Identification of type 1 diabetes risk phenotypes using an outcome-guided clustering analysis

Abstract

Aims/hypothesis: Although statistical models for predicting type 1 diabetes risk have been developed, approaches that reveal the heterogeneity of the at-risk population by identifying clinically meaningful clusters are lacking. We aimed to identify and characterise clusters of islet autoantibody-positive individuals who share similar characteristics and type 1 diabetes risk.

Methods: We tested a novel outcome-guided clustering method in initially non-diabetic autoantibody-positive relatives of individuals with type 1 diabetes, using the TrialNet Pathway to Prevention study data (n=1123). The outcome of the analysis was the time to development of type 1 diabetes, and variables in the model included demographic characteristics, genetics, metabolic factors and islet autoantibodies. An independent dataset (the Diabetes Prevention Trial of Type 1 Diabetes Study) (n=706) was used for validation.

Results: The analysis revealed six clusters with varying type 1 diabetes risks, categorised into three groups based on the hierarchy of clusters. Group A comprised one cluster with high glucose levels (median for glucose mean AUC 9.48 mmol/l; IQR 9.16-10.02) and high risk (2-year diabetes-free survival probability 0.42; 95% CI 0.34, 0.51). Group B comprised one cluster with high IA-2A titres (median 287 DK units/ml; IQR 250-319) and elevated autoantibody titres (2-year diabetes-free survival probability 0.73; 95% CI 0.67, 0.80). Group C comprised four lower-risk clusters with lower autoantibody titres and glucose levels (with 2-year diabetes-free survival probability ranging from 0.84-0.99 in the four clusters). Within group C, the clusters exhibit variations in characteristics such as glucose levels, C-peptide levels and age. A decision rule for assigning individuals to clusters was developed. Use of the validation dataset confirmed that the clusters can identify individuals with similar characteristics.

Conclusions/interpretation: Demographic, metabolic, immunological and genetic markers may be used to identify clusters of distinctive characteristics and different risks of progression to type 1 diabetes among autoantibody-positive individuals with a family history of type 1 diabetes. The results also revealed the heterogeneity in the population and complex interactions between variables.

Keywords: Clustering analysis; Heterogeneity; Risk factors; Type 1 diabetes.

Fig. 1.

MDS map of clusters (center), flanked by the radar charts of each cluster (surrounding the MDS maps), and the Kaplan-Meier survival curves by cluster with their corresponding 95% confidence intervals (outer). ^aGlu MAUC: glucose MAUC. ^bFa Glu: fasting glucose. ^cCpep MAUC: C-peptide MAUC. ^dFa Cpep: fasting C-peptide. ^eN AAB: number of autoantibodies. ^fBMIz: BMI z-score. ^gT1D: type 1 diabetes.

Fig. 2.

Decision rules to assign individuals to clusters. ^aT1D: type 1 diabetes.

Fig. 3.

Comparison of diabetes-free Kaplan-Meier survival curves of clusters in the analysis dataset (TrialNet PTP; black solid lines) and the validation dataset (DPT-1; blue dashed lines). Numbers under the curves are the numbers of individuals at risk in the cluster at each time point P-values on the upper right corners are based on log-rank tests. ^aT1D: type 1 diabetes. ^bTNPTP: TrialNet PTP.