Waist and hip circumferences in different populations

A Molarius et al
118
and population. Regression models were constructed CI: 76.8, 78.6) and Gothenburg (95% CI: 77.4, 78.9)
by adding these explanatory variables one by one into to 91 cm (95% CI: 89.8, 91.3) in the Czech Republic.
the model. Since BMI has been de®ned so that it is Mean hip circumference ranged from 97 cm in
largely unrelated to height, the order in which BMI Glostrup (95% CI: 95.7, 97.4) and Beijing (95% CI:
and height are added into the model should make little 96.0, 97.1) to 108 cm (95% CI: 107.0, 108.3) in the
difference. Here, height was added into the model Czech Republic, and the mean WHR from 0.76 (95%
®rst, because it is not a modi®able variable, and then CI: 0.756, 0.766) in Perth to 0.84 (95% CI: 0.835,
BMI followed by age group and population. In 0.844) in the Czech Republic. The WHR, and espe-
our data, introducing the BMI before height into cially waist circumference, was higher in populations
the model would have decreased the proportion where the median BMI was relatively high compared
explained by the BMI for waist circumference by to other populations, although this was more pro-
1%, while increasing the proportion explained by nounced in women than in men. The mean values
the BMI for the WHR by 1±3%. The change in the were similar to the medians (50th percentile), except
proportion explained by height would have been the for waist and hip circumferences in women, where the
opposite. Height was ®rst introduced as a categorical distributions were slightly positively skewed.
variable (for each 10 cm) to check whether the asso- Next, we calculated the mean values for waist and
ciation was linear. Based on visual inspection, it was hip circumferences and WHR by 10 y age groups for
detected that the association was linear, and height each population. Table 2 gives the mean values in the
was therefore treated as a continuous variable. Like- age group 25±34 y and the difference in means
wise, BMI was modelled only with a linear effect, between the oldest and youngest age group. In gen-
because the second order term for BMI was negligi- eral, all the measures increased with age. The measure
ble. Because the effect of age was not linear, 10 y age which showed the smallest increase with age was hip
groups were used instead of age in years. Regression circumference in men. In women, the effect of age on
coef®cients were calculated using the general linear waist circumference was stronger in the Czech Repub-
model (GLM) procedure of SAS statistical software.25 lic, Friuli and Novi Sad than in the other populations.
The proportions of variation in waist, hip and the Figure 1 shows the population adjusted increase in
WHR explained by the independent variables were waist circumference by age, with and without adjust-
derived from the R-square statistics. Finally, to com- ment for BMI and height. In women, the crude effect
pare the results from the regression analysis based on of age on waist was linear. The increase in waist
individual data to those based on population level circumference with age attenuated when adjusted for
data, we calculated Pearson correlations between age BMI and height. Because height contributed only a
standardized mean BMI and age standardized mean little to this attenuation compared with BMI, it can be
waist circumference for men and women. These concluded that a considerable proportion of the
correlations were ecological where each population increase in waist by age is explained by increasing
presented one observation. BMI, especially in women. Still, there remained a
signi®cant (P < 0.001) increase in waist (> 4 cm in
men and women), which was not due to height or
Results BMI. Figure 2 gives the corresponding effect of age
on the WHR. The increase in the WHR with age was
more pronounced in the younger age groups in men,
Table 1 gives the age standardized percentiles, means whereas it was more pronounced in the older age
and s.d.s for waist and hip circumferences and the groups in women. The difference in the effect of age
WHR, the median BMI and height, and the number of on the WHR between the sexes was statistically
observations by population, for the common age range signi®cant (P < 0.001 for interaction term between
35±64 y in men and women (explanation of the gender and age group).
population abbreviations is given in Appendix 2). Both height and BMI were positively associated
Men had higher values of waist circumference than with waist circumference. The results of the regres-
women, whereas there was little difference in hip sion analyses revealed that height accounted for very
circumference. Therefore, the mean WHR was little (1% in men and < 1% in women) of the variation
higher in men than in women. In men, age standar- in waist circumference, whereas BMI accounted for
dized mean waist circumference ranged from 83 cm about three quarters (Figure 3,a). Age group and
(95% con®dence interval (CI): 82.6, 84.3) in Beijing population explained an additional 5%. Jointly, height,
to 98 cm in the Czech Republic (95% CI: 97.9, 99.1) BMI, age group and population, explained 83% of the
and Halle County (95% CI: 96.9, 99.4). Mean hip variation in waist circumference in men and 79% in
circumference ranged from 94 cm (95% CI: 93.7, women. Waist increased by 0.3 cm (s.e. 0.005 in men
95.0) in Glostrup to 105 cm (95% CI: 103.6, 105.6) and 0.007 in women) for each cm of height in both sexes
in Halle County, and the mean WHR from 0.87 (95% and by 2.4 cm (s.e. 0.01) in men and 2.1 cm (s.e. 0.01) in
CI: 0.866, 0.877) in Beijing to 0.99 (95% CI: 0.988, women for each unit of BMI.
0.996) in Glostrup. In women, age standardized mean The results were fairly similar for hip as for waist
waist circumference ranged from 78 cm in Perth (95% circumference (Figure 3,b). BMI accounted for 64% and