An estimation of deviation towards the mean for serum lipid fractions in patients followed for a varied duration.

ABSTRACT

We evaluated the deviation towards the mean and attempted to quantify it among the different lipid fractions in patients. The study was done retrospectively on patients who were judged to be metabolically stable and had repeated total cholesterol (TC), high density lipoprotein cholesterol (HDL) and triglyceride (TG) measured in a single laboratory with known coefficient of variation for repeated measurements. The patients and their data were separated into 3 groups. Group A (56 patients) evaluated the difference between the first and its average obtained from an average of 4 samples per patient within a mean of 9 months. Group B, examined pairs of data taken an average of 12 months apart. Group C, evaluated 45 patients with at least 3 data points each a year apart. Linear correlations were applied for the repeats versus the first samples. Highly significant correlations were obtained for all the groups. The slopes were less than one (generally between 0.66 and 0.85) and intercepts had positive values. This was seen even for the HDL whose range of values span 25 to 85 mg per cent. These results strongly supported deviation towards the mean such that from our calculation and in this population, a person with an initial TC of 200 mg per cent would have from 37 to 61 per cent chance of obtaining a significantly higher value if the test was repeated. The magnitude of the change would average 30 mg per cent for cholesterol and as much as 30 per cent of the initial values for TG. In this evaluation, the time intervals between repeats did not appear to influence the result. Yearly follow-up also did not seem to exhibit the effect of aging. However, the latter 2 conclusions rested on a small number of observations. It is suggested that several repeated estimations of these lipid fractions be done before a decision is made towards intervening. In instances of epidemiological studies, it is imperative to obtain representative repeated measurements since this deviation towards the mean will alter the slope of the events versus the lipid-variables.