eventually progress to T1D only 35% do so within four years1. This proportion can be enriched to ~60% using the Diabetes Prevention Trial (DPT) risk score2, which is calculated from measures of age, glucose tolerance, beta cell function and body mass index. We sought to identify a gene signature that would improve prediction of progression to T1D in individuals with two or more islet antibody specificities (2Ab+).

We analysed clinical and Immunochip data from 453 2Ab+ TrialNet Pathway to Prevention Study participants enrolled between March 2004 and June 2012. Logistic regression identified 113 SNPs associated with T1D. We separated participants into two HLA-DR-matched groups of 314 (training set) and 139 (test set). A machine-learning algorithm analysed the training set and identified 15 of the 113 SNPs that best predicted progression to T1D. The ability of these 15 SNPs to predict T1D was confirmed in the test set. Moreover, these SNPs predicted T1D in T1D Genetics Consortium (T1DGC) participants. The 15 SNP signature comprised 7 HLA loci and 8 other loci not previously associated with T1D. When combined with the DPT risk score, the gene signature improved discrimination between progressors and non-progressors by identifying those with DPT risk scores less than 7 who progressed more rapidly to diabetes. Data from 2Ab+ TrialNet participants was then modeled to determine the potential impact of the improved prediction tool. We assumed that treatment would halve the rate of progression to diabetes, with a 3-year recruitment window (March 2004 to March 2007) and a further 3 years of follow-up. The combined risk score reduced the required sample size from 177 to 110 without compromising statistical power. 

Conclusion: A novel gene signature contributes to better prediction of progression of 2Ab+ individuals to diabetes and has the potential to improve prevention trial efficiency. Further study of its component genes may identify new markers of disease progression and provide insight into T1D pathogenesis.

  1. 1. JAMA. 2013 309:2473-9 2. Diabetes Care. 2011 34:1785-7