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.