Previous - Next
Table 1 shows that there were, in general, three overall food patterns based on three main groups of staples: first, cassava, maize and beans; second, rice and beans; third, wheat and rice. These food patterns correspond only crudely with the range of income because there is a substantial overlap in income between the first and second categories. When, however, the nutritional content of the diet is assessed, there is no overlap in the categories for mean fat consumption, although individuals varied substantially in their consumption. There was a consistent increase in fat, protein and often alcohol consumption as income rose. A high income was also associated with a greater animal protein and visible fat intake.
When the vectorial analysis was undertaken, the distribution of individuals in relation to the different variables proved surprising. Figure 1 shows a very close concordance between the fat and protein content of the diet and the BMI of the men. The BMI categories of 20-22.9 and 23.0-25.0 lie along the first axis. This means that there is a substantial chance of finding people within this BMI range in any segment of the parabola of dietary relationships. The distance of the high BMI values above the line signifies the close relationship of these high BMIs to conditions where the income is high but where the fat content of the diet is also high. Thus the fat values were >30% of energy in association with BMIs ~27.6. As a contrast, the positions of the lower BMI value, i.e. 17.0-18.4 and <17.0, are poorly related to the nutritional parameters of the diet, implying that the macronutrient content of the diet is unlikely to make an important contribution to low body weights. When more detailed analyses were made with each individual point being displayed in a manner similar to the generic display shown in Fig. 1, then there was a surprising condensation of the individual points so that the BMI related to the nutritional profiles (data not shown) with the scatter diagram showing a remarkable confluence of the data in association with the fat curve set out in Fig. 1. As income increased a greater number of households increased their consumption of both fat and protein. There was a shift from vegetable to animal protein consumption and this was associated with higher fat intakes. However, detailed scrutiny of the data suggested that as income rose the increase in vegetable protein consumption came first with only modest increases in fat; thereafter there was an increase in fat consumption as the switch to animal products occurred with greater incomes. These differences in household purchasing habits explain the displacement of the protein curve from the fat curve in Fig. 1.
In order to reassess this interpretation, a reanalysis was made where protein and fat were considered together in relation to BMI, and once allowance has been made for the different options for protein and fat consumption, then the individual households showed little other variation in behaviour to link with BMI. The fat and protein patterns thus formed a coherent relationship with BMI and the issue of income became much less important as a determinant of adult size.
Table 2. Clustering of dietary patterns in relation to the prevalence of overweight
|
Dietary characteristics % energy |
Nutritional composition (% energy) |
Overweight | ||||||||||||||||
|
|
% male population |
Cereals |
Roots |
Sugar |
Pulses |
Meat fish eggs |
Dairy produce |
Free edible fats |
Green veg fruits |
Carbohydrates | ||||||||
|
Cluster |
Total |
Starch |
Sucrose |
Proteins |
Fats |
% BMI > 25 |
% > 27.5 |
% > 30 |
% of obese males | |||||||||
|
1 |
28.2 |
2.7 |
10.3 |
4.3 |
19.4 |
11.7 |
15.5 |
5.2 |
48.7 |
38.4 |
10.3 |
16.5 |
34.8 |
42.6 |
19.5 |
7.2 | ||
|
2 |
32.3 |
3.0 |
10.6 |
5.1 |
15.7 |
8.7 |
17.5 |
4.8 |
52.6 |
42.0 |
10.6 |
14.5 |
32.9 |
37.0 |
17.0 |
6.9 | ||
|
34.4 |
3.4 |
11.5 |
6.7 |
14.2 |
7.1 |
17.3 |
4.0 |
55.2 |
43.7 |
11.5 |
13.6 |
31.2 |
32.4 |
14.0 |
5.4 | |||
|
Total |
18.1 |
43 | ||||||||||||||||
|
4 |
36.3 |
3.8 |
13.0 |
6.5 |
13.3 |
6.3 |
15.6 |
3.7 |
58.5 |
45.5 |
13.0 |
13.5 |
28.0 |
25 4 |
10.0 |
3.4 | ||
|
5 |
38.2 |
3.9 |
14.3 |
7.5 |
11.2 |
5.8 |
14.3 |
3.5 |
61.5 |
47.2 |
14.3 |
12.9 |
25.6 |
22.9 |
8.8 |
3.4 | ||
|
6 |
38.2 |
5.7 |
14.6 |
7.5 |
11.0 |
5.7 |
12.9 |
3.8 |
63.5 |
48.5 |
14.6 |
12.5 |
24.0 |
19.3 |
7.9 |
3.2 | ||
|
Total |
24.5 |
31 | ||||||||||||||||
|
7 |
40.7 |
7.7 |
14.0 |
9.0 |
9.4 |
4.2 |
11.5 |
3.0 |
67.0 |
53.0 |
14.0 |
12.4 |
20.6 |
15.2 |
6.0 |
2.1 | ||
|
8 |
42.7 |
7.5 |
14.8 |
12.4 |
7.7 |
2.5 |
9.8 |
1.9 |
71.4 |
56.6 |
14.8 |
11.3 |
17.3 |
10.8 |
3.5 |
1.4 | ||
|
Total |
28.4 |
20 | ||||||||||||||||
|
9 |
16.0 |
34.4 |
23.8 |
11.4 |
13.2 |
10.0 |
0.9 |
3.4 |
2.7 |
76.6 |
65.2 |
11.4 |
11.9 |
11.5 |
6.3 |
2.0 |
0.6 |
5 |
|
10 |
13.0 |
28.3 |
35.1 |
9.9 |
15.6 |
4.9 |
1.1 |
2.2 |
2.5 |
82.8 |
72.9 |
9.9 |
9.8 |
7.4 |
4.8 |
0.9 |
0.4 |
2 |
By relating the patterns of diet to BMI and income it proved possible statistically to devise 10 clusters for the whole population which incorporated nutritional and dietary features and the probability of different degrees of adiposity. The characteristics of these clustered groups are shown in Table 2. This shows, in comparatively simple terms, how the prevalence of overweight and obesity relates to the fat and protein content of the diet. Yet a more detailed look, e.g. comparing clusters 3 and 4, shows that the transition here involves little change in protein intake but appreciable differences in dietary fat and, related to this, very clear differences in the prevalence of overweight and obesity. Detailed analyses of the impact of these multiple variables suggested that >80% of the variance in BMI could be explained by the dietary fat content of the diet and that the impact of income and dietary pattern on BMI was best expressed by its fat content.
The evidence for the importance of dietary fat in promoting weight gain is growing steadily and the analyses of this survey were used to contribute a national perspective on this issue during the WHO discussions on the relationship between diet and chronic disease (WHO, 1990). Table 2 and Fig. 1. suggest that, above a fat value of 12%, there is a coherent relationship between the fat content of the diet and the BMI. Below this value the relationship does not hold and other data on energy intakes suggest that the total available energy in low fat diets may be the prime determinant of whether BMIs fall below 18.5. These analyses are unable to provide any insight into the mechanism whereby dietary fat influences the development of overweight, but it seems clear, from a combination of physiological, behavioural and metabolic studies that there is not only a poor thermogenic response to fat but that appetite control is less able to induce the appropriate accuracy of inhibition of food intake when high fat diets are being consumed. Stubbs et al. (1994) in recent studies on young male volunteers showed that when the lifestyle was sedentary a fat content as low as 20% was necessary to induce energy balance in subjects feeding ad libitum, whereas a 40% fat diet might be tolerated when the men were more active. The present epidemiological data deal necessarily with a large number of men subjected to a huge range of social conditions, but the Brazilian population when surveyed could not be considered very sedentary. This suggests that subtle effects of dietary fat on BMI may only become apparent when very large data sets are handled and when the range of fat infakes is sufficiently large, as in the present study, to allow the effect of dietary fat to be discriminated.
Acknowledgement - We thank our colleague, M. T. L. de Vasconcellos, Estatistico Consultant, Rio de Janeiro, Brazil, for help in organizing the data sets and FAO for the original financial support which made these analyses possible. The work was also supported by the Scottish Office Agriculture and Fisheries Department.
Bouchard C & Tremblay A (1990): Genetic effects in human energy expenditure. Int. J. Obes. 14, Suppl. 1, 49-58.
Hill MD (1974): Correspondence analysis: a neglected multivariate method. Appl. Statist. 23, 340-354.
Lebart L, Morineau A & Warwick K (1984): Multivariate descriptive statistical analysis. New York: J. Wiley.
Sorensen TIA, Stunkard AJ & Holst C (1991): Recessive inheritance of a major gene for human obesity. Int. J. Obes. 15, Suppl. 1, 13 (abstract).
Stubbs RJ, Goldberg DR, Murgatroyd PR & Prentice AM (1994): The effect of covert changes in dietary fat and energy density on ad libitum food intake in humans (Abstract). Proc. Nutr Soc. 52, 35A.
WHO (1990): Diet, nutrition and the prevention of chronic disease. Technical Report Series 797. Geneva: WHO.