Machine learning applied to individuals' CBC blood test values over time shows that patients have stable "setpoints" for platelets, hemoglobin and other values; comparing new test results to these patient-specific values allowed for better diagnosis of diabetes and other common conditions