This is the old United Nations University website. Visit the new site at http://unu.edu

Trends in body mass index in developing countries

Abstract
Introduction
Methods
Results
Discussion
References
Appendix. Data sources

David L. Pelletier and Maike Rahn

The authors are affiliated with the Division of Nutritional Sciences at Cornell University in Ithaca, New York, USA.

Mention of the names of firms and commercial products does not imply endorsement by the United Nations University.

Abstract

The purpose of this study was to examine the extent to which adult body mass index (BMI) has changed in developing countries over the past several decades. The analysis is based on a compilation and analysis of mean BMI in 1,432 published samples from developing countries measured between 1957 and 1994. A hierarchical multiple-regression model is applied to these data, controlling for country and study as random covariates and modelling age, sex, socio-economic status, and year as fixed effects. The results reveal a statistically significant increase in mean BMI between 1957 and 1994 in all major regions of the developing world. The size of the increase was 1.4 kg/m² over the 37-year period, with a 95% confidence interval of 0.4 to 2.4 kg/m². Mean BMI appears to have increased in all major regions of the developing world, although the size of the increase varies across regions. Using assumptions about the statistical distribution of BMI within populations and cut-off points recommended by the World Health Organization, the analysis suggests that the increase in mean BMI may have resulted in a slight decrease in the prevalence of underweight (BMI<18.5 kg/m²) but is unlikely to have produced an increase in obesity (BMI>30 kg/m²) in most regions. By contrast, the use of lower cut-off points to define obesity, as is done in many individual studies, would suggest that obesity has increased in developing countries. These results highlight the importance of using standardized definitions for underweight and obesity among adults, the need to assess and consider the prevalence of both conditions simultaneously during planning and policy development, and the need to identify policy instruments appropriate to the nutritional profile within a country.

Introduction

For the world as a whole, the overall rates and causes of death have undergone dramatic changes in the past several decades. Taking the estimates from Bulatao [1] for the years 1970 to 1985 as an example, the total mortality rate for the world has declined from 1,287 to 1,034 per 100,000. This decline is even more marked among developing countries, from a rate of 1,378 in 1970 to 1,035 in 1985. These declines in overall mortality are associated with concomitant declines in virtually all identifiable categories of death for the world and for developing countries as a whole, with the most rapid declines being found for infectious diseases. Another important observation is that, in the developing world, there is an increase in the number and percentage of all deaths attributable to circulatory diseases and neoplasms, trends that are expected to accelerate sharply in the coming years [2].

There are three potential reasons for these changes, which have quite different policy implications and, therefore, are important to distinguish from one another. The first is that the spread of public health measures and knowledge has contributed to a rapid decrease in mortality from communicable diseases, which is most marked among young children and has accelerated since 1960, continuing a trend that began near the turn of the century in developed countries. This has been noted by several authoritative sources and appears well established [3 - 5]. The second is that, fuelled by simultaneous changes in fertility and age-specific mortality rates, a greater proportion of people are living into adulthood, when chronic diseases become manifest. This demographic transition is reflected in significant decreases in total and age-specific fertility rates and increases in life expectancy at birth and at age 15, and is also well established.

A third potential explanation is that the causes of death within various age and sex groups (notably among adults) have undergone changes. According to this explanation, changes in diet, exposure to infectious agents and environmental hazards, and various life-style factors (e.g., tobacco and alcohol use, occupational exposures, sedentary life-style, and psychosocial stress) have led to an increased biological risk for chronic disease. There is a large body of evidence establishing the relevance of these factors to chronic disease in developed countries, but a central question is whether, or to what extent, these changes are having potent and broad-scale effects on mortality in the developing world today. The use of proportional mortality rates (i.e., percentage of total deaths due to different causes) or all-ages mortality rates is inappropriate for testing this explanation, because such comparisons are potentially confounded by the above-mentioned changes in population age structure and the precipitous decrease in infectious diseases that are well established. Unfortunately, many authors have failed to distinguish among these three explanations, and have mistakenly assumed that the sharp increase in the share of total deaths due to chronic disease, or in the rate of death from chronic disease for the population as a whole, implies comparable changes in biological risk for chronic disease among adults in the developing world [6].

To illustrate the importance of methodology for conclusions concerning chronic disease trends in the developing world, table 1 compares the results based on four measures of mortality, adapted from Bulatao and Stephens [7]. The first three measures lead to the clear conclusion that chronic diseases are an emerging problem in developing countries, and, indeed, in a certain sense they are emerging problems. However, the trends in these indicators are being driven by changes in total population size and structure (measure 1), changing age structure (measure 3), and a combination of changing age structure and decline in infectious diseases (measure 2). When these forces are removed by examining age-specific death rates among adults (measure 4), it can be seen that the trend is one of stability or decline, consistent with the overall decline in mortality in these countries. Similar findings were reported for males and females in five developing countries with adequate vital registration systems [6] from 1955 to 1985, and similar findings emerged from the Global Burden of Disease Study [2].

The upward trends in the first three mortality indicators of table 1 have important policy implications for developing countries, even if they cannot be taken as evidence that the biological risk for chronic disease is increasing. These trends do imply that the health systems in these countries will increasingly need to adapt to the prevention and treatment of chronic diseases, because their populations are growing older and are growing in absolute size, and the ratio of deaths due to chronic diseases versus infectious diseases is increasing rapidly. This will create added competition for resources within the health sector, especially if expensive medical intervention strategies are chosen in lieu of primary prevention. However, from a broader policy perspective, it is important to determine the extent to which adults in the developing world are experiencing increases in biological risk for chronic disease as a result of changes in diet, environmental exposures, life-style, and other factors. The answers to this question would help decide the extent to which chronic disease risk should be another consideration in the formulation of food, agriculture, environmental, and nutrition policy. The age-specific mortality trends shown in table 1 (measure 4) suggest that this is not the case, contrary to the forceful impression created by the first three indicators that are more commonly seen and quoted in the journal literature, policy documents, and the media.

This paper extends an earlier report [8] that addresses the “diet, environment, and life-style” question using two quite different approaches. The first approach was to apply a demographic model to predict causes of mortality on an age-specific basis, using a number of other refinements not used in the projections published by the World Bank [7] and the National Research Council [5]. This analysis yielded results broadly similar to those of the Global Burden of Disease Study [2] and are not included in this paper.

TABLE 1. Trends and projections in deaths due to circulatory disease and neoplasms in the developing world according to four mortality measures

Measure	Cause of death	Males				Females
		1970	1985	2000	2015	1970	1985	2000	2015
No. of deaths x 1,000, 45-65 yr	Circul.	989	1,151	1,818	2,812	727	870	1,223	1,744
	Neopl.	380	617	890	1,466	368	521	802	1,299
% of deaths, all ages	Circul.	16.9	18.5	29.4	35.4	16.4	19.2	47.1	35.3
	Neopl.	4.9	7.3	10.8	13.5	4.9	7.0	10.4	13.7
Deaths per 100,000, all ages	Circul.	234	198	244	291	225	192	216	252
	Neopl.	68	78	90	111	68	70	79	98
Deaths per 100,000, 45-65 yr	Circul.	606	479	516	494	459	377	357	314
	Neopl.	0233	257	252	257	232	226	234	234

Source: ref. 7.

Methods

An initial overview of available data in 1990 by WHO [9] located only seven developing countries with national surveys from which the prevalence of obesity could be estimated, none of which had data from two points in time for estimating trends. A more recent review (T. Byers, personal communication, 1996) located another 10 large-scale surveys, but none of these were able to estimate trends over time. That report did note large differences in mean body mass index (BMI: weight/height²) between population groups differing in rural/urban residence or socio-economic conditions, with the urban, upper socio-economic groups having consistently higher mean BMIs and obesity prevalences.

In light of the current dearth of large-scale survey data from two or more points in time in developing countries, the present study uses an alternative approach to estimate time trends in BMI. Specifically, it is based on the compilation and analysis of the many small-scale, community-based surveys in developing countries, as reported in the scientific literature since 1960. Whereas none of these studies is individually able to indicate time trends in BMI, this study was undertaken to determine whether they might, in the aggregate, reveal the presence of such trends when analysed on a country-specific or region-specific basis. In addition, information on age, sex, and socio-economic characteristics was retrieved from these reports in order to analytically remove variation due to these sources and test the extent to which socio-economic variation (a proxy for future trends in developing countries) is associated with variation in adult BMI.

The data for this study were obtained by conducting a literature search for studies published between 1960 and 1995. The criteria for inclusion in the study were the following: data were provided on mean BMI, the prevalence of either high or low BMI, or mean weight and height from which BMI could be calculated; samples were taken from developing countries and represented non-European populations; samples did not represent clinical cases from hospitals or clinics; samples were entirely or largely composed of persons 18 years of age or above.

Three methods were employed for the literature search: computer-assisted search based on keywords, a systematic search of key journals based on geographic or topical focus, and the usual bibliographic branching or “snowballing” technique. These methods yielded 176 papers with data from 66 developing countries (see Appendix). It is suspected that many more papers of this type could have been found if the search had been extended, but this was not possible for the purposes of this paper.

Most of the 176 papers presented data on BMI, weight, and height for several strata from the overall sample, with strata defined according to age, sex, occupation, rural/urban location, education, and other socio-economic variables. Thus, the 176 papers yielded a data set of 1,432 observations, with each observation eventually representing the BMI for one particular stratum, from one study, in one country. The associated age, sex, and socio-economic descriptors for each stratum were recorded and used in the analysis. In recognition of the non-independence of such observations, a hierarchical modelling procedure was used for the analysis, as described below. The median sample size for the 1,432 observations, upon which a mean BMI was calculated, was 77 persons. The lowest quintile had a sample size of 25 or less and the highest quintile had 448 or more.

Although the central interest in this paper is on trends in obesity in developing countries, the available data from these countries usually take the form of mean weight, height, and BMI, as opposed to estimates of the percentage of adults with elevated BMIs, which would be more indicative of obesity. When prevalences are reported, they tend to refer to the percentage of adults with low rather than high BMI values, reflecting the predominant concern for undernutrition in these countries. For these reasons, the primary analysis is based on mean BMI values as the dependent variables. Mean BMIs were directly available for 40% of the observations and could be calculated from mean weight and height in another 40%. The remaining papers provided only prevalence data. Mean BMIs were estimated in these remaining cases by using prediction equations derived from 154 observations in the data set that contained mean BMI data as well as prevalence data.* Although the use of prediction equations for 20% of the sample introduces an additional random error component, there was no difference between the mean BMIs that were directly reported and those estimated through these equations.

* A variety of prediction models were tested, but the final one adopted for this purpose expresses mean BMI as a function of the prevalence above or below a given cut-off point, the cut-off point itself, the square and cube of the prevalence, and an interaction term between prevalence and the cut-off point. This equation has an R² value of 85% and a standard error of the estimate of 1.9 for estimating mean BMI.

A socio-economic index was estimated for each observation by using whatever information was provided in the published papers. Information on occupation, rural/urban residence, or other socio-economic indicators was available for 86.3% of the observations, and this information was used to create a combined socio-economic index. Although the nature and extent of this type of information vary widely across studies, an index was created with three “socio-economic levels.” Depending upon the information available, the lower level may represent rural populations, poor urban or peri-urban populations, traditional or agricultural occupations, or a variety of other descriptors of poverty as used by the authors. The higher socio-economic stratum represents sedentary, skilled, or high-wage occupations, students, “urban” samples not further distinguished by the authors, or groups designated as well-off or of high socio-economic status (SES) by the authors. The middle level includes mixed samples, those employed in the formal sector but not in high-wage jobs, “housewives,” the unemployed, and those designated as mid-level status by the authors. The middle level also includes the 13.7% of observations that could not be clearly classified in the other two levels because no information was available. The resulting socio-economic index is assumed to represent an ordinal variable whose primary validity relates to intra-paper comparisons as presented in the original papers and may have additional validity across papers from the same country or region. However, the index is assumed to have much weaker validity across regions and cannot be used to make strict comparisons across regions.

Table 2 provides a breakdown of the sample according to socio-economic level and region. The distribution of observations by socio-economic index favours the lowest category in the case of sub-Saharan Africa, India, South and South-East Asia, and Australasia, where roughly 50% to 65% of observations fall into this category and the remainder are divided between the middle and highest categories. The opposite pattern is seen with the China samples, in which 74% of observations were classified into the highest level. The remaining regions have more balanced distributions. For reasons given above, these distributions are assumed to reflect the characteristics of the particular sample of studies used in this study, rather than the regions as a whole, and the socio-economic distinctions are primarily assumed to have internal validity (i.e., within-paper and within-country validity).

TABLE 2. Distribution of samples according to socio-economic index and region

Region	Socio-economic indexa
	Lower	Middle/mixed/unstated	Higher
Sub-Saharan Africa	65.4	21.2	13.5
South and South-East Asia	48.7	28.2	23.1
India	54.6	18.2	26.8
China	18.6	7.8	73.6
Australasia	56.7	32.3	10.9
Polynesia and Micronesia	34.6	36.2	29.1
Latin America and Caribbean	31.8	52.5	15.7
North Africa and middle Asia	30.4	42.9	26.8
All regions	46.8	29.4	23.8

a The index is a composite variable based on residence (rural/urban), occupation (10 categories), and/or information provided in the text of each paper; 13.7% of observations could not be classified and were assigned to the middle category. In the final regression analyses, the middle and higher SES categories were combined into a single group to simplify interpretation.

Results

The mean BMI, weight, and height for this sample of studies are shown in table 3 according to region. BMI is lowest in India (20.1) and the rest of South and South-East Asia (20.2), followed by sub-Saharan Africa (21.2), China and Australasia (22.1), North Africa and middle Asia (23.2), and Latin America and the Caribbean (23.3). Polynesia and Micronesia has the highest mean BMI (28.5), which stands out from the rest of the sample.

Table 4 presents the results of three hierarchical regression models to estimate the influence of the two random effects variables (country and paper) and the two fixed effects of greatest interest here (year and region). As revealed by the variance estimates in model 1, country and paper do contribute significantly to the observed variance in BMI. In model 2 it can be seen that the addition of region reduces the apparent influence of country, but country and paper both remain significant variables. Region itself is highly significant in model 2 (F=16.89, p<.0001), and the coefficients for each region conform to the same ranking of regions as that shown in table 3. Model 3 reveals that there is a significant positive coefficient for the variable year (p=.0043), suggesting that for the developing world as a whole, mean BMI has increased by 0.038 kg/m² per year during the period from 1957 to 1994. This implies an increase of 1.41 kg/m² over the entire period, with a 95% confidence interval of 0.44 to 2.37 kg/m².

Table 5 presents the F ratios and significance levels for regression models that incorporate an increasing number of main effects and interaction terms. In reviewing the results of these models, particular attention is given to the main effect of year and its interaction with other main effects, in order to understand the ways in which the temporal trends in mean BMI may vary across regions, age and sex groups, and socio-economic groups. Note that the middle and upper socio-economic groups have been merged into a single category in the results presented here, to simplify the interpretation.

TABLE 3. BMI, weight, and height according to region

Region	Na	BMI (kg/m2)	Weight (kg)	Height (cm)	Nb
Sub-Saharan Africa	312	21.2±2.5	55.7±6.3	164.8±6.9	196
South and South-East Asia	156	20.2±1.8	49.3±5.1	156.8±7.0	50
India	203	20.1±2.4	47.3±8.0	156.2±9.2	102
China	128	22.1±1.4	52.7±3.1	161.8±6.2	6
Australasia	201	22.1±2.3	51.2±7.0	154.2+6.6	116
Polynesia and Micronesia	127	28.5±3.2	73.0±8.4	164.5±6.4	42
Latin America and Caribbean	231	23.3±1.9	56.4±6.8	156.1±7.4	70
North Africa and middle Asiac	54	23.2±2.0	56.6±6.4	158.6±7.9	23
All regions	1,412	22.2±3.2	54.1±9.2	159.4±8.6	605

Plus-minus values are means ± SD.

a. Sample sizes refer to the total number of observations in the data set, with the number of observations from each published paper being determined by the number of persons of each age, sex, and SES for which BMI, weight, or height was reported.

b. Sample sizes refer to observations reporting mean weight and height instead of, or in addition to, mean BMI or BMI prevalences.

c. This region is excluded from subsequent analyses because most observations arise from only three countries, each measured at only one point in time.

Model	Independent variable	Coefficient	P
1	Country	5.03	.0001
	Paper	1.94	.0001
	Residual	2.22	.0001
	Intercept	22.59	.0001
2	Country	1.58	.0015
	Paper	2.02	.0001
	Residual	2.18	.0001
	Intercept	20.81	.0001
	Sub-Saharan Africa	0.74	.2709
	South and South-East Asia	-0.11	.8414
	China (full)8	-0.13	.8479
	Australasia	2.32	.0001
	Polynesia and Micronesia	6.59	.0001
	Latin America and Caribbean	2.66	.0001
	India (reference)	-	-
3	Country	1.64	.0012
	Paper	2.00	.0001
	Residual	2.17	.0001
	Intercept	20.61	.0001
	Sub-Saharan Africa	0.74	.2748
	South and South-East Asia	-0.12	.8306
	China (full)a	-0.16	.8117
	Australasia	2.28	.0001
	Polynesia and Micronesia	6.66	.0001
	Latin America and Caribbean	2.73	.0001
	India (reference)	-	-
	Year	.038	.0043

^a Includes all data points from China.

Model 5 includes an interaction term between region and year, to test whether the apparent increase in mean BMI over time varies across regions. The F ratio for the interaction term is statistically significant and restores the F ratio for year to statistically significant levels (F=7.14, p =.0077). Model 6 further includes the interactions between region and each of the other explanatory variables (age, sex, and SES). The F ratio for each of these is highly significant, indicating that these variables are differentially associated with mean BMI across the regions. The F ratio for the year main effect remains significant (F=4.36,p=.0370).

Model 7 includes the three-way interactions involving year and region with each of the other variables in turn (age, sex, and SES). The three-way interactions involving sex and age-squared are not statistically significant. Those involving SES and age are significant, indicating that the region-specific time trends in mean BMI vary with age and SES group. The main effect of year remains statistically significant in this model (F=5.59, p>=.0182). Model 8 includes the two-way interactions between year and each of the other explanatory variables, along with two-way interactions involving age, sex, and SES. The statistically significant interactions among this group are year by age, age by SES, and SES by sex. As before, the year main effect remains statistically significant (F=4.63, p=.0316).

In order to examine the implications of these results for time trends in mean BMI, a series of predicted means was generated for men and women from each region, stratifying by the SES variables. The results from model 6 were used for this purpose in order to incorporate the strong effects of the interactions in that model. Although some of the interaction terms in models 7 and 8 are also significant, the precision of the time-trend estimates from these models is likely to be much lower, and therefore they were not used for this exercise. The results from model 6 were used to estimate the mean BMI levels in various groups in the years 1960 and 1990. The SES stratification contrasts the low socio-economic group with all others (middle and high combined). The results are shown in table 6 and figure 1.

TABLE 5. Results of hierarchical regression analysis on BMI incorporating age, sex, SES, and interactions (F values)^a

Independent variable	df	Model
		3	4	5	6	7	8
Year	1	8.19***	2.97*	7.14***	4.36**	5.59**	4.63**
Region	6	16.87****	18.82****	15.67****	16.03****	14.88****	14.56****
Sex	1		37.87****	38.19****	50.80****	16.25****	14.22****
SES	1		179.45****	174.14****	186.14****	43.21****	49.11****
Age	1		2.71*	2.58	2.84	0.00	0.00
Age2	1		83.49****	83.76****	100.81****	57.55****	54.66****
Year x region	6			2.60**	1.9*	2.00*	2.06*
Sex x region	6				16.27****	15.61****	15.05****
SES x region	6				6.67****	2.21**	2.10**
Age x region	6				12.90****	9.24****	6.21****
Age2 x region	6				6.84****	6.13****	6.37****
Year x region x sex	7					1.75*	2.13**
Year x region x SES	7					2.12**	2.59**
Year x region x age	7					2.23***	1.92*
Year x region x age2	7					0.09	0.23
Year x age	1						2.77*
Year x age2	1						.28
Year x sex	1						1.49
Year x SES	1						.01
Age x sex	1						.60
Age2 x sex	1						.44
Age x SES	1						28.05****
Age2 x SES	1						1.86
Sex x SES	1						15.83****

a. F values refer to type III sums of squares in which all other model variables are controlled simultaneously. All models include country and paper as random effects, as in table 6, which were significant in all cases. Includes all data points from China.

p >.100 for figures with no asterisk.
*.05<p<.10.
**.01<p<.05.
***.001<p<.01.
*”* p<.001.

It should be noted that the results for China described above are based on all data points available for this analysis, i.e., the row labeled “China (full).” Model 6 was also tested after excluding eight data points from the early 1970s in China, which are the only other data points available for that country before the 1980s. The deletion of these data points has a dramatic effect on the slope estimate (B=1.477) and on the estimates of mean BMI for China in 1960 and 1990. This suggests the possibility that a modest increase in BMI may have taken place during the 1970s in China, but a much more rapid increase took place during the 1980s.

TABLE 6. Predicted mean BMI according to region, year, sex, and SES (from model 6)

Region	Slopea	High SES		Low SES
		1960	1990	1960	1990
Women
Sub-Saharan Africa	.207	23.2	23.8	21.4	22.0
South and South-East Asia	.169	20.9	21.4	20.3	20.8
China (full)b	.106	20.1	23.3	18.0	21.1
India	.048	21.6	21.8	19.5	19.7
Australasia	-.295	24.1	23.2	22.7	21.9
Polynesia and Micronesia	.957	27.9	30.8	25.7	28.6
Latin America and Caribbean	.112	24.8	25.1	23.8	24.1
China (1980s)	1.477	18.5	22.9	17.7	22.1
Men
Sub-Saharan Africa	.207	22.1	22.7	20.3	20.9
South and South-East Asia	.169	20.9	21.4	20.4	20.9
China (full)b	.106	19.8	22.9	17.4	22.2
India	.048	21.5	21.6	19.4	19.5
Australasia	-.295	24.6	23.7	23.2	22.4
Polynesia and Micronesia	.957	26.0	28.9	23.8	26.6
Latin America and Caribbean	.112	24.2	24.6	23.2	21.8
China (1980s)	1.477	18.1	22.6	17.4	21.8

a Represents the estimated change in mean BMI per decade.
b Includes all data points from China.

FIG. 1. Predicted mean BMI according to region, year, SES, and sex (from model 6) (b)

FIG. 1. Predicted mean BMI according to region, year, SES, and sex (from model 6) (c)

FIG. 1. Predicted mean BMI according to region, year, SES, and sex (from model 6) (d)

TABLE 7. Calculated prevalence of underweight, overweight, and obesity according to region, for high-SES women and low-SES men

Region	Underweight (BMI < 18.5 kg/m2)		Overweight (BMI > 27 kg/m2)		Obesity (BMI > 30 kg/m2)
	1960	1990	1960	1990	1960	1990
High-SES women
Sub-Saharan Africa	7.1	6.1	11.7	17.4	1.7	3.5
South and South-East Asia	15.1	12.5	0.5	1.3	0.0	0.0
China (full)	21.8	7.0	0.0	12.7	0.0	1.9
India	11.7	10.4	1.9	2.6	0.1	0.1
Australasia	5.7	7.1	20.6	11.7	4.8	1.7
Polynesia and Micronesia	2.9	2.1	57.1	73.6	33.7	55.2
Latin America and Caribbean	4.9	4.6	28.1	31.2	8.5	10.5
China (1980s)	50.0	7.8	0.0	9.2	0.0	1.1
Low-SES men
Sub-Saharan Africa	19.8	15.1	0.1	0.5	0.0	0.0
South and South-East Asia	19.0	15.1	0.1	0.5	0.0	0.0
China (full)	35.8	10.5	0.0	4.5	0.0	0.3
India	30.5	29.1	0.0	0.0	0.0	0.0
Australasia	7.1	8.9	11.7	5.6	1.7	0.4
Polynesia and Micronesia	6.1	3.5	17.4	47.4	3.5	22.4
Latin America and Caribbean	7.1	6.4	11.7	15.6	1.7	2.8
China (1980s)	35.8	11.0	0.0	2.6	0.0	0.1

Finally, the data in table 6 were converted into order-of-magnitude estimates of the prevalence of underweight, overweight, and obesity to gain some perspective on the extent to which changes in mean BMI are associated with changes in prevalences. This conversion was accomplished through the use of normal curve properties after making some assumptions about the standard deviation of BMI.* Cut-off values of 18.5 and 30.0 were used for estimating underweight and obesity, respectively [12], and an arbitrary cut-off value of 27.0 was used for estimating overweight. This analysis was performed for high-SES women and low-SES men, which represent the groups with the highest and lowest mean BMIs, respectively, in table 6. As shown in table 7, the estimated prevalence of underweight is far greater than the prevalence of obesity in most of the world, even in 1990. This is seen in sub-Saharan Africa, South and South-East Asia, China, India, and (among low-SES men) Latin America. High-SES women in Latin America and the Caribbean are an exception, in that the estimated prevalence of obesity is somewhat greater than the prevalence of underweight. Polynesia and Micronesia also differ (markedly so) from the rest of the world, in displaying much higher prevalences of obesity than underweight. Even when the much lower cut-off point of 27.0 is used, these estimates suggest that underweight remains the more common condition among low-SES men in most regions, even in 1990. Among high-SES women, however, the use of this lower cut-off point creates the appearance that overweight is surpassing underweight as the more common condition in sub-Saharan Africa, China, Australasia, and (for both sexes) Latin America. This lower cut-off point has little effect on the prevalence estimates in India and South and South-East Asia.

* The standard deviation of BMI was assumed to vary in a linear fashion from 2.0 in low-BMI populations (i.e, BMI=20, as in Norgan [11]) to 3.5 in medium-BMI populations (i.e., BMI=24, as in WHO [12], p. 337) to 5.0 in high-BMI populations (i.e., BMI=28, as in Pawson [13]). By further assuming that BMI is normally distributed, the normal variate (Z) was calculated for each mean BMI shown in table 6 as follows:

Z=(cut-off - mean BMI)/[(.375 x mean BMI) - 5.5]

The percentages of individuals with BMIs above or below the cut-offs specified in table 7 were then estimated using tabled values for the cumulative normal frequency distribution.

Discussion

The present analysis represents a second-best strategy for estimating time trends in BMI in developing countries, the choice of which is necessitated by the virtual absence of large-scale, representative surveys at two points in time in such countries. That said, the approach has several strengths, including the existence of a large number of small-scale surveys from a wide variety of countries in all major regions; the availability of ancillary information on age, sex, and socio-economic characteristics to refine the estimate of time trends and explore a variety of interactions; and the use of a hierarchical modelling procedure that permits the statistical removal of any “extraneous” variation in BMI that may be associated with a particular country or survey (e.g., differences in level of economic development, genetic or climate-related influences on adult physique, and survey methods that may vary across studies).

If the results concerning time trends had supported the null hypothesis (i.e., no evidence of trends in BMI), they would have had numerous possible explanations, notably the existence of excessive random error in the mean BMI data. As it stands, the results suggest an increase in mean BMI for most regions of the world, which is unlikely to have arisen due to random error in the mean BMI data. Thus, the only serious threat to the validity of the time-trend findings would be if more recent surveys in each country or region were purposely conducted (or published) in populations at risk for higher mean BMI or obesity, whereas the older studies were conducted in populations less prone to obesity. In all likelihood, this would have taken the form of more frequent surveys in the urban or high-SES in recent years, but the use of an SES index in the analysis serves as a control against this possibility. It is notable (from table 5) that the main effect for year remains statistically significant, regardless of whether the SES main effect or its interaction with year is included in the model.

Of perhaps greater importance for interpreting the present results are the biological considerations related to BMI and, in particular, the extent to which increases in obesity can be inferred from these findings. BMI is a convenient indicator of obesity at high cut-off points (e.g., a value of 30 as recommended by WHO [12]), but it is much less useful at lower values when applied at an individual level. For instance, among 138 male Italian shipyard workers with an average BMI of 25.3 kg/m² and 22.3% body fat (estimated through densitometry), the standard deviation in percentage of body fatness at a given BMI was 4.0% [14]. Although BMI in that sample was highly correlated with percentage of body fat (r=.75), it was equally correlated with fat-free mass (r=.68). Similar results are observed among men and women from Papua New Guinea and Ethiopia, whose mean BMIs are at the lower end of the range for developing countries seen in the present study (19-22 kg/m²) [11]. Thus, among adults in developing countries, variation in BMI may be as much a reflection of variation in fat-free mass as fat mass and cannot be reliably used to infer percentage of body fat at the individual level across the low-to-moderate range of BMI values.

In contrast to these interpretational difficulties associated with cross-sectional, inter-individual variation within the moderate range of BMI values, a different set of considerations applies to this paper, which focuses on the group level and on trends in mean BMI over time. First, through an analysis of 285 samples of adult men and women from developing countries, Norgan [15] concluded that, at least over the range of BMIs from 20 to 25 kg/m², the slight inter-population differences in the relationship between BMI and body composition are unlikely to be meaningful in epidemiologic studies. This conclusion is even more valid in the present case, because the focus is on trends over time within countries and regions, rather than cross-sectional differences among countries. Second, at the population level, the mean BMI is an indicator of the central tendency of a distribution of individual BMIs, with some individuals exceeding conventional cut-off points for obesity (e.g., 30 kg/m²). Groups with relatively high mean BMI are likely to have a higher prevalence of obesity thus defined. The only caveat would be if the variance in BMI becomes compressed as mean BMI increases, but the common observation is just the opposite: groups with high mean BMI tend to have an expanded variance above the median due to greater variance in the upper tail, suggesting that the prevalence of obesity in a population increases in an accelerating fashion as the mean BMI increases (see footnote on page 231).

Applying a population-level interpretation to time trends in mean BMI provides some basis for investigating possible increases in obesity. As shown in table 6, mean BMI appears to have increased in most regions of the world from 1960 to 1990 (with the exception of Australasia), although the size of this increase varies widely across regions. Although this might seem to support the notion that obesity is increasing in the developing world, the results in table 7 suggest otherwise. Specifically, those results suggest that the prevalence of obesity, as defined by a cut-off BMI value of 30 kg/m², is not estimated to have increased substantially in most major regions of the world during this period. The reason for this apparent contradiction is that the mean BMI of adults in most regions remains well below the cut-off point of 30 kg/m², even in 1990. These results further suggest that the prevalence of underweight (defined as BMI< 18.5 kg/m²) has shown a slight decline in most regions but remains far higher than the prevalence of obesity.

For reasons given above, the results of the hierarchical regression analysis that is the foundation for this study are likely to be valid for estimating the overall trends in mean BMI in the past three decades, but there is likely to be much lower precision in the estimates of region-specific trends (because of a smaller number of observations in each region). Thus, the most conservative and defensible conclusion from these data is simply that mean BMI has increased in developing countries since 1957 and that this probably reflects a slight increase in overweight (as defined by a moderate cutoff point of BMI > 27 kg/m²) in certain regions of the developing world, but the increases in mean BMI seen to date do not appear to be associated with an increase in obesity (as defined by the recommended cut-off point of BMI > 30 kg/m²). To the contrary, the increase in mean BMI is likely to be associated with a slight decline in underweight (defined as BMI<18.5 kg/m²), but underweight remains the more common condition in most developing regions.

Notwithstanding the above conclusions, four caveats are in order. First, the overall trends suggested in this study refer to large geographic regions, and in no way do they preclude the possibility that obesity exists and is on the increase in smaller geographic areas and/or socio-economic groups. Second, this study does provide evidence that the general upward trend in mean BMI is likely to continue in the future if the underlying behavioural and socio-economic trends persist. This is suggested by the strong cross-sectional association between mean BMI and the socio-economic index in the present study, which is especially marked in certain regions. Third, the cut-off values used here to define underweight and obesity conform to WHO recommendations [12], but these precise cut-off values are not equivalent in statistical and functional terms (e.g., as risk factors for infectious and/or chronic disease in different environments), and the social implications of these two conditions differ markedly (e.g., in terms of social equity in access to food). Finally, this study (specifically, table 7) underscores the common knowledge that estimates of the prevalence of “obesity” (and changes over time) can vary widely, depending upon the cut-off point used to define that condition. The use of low cut-off points to define obesity (e.g., as low as 25 kg/m² in some papers) may create the appearance of a current or emerging obesity problem in developing countries and contribute to a shift in policy attention from undernutrition to overnutrition when the social and health conditions may not warrant it.

Given the long time lags required for the development and implementation of preventive policies and programmes, the strong association between socio-economic factors and mean BMI in this study does reinforce the belief that developing countries eventually will face greater heterogeneity in nutrition-related problems. One of the challenges this creates at the policy level is to continue to find ways to meet the energy and nutrient needs of the undernourished segments of rural and urban populations in developing countries, while simultaneously ensuring healthful diets and physical activity patterns to prevent chronic disease in some of these same populations [16]. Although several categories of policy instruments have been identified for intervening in the quality of the diet of populations [17], in the current policy climate the more powerful of these may be construed as trade barriers and/or questionable interference in domestic markets (e.g., legislation, regulation, taxes, and subsidies). In addition, given the administrative weaknesses in many developing countries, it may be difficult to design these instruments in such a way that control of one form of malnutrition (e.g., protein-energy malnutrition) is not compromised by attempts to reduce or prevent another form (e.g., chronic disease-related malnutrition). The reconciling of these trade-offs at the analytical and operational levels is likely to be one of the key challenges for nutrition policy in the coming decades.

Acknowledgements

The work described here is an extension of research conducted for the International Food Policy Research Institute as part of the 2020 Vision Initiative. The assistance of Dr. Laura Kettel-Khan and Dr. Ed Frongillo is gratefully acknowledged.

References

1. Bulatao RA. Mortality by cause. In: Gribble JN, Preston SH, eds. The epidemiological transition: policy and planning implications for developing countries. Washington, DC: National Academy Press, 1993:42-68.

2. Murray CJL, Lopez AD. Alternative projections of mortality and disability by cause 1990-2020: global burden of disease study. Lancet 1997;349:1498-1504.

3. World Bank. The world development report 1993: investing in health. New York: Oxford University Press, 1993.

4. Jamison DT, Mosley WH, Measham AR, Bobadilla JL, eds. Disease control priorities in developing countries. Oxford: Oxford University Press, 1993.

5. Gribble JN, Preston SH, eds. The epidemiological transition: policy and planning implications for developing countries. Washington, DC: National Academy Press, 1993.

6. Murray CJL, Yang G, Qiao X. Adult mortality: levels, patterns and causes. In: Feachem RGA, Kjellstrom T, Murray CJL, Over M, Phillips MA, eds. The health of adults in the developing world. New York: Oxford University Press, 1992:23-112.

7. Bulatao RA, Stephens PW. Global estimates and projections of mortality by cause, 1970-2015. Working Paper. Washington, DC: World Bank, 1992.

8. Pelletier DL, Gage T. Recent trends in obesity and chronic disease in developing countries. Washington, DC: International Food Policy Research Institute, 1995.

9. Gurney M, Gorstein J. The global prevalence of obesity - an initial overview of available data. World Health Stat Quart 1990:251-4.

10. Statistical Analysis System (SAS), version 6.0. Cary, NC, USA: SAS.

11. Norgan NG. Body mass index and body energy stores in developing countries. Eur J Clin Nutr 1990;44:79-84.

12. World Health Organization. Physical status: the use and interpretation of anthropometry. Technical report series no. 854. Geneva: WHO, 1995.

13. Pawson IG. The morphological characteristics of Samoan adults. In: Baker PT, Hanna JM, Baker TS, eds. The changing Samoans: behavior and health in transition. Oxford: Oxford University Press, 1986:254-74.

14. Norgan NG, Ferro-Luzzi A. Weight-height indices as estimators of fatness in men. Hum Nutr Clin Nutr 1982;36C:363-72.

15. Norgan NG. Population differences in body composition in relation to the body mass index. Eur J Clin Nutr 1994;48:S10-S27.

16. Mosley WH, Bobadilla JL, Jamison DT. The health transition: implications for health policy in developing countries. In: Jamison DT, Mosley WH, Measham AR, Bobadilla JL, eds. Disease control priorities in developing countries. Oxford: Oxford University Press, 1993:673-99.

17. Milio N. Nutrition policy for food-rich countries: a strategic analysis. Baltimore, Md, USA: Johns Hopkins University Press, 1990.

Appendix. Data sources

Abamtp ZU, de Guzman PE, Dona’o DS, Cuaderna PC, Jandayan MO, Agustin CP. Height and weight measurements of high school students in Central Luzon. Philippine J Nutr 1981;34(1):21-27,48.

Afifi ZEM. Determinants of growth of infants in an Egyptian village: maternal anthropometry, birth interval, solid food and death of siblings. Hum Biol 1985;57:649-58.

Al-Shoshan AA. The affluent diet and its consequences: Saudi Arabia - a case in point. World Rev Nutr Diet 1992;69:113-165.

Akinkugbe OO. Epidemiology of cardiovascular disease in developing countries. J Hypertens 1990;8:S233-8.

Ayatollahi SMT, Carpenter RG. Height, weight, BMI and weight-for-height of adults in southern Iran: How should obesity be defined? Ann Hum Biol 1993,20:13-9.

Bailey S, Campos H, Schcsinsky K, Mata L. Relationship of upper body fat distribution to serum glucose and lip-ids in a Costa Rican population. Am J Phys Anthropol 1987;73:111-7.

Beall CM, Goldstein MC. High prevalence of excess fat and central fat patterning among Mongolian pastoral nomads. Am J Hum Biol 1992;4:747-56.

Beevers DG, Prince JS. Hypertension: an emerging problem in tropical countries. R Soc Trop Med Hyg 1991;85:324-6.

Benefice E, Simondon K. Agricultural development and nutrition among rural populations: a case study of the middle valley in Senegal. Ecol Food Nutr 1993;31:45-66.

Berdasco A. Body mass index values in the Cuban adult population. Eur J Clin Nutr 1994;48:S155-64.

Berrios XC, Jadue LH, Zenteno JA, Ross MIA, Rodriguez RP. Prevalence of risk factors for chronic diseases: a population study in the metropolitan area of Santiago, Chile. Rev Med Chile 1990;118:597-604.

Berry JN, Khattri HN, Bidwai PS, Gupta HD. Height and weight of male college and polytechnic students in northern India. Indian J Med Res 1968;56:1783-99.

Berry JN, Sarvotham SG, Gupta HD. Skinfold thickness and obesity in North Indian urban men and women. J Assoc Physicians India 1973;21:743-51.

Bharati P. Variation in adult body dimensions in relation to economic condition among the Mahishyas of Howrah district, West Bengal, India. Ann Hum Biol 1989;16:529-41.

Bhuiya A, Mostaafa G. Levels and differentials in weight, height and body mass index among mothers in a rural area of Bangladesh. J Biosoc Sci 1993;25:31-8.

Bindon JR, Baker PT. Modernization, migration and obesity among Samoan adults. Ann Hum Biol 1985;12:67-76.

Bindon JR, Crews DE. Changes in some health status characteristics of American Samoan men: preliminary observations from a 12-year follow-up study. Am J Hum Biol 1993:5:31-7.

Bleiberg FM, Brun TA, Goihman S. Duration of activities and energy expenditure of female farmers in dry and rainy seasons in upper-Volta. Br J Nutr 1980;43:71-82.

Borghesi LL, Maiz AG, Arteaga ALI. Correlation of changes in body weight and cardiovascular risk factors. Rev Med Chile 1992;120:741-7.

Bourne LT, Langenhoven ML, Steyn K, Jooste PL, Laubscher JA, Van der Vyver E. Nutrient intake in the urban African population of the Cape Peninsula, South Africa. The Brisk Study. Cent Afr J Med 1993;39:238-47.

Branca F, Pastore G, Demissie T, Ferro-Luzzi A. The nutritional impact of seasonality in children and adults of rural Ethiopia. Eur J Clin Nutr 1993;47:840-50.

Brush G, Boyce AJ, Harrison GA. Associations between anthropometric variables and reproductive performance in a Papua New Guinea highland population. Ann Hum Biol 1983;10:223-34.

Bursztyn PG. A diet survey in Zimbabwe. Hum Nutr Appl Nutr 1985;39A:376-88.

Campbell TC, Junshi C, Brun T, Parpia B, Yinsheng QU, Chumming C, Geissler C. China: from diseases of poverty to diseases of affluence. Policy implications of the epidemiological transition. Ecol Food Nutr 1992;27:133-44.

Castle WM. Coronary heart disease risk factors in black and white men in Zimbabwe and the effect of living standards. S Afr Med J 1982;61:926-9.

Chadha SL, Radhakrishnan S, Ramachandran K, Kaul U, Gopinath N. Epidemiological study of coronary heart disease in an urban population of Delhi. Indian J Med Res 1992;92:424-30.

Chadha SL, Gopinath N, Ramachandran K. Epidemiological study of coronary heart disease in Gujaratis in Delhi (India). Indian J Med Res 1992;96:115-21.

Collins V, Dowse G, Zimmet P. Prevalence of obesity in Pacific and Indian Ocean populations. London: Elsevier, 1990:29-32.

Crews DE. Body weight, blood pressure and the risk of total and cardiovascular mortality in an obese population. Hum Biol 1988:60:417-33.

Curran-Everett LS. Accordance between V_O2max and behavior in Ngisonyoka Turkana. Am J Hum Biol 1994;6:761-71.

Daly J, Tsomondo E, Jones J. Height, weight and skin fold thickness of African primary school children and adult men. Cent Afr J Med 1973;19:234-6.

Darlu P, Couilliot MF, Drupt F. Ecological and cultural differences in the relationships between diet, obesity and serum lipid concentrations in a Polynesian population. Ecol Food Nutr 1984;14:169-83.

Da Rocha FJ, Salzano FM. Anthropometric studies in Brazilian Cayapo Indians. Am J Phys Anthropol 1972;36:95-102.

Das Gupta P, Mukherjee BN, Das SK, Pakrasi K, Mukherjee D. A study of anthropometry and genetic markers amongst the Kaoras of West Bengal. Indian J Phys Anthropol Hum Genet 1981;7:19-37.

Date C, Baba M, Okuda T, Kajiwara NM, Minamide T, Fujita Y, Ichikawa M, Miyatani S, Hayashi M, Tanaki H, Heywood P, Alpers M, Koishi H. Nutritional status of some Papua New Guinea highlanders as assessed by physical measurements and blood analysis. Ecol Food Nutr 1988;20:185-96.

Day J, Carruthers M, Bailey A, Robinson D. Anthropometric, physiological and biochemical differences between urban and rural Maasai. Atherosclerosis 1976,23:357-61.

Day J, Bailey A, Robinson D. Biological variations associated with change in lifestyle among the pastoral and nomadic tribes of East Africa. Ann Hum Biol 1979;6:29-39.

Davies CTM, Mbelwa D, Crockford G, Weiner JS. Exercise tolerance and body composition of male and female Africans aged 18-30 years. Hum Biol 1973:45:31-40.

Delisle H, Alladoumgue M, Begin F, Nandjingar K, Lasorsa C. Household food consumption and nutritional adequacy in Wadi sones of Chad, Central Africa. Ecol Food Nutr 1991;25:229-48.

Delpeuch F, Cornu A, Massamba JP, Traissac P, Maire B. Is body mass index sensitively related to socio-economic status and to economic adjustment? A case study from the Congo. Eur J Clin Nutr 1994;48:S141-7.

Dettwyler KA. Nutritional status of adults in rural Mali. Am J Phys Anthropol 1992;88:309-21.

de Vasconcellos MTL. Body mass index: its relationship with food consumption and socioeconomic variables in Brazil. Eur J Clin Nutr 1994;48:S115-23.

De Villiers H. A study of morphological variables in urban and rural Venda male populations. Hum Biol Environ Change 1971;110-13.

Dhurandhar NV, Kulkarni PR. Prevalence of obesity in Bombay Int J Obes 1992;16:367-75.

Diaz E, Gonzalez-Cossio T, Rivera J, Immink MDC, Mendoza RD, Flores CR. Body composition estimates using different measurement techniques in a sample of highland subsistence farmers in Guatemala. Am J Hum Biol 1991;3:525-30.

Donaldson JF. Changes in African disease patterns with western acculturation. Cent Afr J Med 1971;17:51-4.

Dowse G, Collins G, Zimmet P, Finch C. Cross-sectional and longitudinal relationship between obesity, hypertension and coronary heart disease in Micronesian Nauruans. London: Elsevier, 1990:179-83.

Dua A, Seth V. Obesity-prevalence and association with food behaviour in married women (25-40 years). Indian J Nutr Diet 1988;25:338-44.

Durnin JVGA, Drummond S, Satyanarayana K. A collaborative EEC study on seasonality and marginal nutrition: the Glasgow Hyderabad (S. India) Study. Eur J Clin Nutr 1990;44:19-29.

Egun GN, Atinmo T. Protein requirement of young adult Nigerian females on habitual Nigerian diet at the usual level of energy intake. Br J Nutr 1993;70:439-48.

Egusa G, Bennett PH, Aleck K, Taylor R, Howard BV. Hyperlipemia and arteriosclerotic cardiovascular disease in the Polynesian population of Rarotonga. Atherosclerosis 1984;53:241-54.

Eveleth PB, Tanner JM. Worldwide variation in human growth. Cambridge: Cambridge University Press, 1976.

Ferro-Luzzi A, Scaccini C, Taffese S, Aberra B, Demeke T. Seasonal energy deficiency in Ethiopian rural women. Eur J Clin Nutr 1990;44:7-18.

Francois PJ, James WPT. An assessment of nutritional factors affecting the BMI of a population. Eur J Clin Nutr 1994;48:S110-S4.

Frisancho AR, Baker PT. Altitude and growth: a study of the patterns of physical growth of a high altitude Peruvian Quechua population. Am J Phys Anthropol 1970;32:279-92.

Froment A, Hiernaux J. Climate-associated anthropometric variation between populations of the Niger bend. Ann Hum Biol 1984;11:189-200.

Garcia M, Kennedy E. Assessing the linkages between low body mass index and morbidity in adults: evidence from four developing countries. Eur J Clin Nutr 1994;48:S90-7.

Garcia PA, Luther T, Serrano FJ, Tan AG. Nutrition and health of the rural elderly in five selected Barangays in the Philippines. Nutr Res 1982:2:545-60.

Garcia-Palmieri MR, Tillotson J, Cordero E, Costas R, Sorlie P, Gordon T, Kannel WB, Colon AA. Nutrient intake and serum lipids in urban and rural Puerto Rican men. Am J Clin Nutr 1977;30:2092-100.

Gardner GW, Edgerton VR, Senewiratne B, Barnard RJ, Ohira Y. Physical work capacity and metabolic stress in subjects with iron deficiency anemia. Am J Clin Nutr 1977;30:910-17.

Garg BS, Singh JV. Nutritional status of aged in an Urban area. Indian J Nutr Diet 1983;20:316-22.

Geppert J, Splett PL. Summary document of nutrition intervention in obesity J Am Diet Assoc 1991;S31(5):31-5.

Giay T, Khoi HH. Use of body mass index in the assessment of adult nutritional status in Vietnam. Eur J Clin Nutr 1994;48:S124-30.

Glanville EV, Geerdink RA. Skinfold thickness, body measurements and age changes in Trio and Wajna Indians of Surinam. Am J Phys Anthropol 1970:32:455-562.

Gopaldas T, Gupta A, Saxena K. The impact of sanskritization in a forest-dwelling tribe of Gujarat, India. Ecology, food consumption patterns, nutrient intake, anthropometric, clinical and hematological status. Ecol Food Nutr 1993;12:217-27.

Harrison GA, Kuchemann CF, Moore MAS, Boyce AJ, Baju T, Mourant AE, Godber MJ, Glasgow BG, Kopec AC, Tills D. The effects of altitudinal variation in Ethiopian populations. Phil Trans R Soc Lond 1969;256:147-82.

Harrison GG, Galal OM, Ritenbaugh C, Shaheen FM. Wahba SAA, Kirksey A, Jerome NW. Dependability and precision of anthropometric measures in a longitudinal field study in an Egyptian village. Am J Hum Biol 1991,3:479-87.

Hiernaux J. Heredity and environment: their influence on human morphology. A comparison of two independent lines of study Am J Phys Anthropol 1963:21:575-89.

Hiernaux J, Rudan P, Brambati A. Climate and the weight/height relationship in sub-Saharan Africa. Ann Hum Biol 1975;2:3-12.

Hiernaux J, Hartono DB. Physical measurements of the adult Hadza of Tanzania. Ann Hum Biol 1980;7:339-46.

Hornabrook RW, Serjeantson S, Stanhope JM. The relationship between socioeconomic status and health in two Papua New Guinea populations. Hum Ecol 1977;5:369-82.

Hornick CA, Fellmeth BD. High density lipoprotein cholesterol, insulin and obesity in Samoans. Atherosclerosis 1981;39:321-8.

Hoyos MD, Fraser HS. Obesity and the use of health services in a Barbadian community. Hum Nutr Clin Nutr 1984;38C:77-8.

Htay H, Po L, Mya-Tu M. A comparative study of the physical performance capacity of rural and urban Burmese women and aged 19-24 years. Southeast Asian J Trop Med Public Health 1976:7:452-9.

Huizinga J. Human biological observations on some African populations of the thorn Savanna belt 1. Kroninklijke Nederlandse Akademie Van Wetenschappen Proceedings 1968;71:356-72.

Huss-Ashmore R, Curry JJ. Diet, nutrition and agricultural development in Swaziland: agricultural ecology and nutritional status, Ecol Food Nutr 1989;23:189-209.

Huss-Ashmore R, Goodman JL, Sibiya TE, Stein TP. Energy expenditure of young Swazi women as measured by the doubly-labelled water method. Eur J Clin Nutr 1989:43:737-48.

Immink MDC, Flores RA, Diaz EO. Body mass index, body composition and the chronic energy deficiency classification of rural adult populations in Guatemala. Eur J Clin Nutr 1992:46:419-27.

Inaoka T, Suzuki T, Ohtsuka R, Kawabe T, Akimichi T, Takemori K, Sasaki M. Salt consumption, body fatness and blood pressure of the Gidra in Lowland Papua. Ecol Food Nutr 1987:20:55-6.

Ivanovic D, Alvarez MDLL, Barreray G, Muzzo S. Influence of the socioeconomic level on nutritional status of students graduated from basic and high school. Rev Med Chile 1984:112:1165-71.

Ivanovic D, Olivares M, Ivanovic R. Nutritional status of school children in the metropolitan area of Santiago, Chile. Rev Med Chile 1990:118:916-24.

James WPT. Introduction: the challenge of adult chronic energy deficiency. Eur I Clin Nutr 1994;48:S1-S9.

Johnston FE, Gindhart PS, Jantz RL, Kensinger KM, Walker GF. The anthropometric determination of body composition among the Peruvian Cashinahua. Am J Phys Anthropol 1971:34:409-16.

Jones JJ. A comparative study of the prevalence of adult obesity in the three racial groups of Kuala Lumpur. Med J Malaysia 1976;30(4):256-60.

Katsivo MN, Apeagyie F. Hypertension in Kitui district: a comparative study between urban and rural populations. E Afr Med J 1991:68:531-8.

Kauffmann RQ, Hidalgo CR, Roessler EB, Aicardi VS, Bunout DB, Rodriguez RS, Canas LC. Preventive health examination in asymptomatic workers: design and initial results of an institutional program. Rev Med Chile 1991:119:617-25.

Kaur B, Singh R. One year follow-up study of stature, weight, emergence of dentition, and sexual maturation of well-nourished Indian girls from birth to 20 years. Am J Hum Biol 1994:6:425-36.

Kaur DP, Singh R. Parent-adult offspring correlations and heritability of body measurements in a rural Indian population. Ann Hum Biol 1981:8:333-9.

Kesterloot H, Oviasu VO, Obasohan AO, Olomu A, Cobbaert C, Lissens W. Serum lipid and apolipoprotein levels in a Nigerian population sample. Atherosclerosis 1989:78:33-8.

Knuiman JT, West CE, Burema J. Serum total and high density lipoprotein cholesterol concentrations and body mass index in adult men from 13 countries. Am J Epidemiol 1982:116:631-42.

Koong SL, Malison MD, Nakashima AK. A prevalence survey of behavioural risk factors in Taipei City, Taiwan. Int J Epidemiol 1990:19:154-9.

Kullah KM, Ramnath T. Nutritional status of the aged in rural areas of Andhra Pradesh. Indian J Nutr Diet 1985:22:330-6.

Kutty VR, Balakrishnan KG, Jayasree AK, Thomas J. Prevalence of coronary heart disease in the rural population of Thiruvananthapuram district, Kerala, India. Int J Cardiol 1993:39:59-70.

Lara-Pantin E. Obesity in developing countries. In: Berry EM, Blondheim SH, Eliahou HE, Shafrir E, eds. Recent advances in obesity research. Proceedings of the Fifth International Congress on Obesity. Paris and London: John Libbey, 1987:5-8.

Lester FT. Blood pressure levels in Ethiopian outpatients. Ethiop Med J 1973:11:145-54.

Little MA, Galvin K, Mugambi M. Cross-sectional growth of nomadic Turkana pastoralists. Hum Biol 1983;55: 811-30.

Little MA, Johnson BR. Grip strength, muscle fatigue, and body composition in nomadic Turkana pastoralists. Am J Phys Anthropol 1986;69:335-44.

Little MA, Leslie PA, Campbell KL. Energy reserves and parity of nomadic and settled Turkana women. Am J Hum Biol 1992;4:729-38.

Maddocks I, Rovin L. A new Guinea population in which blood pressure appears to fall as age advances. Papua New Guinea Med J 1965;8:17-21.

Malik P, Singh R. Familial correlations and heritability of body measurements in high and low socio-economic groups of an Indian population. Indian J Phys Anthropol Hum Genet 1991;17:93-104.

Malik SL, Singh IP. Growth trends among male bods of Ladakh - a high altitude population. Am J Phys Anthropol 1978;48:171-6.

Malville NJ. Hemoglobin concentration and working capacity of nonpregnant rural women of the Kathmandu valley of Nepal. Am J Hum Biol 1991:3:377-87.

Manocha S, Gupta MC, Gandhi BM. Difference in dietary intake of obese females belonging to different income groups. Indian J Nutr Diet 1985:22:52-6.

Markovic N, Olomu IN, Bunker CH, Huston SL, Ukoli FAM, Kuller LH. Adequacy of a single visit for classification of hypertensive status in a Nigerian civil servant population. Int J Epidemiol 1994:23:723-9.

Marks C, Mahalu W, Kuskov S. Management of cardiovascular disorder in Zimbabwe: 1989-1992. Cent Afr J Med 1994:40:1-4.

Martin FIR, Wyatt GB, Griew AR, Haurahelia M, Higginbotham L. Diabetes mellitus in urban and rural communities in Papua New Guinea. Diabetologia 1980;18:369-74.

Matawaran AJ, Gervasio CC. Average heights and weights of Filipinos. Philippine J Nutr 1971,24:74-92.

Mazess RB. The oxygen cost of breathing in man: effects of altitude, training and race. Am J Phys Anthropol 1968:29:365-76.

McGarvey ST. Obesity in Samoans and a perspective on its etiology in Polynesians. Am J Clin Nutr 1991;53:1586S-94S.

McGarvey ST, Levinson PD, Rausserman L, Galanis DJ, Hornick CA. Population change in adult obesity and blood lipids in American Samoa from 1976-1978 to 1990. Am J Hum Biol 1993;5:17-30.

McKay DA, Wade TL. Nutrition, environment and health in the Iban longhouse. Southeast Asian J Trop Med Public Health 1970;1:68-77.

Miller DS, Baker J, Bowden M, Evans E, Holt J, McKeag RJ, Meinertzhagen L, Mumford PM, Oddy DJ, Rivers JPWR, Sevenhuysen G, Stock MJ, Watts M, Kebede A, Wolde-Gabriel Y, Wolde-Gabriel Z. The Ethiopia applied nutrition project. Proc R Soc Lond 1976;194:23-48.

Miller JE, Huss-Ashmore R. Do reproductive patterns affect maternal nutritional status? An analysis of maternal depletion in Lesotho. Am J Hum Biol 1989;1:409-19.

Milton K. Morphometric features as tribal predictors in North-Western Amazonia. Ann Hum Biol 1983;10:435-40.

Minghelli G, Schutz Y, Charbonnier A, Whitehead R, Jequier E. Twenty-four-hour energy expenditure and basal metabolic rate measured in a whole-body indirect calorimeter in Gambian men. Am J Clin Nutr 1990;51:563-70.

Mock NB, Abdoh AA, Magnani RJ, Konde MK. The relationship between maternal and child nutritional status in rural Guinea. Ecol Food Nutr 1993;30:39-49.

Murray CJL, Feaschem RGA. Adult mortality in developing countries. Trans R Soc Trop Med Hyg 1990;84:1-2.

Musaiger AO. Nutritional status and dietary habits of adolescent girls in Oman. Ecol Food Nutr 1994;31:227-37.

Mustaffa BE. Diabetes mellitus in peninsular Malaysia: ethnic differences in prevalence and complications. Ann Acad Med 1985;14:272-6.

Nag PK. Maximal oxygen uptake of agricultural men and women in India. Am J Phys Anthropol 1987;74:149-53.

Nair P, Nyamphisi M, Yarnell JWG. Lack of difference in blood pressure between the urban and rural population in Lesotho, Africa. Cent Afr J Med 1994;40(10):278-81.

Norgan NG, Ferro-Luzzi A, Durnin JVGA. The energy and nutrient intake and the energy expenditure of 204 New Guinean adults. Phil Trans R Soc Lond 1974,268:309-48.

Norgan NG, Ferro-Luzzi A, Durnin JVGA. The body composition of New Guinea adults in contrasting environments. Ann Hum Biol 1982;9(4):343-53.

Nurse GT. The body weight of African village men. Cent Afr J Med 1968;14(5):94-6.

Osaka T, Konomi S, Ogaki T, Yoshimizu Y, Kawasaki T, Uezono K, Ito K, Chijiiwa C. A comparative study of percent fat and living style on Nepalese. Ann Phys Anthropol 1990:9:265-73.

Padmavati S. Epidemiology of cardiovascular disease in India. Circulation 1962;25:711-17.

Pakpeankitvatana R, Tanphaichitr V, Lochaya S,Viruenchavee N, Tamwiwat C, WatthanaKasetr S. Clinical and epidemiological studies on hyperlipidemia in urban Thais. In: Yasumoto K, Hokawa Y, Koishi H, Sanno Y, eds. Proceedings of the Fifth Asian Congress of Nutrition. Tokyo: Center for Academic Publications, 1987:430-3.

Panter-Brick C, Miller DG, Eggerman M. Hemoglobin levels and step test performance of men and women in Nepal. Am J Hum Biol 1992;4:481-91.

Papoz L, Khalifa FB, Eschwege E, Ayed HB. Diabetes mellitus in Tunisia: description in urban and rural populations. Int J Epidemiol 1988;17:419-22.

Pearson JD, Hanna JM. Selective migration as a bias in modernization studies: premigration differences in morphology and blood pressure among 15-30 year-old American Samoans. Am J Hum Biol 1989;1:263-9.

Pellett PL. The U.S. surgeon general’s report and the third world. Ecol Food Nutr 1989;23:237-42.

Podar T, Tuomilehto-Wolf E, Tuomilehto J, LaPorte RE, Adojaan B. Insulin-dependent diabetes mellitus in native Estonians and immigrants to Estonia. Am J Epidemiol 1992,135:1231-6.

Prentice AM, Whitehead RG, Roberts SB, Paul AA. Long-term energy balance in childbearing Gambian women. Am J Clin Nutr 1981,34:2790-9.

Prior IAM, Harvey HPB. The health of two groups of Cook Island Maoris. CNZ Department of Health, Special Report Series No. 261. Wellington, New Zealand: Government Printer, 1966.

Prior I. Epidemiology of cardiovascular diseases in Asian-Pacific region. Cardiovascular epidemiology in Polynesians in the pacific. Singapore Med J 1973;14:223-7.

Pryer JA. Body mass index and work-disabling morbidity: results from a Bangladesh case study. Eur J Clin Nutr 1993;47:653-7.

Qamra SR, Deodhar SD, Jit I. A metric study of foot of North-West Indians and its relationship to body height and weight. Indian J Phys Anthropol Hum Genet 1986;12:131-8.

Raja C, Singh R, Bharadwaj H. Anthropometric determination of body volume, body density and segmental volume in adult Indian women. Ann Hum Biol 1977;4:565-75.

Rao DH, Sastry JG. Growth pattern of well-to-do Indian adolescents and young adults. Indian J Med Res 1977;66:950-6.

Rao KV, Rau P, Thimmayamma BVS. Nutritional anthropometry of Indian adults. Indian J Nutr Diet 1986,23:239-56.

Riley AP. Dynamic and static measures of growth among pre-postmenarcheal females in rural Bangladesh. Am J Hum Biol 1990;2:255-64.

Rosetta L. Sex differences in seasonal variations of the nutritional status of Serere adults in Senegal. Ecol Food Nutr 1986;18:231-44.

Rothhammer F, Spielman RS. Anthropometric variation in the Aymara: genetic, geographic, and topographic contributions. Am J Hum Genet 1972;24:371-80.

Sanjeev, Indech GD, Jit I, Johnston FE. Skinfold thicknesses, body circumferences and their relationship to age, sex and socioeconomic status in adults from Northwest India. Am J Hum Biol 1991;3:469-77.

Sarkar JM, Das NK, Bhattacharya KK, Ghosh AK. A study of stature and weight in relation to some haemoglobin genotypes. Indian J Phys Anthropol Hum Genet 1984;10:175-8.

Satwanti, Singh IP, Bharadwaj H. Fat distribution in lean and obese young Indian women: a densitometric and anthropometric evaluation. Am J Phys Anthropol 1980;53:611-6.

Schultink WJ, Klaver W, Wijk HV, van Raaij JMA, Hautvast JGAJ. Body weight changes and basal metabolic rates of rural Beninese women during seasons with different energy intakes. Eur J Clin Nutr 1990;44:31-40.

Schultink WJ, Lawrence M, van Raaij JMA, Scott WM, Hautvast JGAJ. Body composition of rural Beninese women in different seasons assessed by skinfold thickness and bio-electrical impedance measurements and by a deuterium oxide dilution technique. Am J Clin Nutr 1992;55:321-5.

Schultink WJ, van Raaij JMA, Hautvast JGAJ. Seasonal weight loss and metabolic adaption in rural Beninese women: the relationship with body mass index. Br J Nutr 1993;70:689-700.

Seftel HC, Johnson S, Muller EA. Distribution and biosocial correlations of blood pressure levels in Johannesburg blacks. S Afr Med J 1980:313-20.

Sever PS, Gordon D, Peart WS, Beighton P. Blood-pressure and its correlates in urban and tribal Africa. Lancet 1980;2(8185):60-4.

Shack KW, Grivetti LE, Dewey KG. Effects of resettlement on nutritional status of mothers and children in lowland Papua New Guinea. Ecol Pood Nutr 1990;24:37-54.

Sharma K, Talwar I, Sharma N. Age at menarche in relation to adult body size and physique. Ann Hum Biol 1988;15:431-4.

Sichieri R, Coitinho DC, Leao MM, Recine E, Everhart JE. High temporal, geographic, and income variation in body mass index among adults in Brazil. Am J Public Health 1994;84:793-8.

Singh R. Nutritional anthropometric measurements of male Tamil clerks in Madras (sea coast plains) and Ooty (hills). Ann Hum Biol 1975;2:301-4.

Sinha DP, Mclntosh CE. Changing nutritional patterns in the Caribbean and their implications for health. Food Nutr Bull 1992;14:88-96.

Sitthi-Amorn C, Chandraprasert S, Bunnag SC, Plengvidhya CS. The prevalence and risk factors of hypertension in Dlong Toey slum and Klong Toey government apartment houses. Int J Epidemiol 1989;18:89-94.

Sloan AW, Masali M. Anthropometry of Sherpa men, Ann Hum Biol 1978;5:453-8.

Sood AK, Kapil U, Gupta MC. Epidemiology of obesity in an urban community. Indian J Nutr Diet 1985;22:42-8.

Spurr GB, Reina JC, Dufour DL, Narvaez JV. VO2 max and nutritional status in urban Colombian girls and women. Am J Hum Biol 1994;6:641-9.

Spurr GB, Reina JC, Li SJ, de Orozco B, Dufour DL. Body composition of Colombian women. Am J Clin Nutr 1994;60:279-85.

Strickland SS, Ulijaszek SJ. Resting energy expenditure and body composition in rural Sarawaki adults. Am J Hum Biol 1993;5:341-50.

Strickland SS, Ulijaszek SJ. Iban energy nutrition and shifting agriculture. Ecol Food Nutr 1994,33:75-92.

Sukkar MY. Skinfold thickness and body fat in adult for men and women of Western Sudan. Hum Biol 1976;48:315-21.

Swai AB, McLarty DG, Kitange HM, Kilima PM, Tatalla S, Keen N, Chuwa LM, Albert KGMM. Low prevalence of risk factors for coronary heart disease in rural Tanzania. Int J Epidemiol 1993;22:651-9.

Tai TY, Yang CL, Chang CJ, Chang SM, Chen YH, Lin BJ, Ko LS, Chen MS, Chen CJ. Epidemiology of diabetes mel-litus among adults in Taiwan, ROC. J Med Assoc Thai 1987;70:42-8.

Tai TY, Yang CL, Chang CJ, Chang SM, Chen YH, Lin BJ, Ko LS, Chen MS, Chen CJ. Epidemiology of diabetes mel-litus in Taiwan, ROC - comparison between urban and rural areas. J Med Assoc Thai 1987;70:49-53.

Tai TY, Chuang LM, Wu HP, Chen CJ. Association of body build with non-insulin dependent diabetes mellitus and hypertension among Chinese adults: a 4-year follow-up study. Int J Epidemiol 1992;21:511-7.

Tan RH, DeGuzman PE, Lingao AL, Del Rosario TB, Cuaderno FC. Weight, height and arm skinfold measurements of high school students in Metropolitan Manila. Philippine J Nutr 1974:27:27-33.

Taylor R, Pam P, Zimmet P, Raper LR, Ringrose H. Physical activity and prevalence of diabetes in Melanesian and Indian men in Fiji. Diabetologia 1984;27:578-82.

Taylor R, Zimmet P, Levy S, Collins V. Group comparisons of blood pressure and indices of obesity and salt intake in Pacific populations. Med J Aust 1985;142:499-501.

Taylor R, Badcock J, King H, Pargeter K, Zimmet P, Fred T, Lund M, Ringrose H, Bach F, Wang R, Sladden T. Dietary intake, exercise, obesity and noncommunicable disease in rural and urban populations of three Pacific island countries. J Am Coil Nutr 1992;11:283-93.

Tracer DP. Fertility-related changes in maternal body composition among the Au of Papua New Guinea. Am J Phys Anthropol 1991;85:393-405.

Ulijaszek SJ, Lourie JA, Taufa T, Pumuye A. The Ok Tedi health and nutrition project, Papua New Guinea: adult physique of three populations in the North Fly region. Ann Hum Biol 1989;16:61-74.

Underwood BA, Hepner R, Cross E, Mirza AB, Hayat K, Kallue A. Height, weight, and skinfold thickness data collected during a survey of rural and urban populations of West Pakistan. Am J Clin Nutr 1967;20:694-701.

Valencia ME, Moya SY, McNeill G, Haggarty P. Basal metabolic rate and body fatness of adult men in northern Mexico. Eur J Clin Nutr 1994;48:205-11.

Verma BL, Kumar A, Srivastava RN. Measurement of body-build based on weight and height an index for adults in an Indian population. Indian J Public Health 1982;26:133-43.

Weigel MM, Armijos RX, Monaco MH, Izurieta R, Racines RJ, Zurita C, Jaramillo G. Nutritional and health status of rural women colonists in the subtropical lowlands of Northwest Ecuador. Ecol Food Nutr 1992:29:25-43.

WHO Monica Project: A Worldwide Monitoring System for Cardiovascular Disease. World Health Statistics Annual. Geneva: World Health Organization, 1989:27-59.

Wiley AD. Neonatal and maternal anthropometric characteristics in a high altitude population of the western Himalaya. Am J Hum Biol 1994;6:499-510.

Winkler EM, Kirchengast S. Body dimensions and differential fertility in Kung San males from Namibia. Am J Hum Biol 1994;6:203-13.

Yassin Z, Terry RD. Anthropometric characteristics of rural elderly females in Malaysia. Ecol Food Nutr 1991;26:109-17.

Yonga GO, Ogola EN, Juma FD. Cardiovascular risk factor profiles in mild to moderate hypertensives seen at Kenyatta National Hospital. E Afr Med J 1993;70:693-5.

Zimmet P, Faaiuso S, Ainuu J, Whitehouse S, Milne B, DeBoer W. The prevalence of diabetes in the rural and urban Polynesian population of Western Samoa. Diabetes 1981;30:45-51.

Zimmet PZ, King HOM, Bjorntorp SPA. Obesity, hypertension, carbohydrate disorders and the risk of chronic diseases: Is there any epidemiological evidence for integrated prevention programmes? Med J Aust 1986;145:256-62.

Zinyowera TS, Msamati BC. Obesity: Is it a problem in black Africans in Zimbabwe? Cent Afr J Med 1994:40:33-8.