Abstract
Epidemiological studies using lipidomic approaches can identify lipids associated with exposures and diseases. We evaluated the sources of variability of lipidomic profiles measured in blood samples and the implications when designing epidemiologic studies. We measured 918 lipid species in nonfasting baseline serum from 693 participants in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, with 570 participants having serial blood samples separated by 1-5 years and 72 blinded replicate quality control samples. Blood samples were collected during 1993-2006. For each lipid species, we calculated the between-individual, within-individual, and technical variances, and we estimated the statistical power to detect associations in case-control studies. The technical variability was moderate, with a median intraclass correlation coefficient of 0.79. The combination of technical and within-individual variances accounted for most of the variability in 74% of the lipid species. For an average true relative risk of 3 (comparing upper and lower quartiles) after correction for multiple comparisons at the Bonferroni significance threshold (α = 0.05/918 = 5.45 ×10-5), we estimated that a study with 500, 1,000, and 5,000 total participants (1:1 case-control ratio) would have 19%, 57%, and 99% power, respectively. Epidemiologic studies examining associations between lipidomic profiles and disease require large samples sizes to detect moderate effect sizes associations.