FIG. 5.

Clustering similar patients in feature space produces monitored cohorts. At the healthcare level, the national pool of 350M individuals is partitioned into population cohorts of 350K persons each, where each cohort is clustered in 30K-dimensional feature space of health determinants. The clustering is thus feature derived rather than disease derived. At the infrastructure level, each cohort is now reduced to the scale of present national surveys, where there are 350 demographic cohorts of 1000 persons each. That is, each light-colored shape represents the human variation of 1000 persons with similar health features and similar population demographics, so they are sufficiently accurate to discover effective treatments.