Using Reduced Rank Regression to Identify Patterns of Diet and Serum Biomarkers of Obesity to Predict Risk of Cardiovascular Disease and All-Cause Mortality – A New Approach Applied in Epidemiology

Objective: The present study aims to introduce a novel math model, the reduced rank regression (RRR) to classify patterns of multiple diet and serum markers of obesity and to test associations between these patterns and risk of cardiovascular disease (CVD) and all-cause mortality.

Methods: Data (n=13,687 in subjects aged 20 and older) from the third National Health and Nutrition Examination Survey Dietary markers are analyzed. Multiple dietary indicators (vitamins C and D, magnesium, calcium, sodium, vegetable protein, dietary fiber, animal protein, and total fat) obtained using 24-hour Food History Questionnaire, and serum biomarkers (total cholesterol, LDL-C, HDL-C, triglycerides, HbA1c, and CRP) were clustered using traditional and newly proposed RRR approaches to classify patterns of these factors, and associations of RRR1 with CVD and all-cause mortality are analyzed using Cox's regression models.

Results: During an average 12-year follow-up, 3,425 subjects died from all-causes (17.68%) and 1,375 died from CVD (6.90%). As compared to the first quintile of RRR1 score, hazard ratios (HRs,95%CI) of the second to fifth quintiles of RRR1 for CVD mortality were 0.91 (0.68-1.21), 1.02 (0.74-1.39), 1.16 (0.89-1.51) and 1.88 (1.49-2.38), respectively (test for trend, p<0.001). Blacks had 23% higher risk of CVD mortality (HR: 1.23, 95%CI: 1.01-1.52) than Whites after controlling multiple covariates. Similar results were observed in the associations between RRR1 and all-cause mortality. As compared to traditional approaches, RRR1 had a higher predictive power.

Conclusions: RRR provides a better tool to identify patterns of diet and biomarkers of obesity, and improves the predictive power of the associations between exposures and outcomes.