Cluster analysis of an insulin-dependent diabetic cohort towards the definition of clinical subtypes

Clinical and biochemical data on 111 consecutive insulin-dependent diabetic children enrolled in a longitudinal prospective study were analyzed to determine if more than one clinical expression of Type I diabetes exists. Use of multivariate statistical methods, including Correspondence Analysis, kappa-means clustering and RECPAM (RECursive Partition and AMalgamation), show that there are two well differentiated clinical expressions of IDDM each characterized by a cluster. One is characterized by later age, less severe onset, longer symptom duration, less beta-cell disappearance after 12 months, more females; the other by earlier age, more sudden and severe onset, DR 3/4, earlier disappearance of beta-cell function and more males. RECPAM analysis provides further insight into the structure of the two clusters. An other RECPAM tree identifies low, medium and high risk groups of disappearance of beta-cell function at 12 months after diagnosis.