Commingling in distributions of lipids and related variables

In a sample of nearly 8,000 Japanese males, the distributions of casual cholesterol and triglyceride, hematocrit, glucose, uric acid, diastolic blood pressure, and weight (covariance adjusted) could not be normalized by a power transform and were significantly better fitted by a mixture of distributions. The evidence for admixture was nonsignificant for systolic blood pressure, significant but unimpressive for height and weight, and strong for the remaining variables. The minor component corresponded to high values, in low frequency except for triglyceride and glucose. These results favor an interpretation of elevated levels in terms of distinct entities, genetic or environmental or both, rather than cumulative small effects only. These entities appear to be megaphenic (i.e., with effects exceeding one phenotypic standard deviation). Consequences of this hypothesis are discussed.