Predictive modeling of health care costs: Do cardiovascular risk markers improve prediction?

To investigate the ability of multiple cardiovascular disease (CVD) markers to predict future health care costs. CVD markers included traditional risk factors (smoking status, body mass index, waist circumference, alcohol intake, diabetes, total : high-density lipoprotein cholesterol ratio, actual hypertension, physical activity) and newer markers (carotid intima-media thickness, hemoglobin A1c, apolipoprotein B : apolipoprotein A-1 ratio, lipoprotein (a), leukocyte count, high-sensitive C-reactive protein, plasma fibrinogen, estimated glomerular filtration rate, urinary albumin : creatinine ratio). The study sample consisted of 2233 participants without history of myocardial infarction, stroke, heart failure, and angina pectoris at baseline (50.6% women; mean age 60.9 years; age range 45-81 years) from the cohort Study of Health in Pomerania, Germany (median follow-up 5 years). Predictive modeling revealed that a basic model with sex, age, years of school education, insurance status, and income explained 0.9% in baseline total cost variation and 1.5% in total cost variation at 5-year follow-up. The incorporation of a combination of significant CVD markers resulted in an increase in the R2 for total costs of 70% at baseline and 69% after 5 years, with a final R2 of 0.030 at baseline and an R2 of 0.048 at 5-year follow-up. Our data suggest that for individuals without history of CVD, the simultaneous addition of several CVD risk markers improves predictive modeling of future health care costs beyond that of a model that is based on established health care predictors