The quality of clusters obtained with different pre-processing-clustering pipelines.

For each pre-processing/clustering pipeline, the number of clusters (N) and the homogeneity-separation difference (HA-SA) are provided. The HA-SA measure reflects the difference in variation between individuals in different clusters (SA) and individuals in the same cluster (HA). For the validation set, the mean±SD of the number of clusters and HA-SA were calculated for 10 re-sampled datasets. See Materials and Methods for a complete description of the algorithms and pre-processing methods. K-mns = K-means; NoramTranf = transformation to normal; Z-score = Z-score normalization; AgeAdj = age adjustment followed by Z-score normalization. The analysis was performed separately for the NHANES training and validation sets, using all individuals (“All Patients”) or restricting the analysis to diabetic patients (“Diabetes”).