Visualization of patient clusters within the training cohort
(A) Using a 4-D embedding as input, patients within the training cohort were classified into 29 clusters by fitting a GMM. Each 4-D cluster was then visualized by labeling the training cohort with colors on a 2-D embedding.
(B) The fitted GMM was then applied to classify patients from waves one and two of the validation cohort into the same set of 4-D clusters. The proportion of patients assigned to each cluster was calculated within each cohort.