Does diabetes appear in distinct phenotypes in young people? Results of the diabetes mellitus incidence Cohort Registry (DiMelli)

Abstract

Introduction: The diabetes mellitus Incidence Cohort Registry (DiMelli) aims to characterize diabetes phenotypes by immunologic, metabolic, and genetic markers. We classified patients into three groups according to islet autoantibody status and examined whether patients with multiple diabetes-associated autoantibodies, one autoantibody, or without autoantibodies differed with respect to clinical, metabolic, and genetic parameters, including an insulin sensitivity (IS) score based on waist, HbA1c, and triglycerides. We also assessed whether metabolic markers predicted the immune status.

Materials and methods: As of June 2012, 630 patients in Bavaria, Germany, aged <20 years diagnosed with any type of diabetes within the preceding 6 months were registered in DiMelli. We compared the clinical and laboratory parameters between islet autoantibody status defined patient groups. Parameters showing the strongest associations were included in principal component analysis. Receiver operating characteristic curves were used to assess the ability of the IS Score to predict islet autoantibody status.

Results: Patients with multiple islet autoantibodies, one autoantibody, or without autoantibodies were significantly different in terms of BMI percentile, weight loss before diagnosis, fasting C-peptide (all, P<0.001), and IS Score (P=0.034). However, principal component analysis revealed no distinct patterns according to autoantibody status. At the optimal IS Score cut-off for predicting islet autoantibody positivity (single compared to none), the specificity was 52.0% and the sensitivity was 86.8%. With respect to prediction of multiple autoantibodies (compared to none), specificity and sensitivity were slightly lower and in combination inferior to those obtained using the BMI percentile and fasting C-peptide.

Discussion: The DiMelli study indicated that patients with and without islet autoantibodies differed with respect to metabolic and genetic markers but there was considerable overlap of phenotypes, and autoantibody status could not be predicted by these parameters. Thus, our results suggest that refined diabetes classification may require both immune and metabolic phenotyping.

Figure 1. Empirical distributions of Insulin Sensitivity (IS) Score, BMI percentile, weight loss before diagnosis, and fasting C-peptide according to islet autoantibody status.

Figure 2. Biplots of the first two principal components (PC) determined by principal component analyses using BMI percentile, weight loss before diagnosis, fasting C-peptide, and insulin sensitivity score, according to the number of autoantibodies in the whole dataset (top) and in children aged 10–20 years (bottom).

The variance explained by PC1 and PC2 is shown in parentheses along each axis.

Figure 3. Receiver operating characteristic curves for number of islet autoantibodies status (dashed line: one compared to none, solid line: multiple compared to none) according to Insulin Sensitivity (IS) Score, BMI percentile, weight loss before diagnosis, and fasting C-peptide.

Optimal cut-off values determined by the Youdenindex with specificity and sensitivity in parentheses are displayed.