Spherical Coordinate System for Dyslipoproteinemia Phenotyping and Risk Prediction.

Background/Objectives: The factors contributing to residual atherosclerotic cardiovascular disease (ASCVD) risk in individuals are not fully understood, but knowledge of the specific type of dyslipoproteinemia may help further refine risk assessment. We developed a novel phenotyping and risk assessment system that may be applied automatically using standard lipid panel parameters. Methods: NHANES data collected from 37,056 individuals during 1999-2018 were used to develop a three-dimensional dyslipidemia phenotype classification system. ARIC data from 14,632 individuals were used to train and validate the risk model. Three-dimensional Cartesian coordinates were converted to spherical coordinates, which were used as features in a logistic regression model that provides a probability of ASCVD. UK Biobank data from 354,344 individuals were used to further validate and test the model. Results: Nine lipidemia phenotypes were defined based on the concentrations of HDLC, non-HDLC and TG. These phenotypes were related to the prevalence of metabolic syndrome, pooled cohort equation (PCE) score and ASCVD-free survival. A logistic regression model including age, sex and the spherical coordinates of the phenotype provided a composite risk score with predictive accuracy comparable to that of the PCEs. Conclusions: We provided an example of how a multidimensional coordinate system may be used to define a novel lipoprotein phenotyping system to examine disease associations. When applied to an ASCVD risk model, the composite spherical coordinate risk marker, which can be fully automated, provided an F1 performance score almost as good as the PCEs, which requires other risk factors besides lipids.