Magnitude of the Regression to the Mean within One-year Intra-individual Changes in Serum Lipid Levels among Japanese Male Workers

To investigate the magnitude of the 'regression to the mean' effect for the changes in serum total cholesterol (TC), triglycerides (TG), and high density lipoprotein cholesterol (HDLC) levels during one-year interval between annual health check-ups in occupational settings, the relationships between the baseline level and subsequent one-year change in TC, TG or HDLC were analyzed using paired health check-up data in 1998 and 1999 of 547 Japanese male clerical workers. After adjustment for age, body mass index (BMI), yearly change in BMI, drinking score and smoking score by the multivariate analyses, the yearly changes in each serum lipid (ATC, A Ln(TG) or A HDLC) were clearly inversely associated with the lipid levels in 1998. For example, in the multiple linear regression analyses setting A value in each serum lipid as a dependent variable, the partial regression coefficients for the baseline lipid levels (f3,) were 0.21 (p<0.001) for the TC, 0.39 (p<0.001) for the Ln(TG) and 0.15 (p<0.001) for the HDLC, respectively. These results suggest that the observed yearly change in each serum lipid level may largely reflect the 'regression to the mean' effect in addition to the real yearly biological change. J Epidemiol, 2001 ; 11:61-69

Therefore, the measured levels at the baseline show the regression to the mean of the population as the observation period proceeds.In order to appropriately evaluate the measured risk factor levels at the baseline or their changes after a particular interval, it is potentially important to make relevant adjustment for such 'regression to the mean' effect.Some previous reports in Western countries well illustrate the influences of this effect using the examples of blood pressure 1-5) or serum total cholesterol (TC) level 1,3,4).However, the magnitude of this effect has not yet been elucidated in serum triglyceride (TG) or HDL cholesterol (HDLC) levels.And in Japan, there have not been epidemiological reports on this effect.Therefore, in the present study, the magnitude of the 'regression to the mean' effect for one-year change in each of TC, TG and HDLC levels was investigated using annual health checkup records for Japanese male workers.The results of this study will contribute to the appraisal of the regression dilution bias for risk assessment in a cohort study setting serum lipid levels as risk factors at baseline in Japan .

Subjects
The Mean and SD (in parentheses) are listed except for "Age-range".Unit of B: * : mg/(dl X year), ** : mm2/dl, * * * : mg/dl TC : Total cholesterol level, Ln(TG) : Natural logarithm of triglycerides level, HDLC : High-density-cholesterol level, BMI : Body mass index of the population, the value is likely to be more often situated near the bottom of the physiological curve rather than the top.While in the former case the measured value tends to decrease by the remeasurement after a particular time interval, in the latter case it tends to increase by the replicate measurement (i.e.The return to the usual value can be found.).Accordingly, the measurement error and the physiological fluctuation may be major causes for the regression to the mean within the intra-individual changes in bio-markers during a particular time interval.Moreover, the 'regression to the mean' effect resulted from these unintentional factors may inevitably introduce the regression dilution bias for risk assessment in a cohort study setting the bio-markers as risk factors at baseline.
As a result of the present study, such 'regression to the mean' effects could be found for any of the three serum lipids.The shrinkages of the range between the mean values of the top and bottom quintiles of the baseline level in each serum lipid were clearly noted after one-year interval.Clarke u suggested that the ratio of the range observed after a follow-up to the range at the baseline corresponds to the nonparametric estimate of the 'regression dilution ratio' .According to this concept, the 'regression dilution ratios' after one-year interval were 0.78(=72/92) for the TC, 0.64(=1.073/1.665)for the Ln(TG), and 0.86(=31/36) for the HDLC, respectively.

Figure 1 .
Figure 1.Comparison of mean TC in 1998 and those in 1999 by quintiles of TC in 1998.TC : Total cholesterol level.

Figure 2 .
Figure 2. Comparison of mean Ln(TG) and geometric mean of TG (in parentheses) in 1998 and those in 1999 by quintiles of Ln(TG) in 1998.TG : Triglycerides level, Ln(TG) : Natural logarithm of triglycerides level.

Figure 3 .
Figure 3.Comparison of mean HDLC in 1998 and those in 1999 by quintiles of HDLC in 1998.HDLC : High -density -lipoprotein cholesterol level.

Figure 4 .Figure 5 .
Figure 4. Adjusted mean levels of one -year change in TC by quintiles of the level in 1998 .TC : Total cholesterol level.

Table 1 .
Comparison of paired health check-up data (data in 1998 and those in 1999) among 547 subjects.
and bottom quintiles clearly shrank after one-year interval in these three lipids.For example, the mean TC levels of persons who were in the top quintile initially (i.e.those with a TC value Adjusted mean levels of one -year change in HDLC by quintiles of the level in 1998.