Contents - Previous - Next

Assessing the linkages between low body mass index and morbidity in adults: evidence from four developing countries

Introduction
BMI trends
Morbidity experience
Correlation between low BMI and morbidity
Conclusions and summary
References
Discussion

M. Garcia and E. Kennedy

International Food Policy Research Institute, 120017th Street, NW Washington DC 20036, USA

The study explores the relationship between low BMI and illness patterns in four developing countries. Using probit analysis, the study finds small but statistically significant effect of low BMI on proneness to morbidity in Pakistan and Kenya but none in the Philippines and Ghana. The circularity of causation between low BMI and illness was addressed by using the instrumental variable approach. The study also finds that the threshold at which morbidity begins to rise is not consistent with the suggested cut-off of BMI 18.5 kg/m². The cut-offs were meaningful only in the case of Pakistan, but even here, the sizes of relationships are quite small although statistically significant.

Introduction

Adult anthropometry is frequently used for assessing the state of health and nutritional status of women. One of the specific measures that uses anthropometry is that of body mass index (BMI), which is an indicator of thinness or fatness and is defined as weight divided by the square of height (kg/m²). For many years BMI - also called the Quetelet's index - has been used in developed countries for assessing obesity and its associated risks of chronic diseases which are common in the affluent societies. In adults, obesity is a risk factor for a whole range of conditions including cardiovascular diseases, diabetes mellitus, arthritis, gallbladder diseases, cancer and respiratory dysfunction (Gurney & Gorstein, 1988); and it is no wonder that actuarial estimations include BMI in the risk calculation of insurance premiums in adults.

Recently, interest has expanded in using BMI for assessing the chronic dietary deficiencies, and morbidity patterns, at the lower end of the range of the BMI scale. Indeed for policy and diagnosis of problems, it would be important to know the acceptable lower limit at which bodily functions start to fail and subsequently cause illness and dysfunction. The various ranges of BMI suggested by James, Ferro-Luzzi & Waterlow (1988) for example suggest a cut-off of 18.5 BMI to determine the presence of chronic energy deficiency (CED) at the population level.

The purpose of this paper is to assess the links between low BMI and morbidity status using data from four developing countries. It has been shown that in children there is a problem of circularity (endogeneity) in causation (Garcia & Alderman, 1989), that low weights can cause illness, and that illness episodes can cause low weights. In order to understand the functional significance of low BMI on health, it is necessary to carefully control for this two-way interaction. The circularity of low weight-morbidity relationships is more difficult to test in adults compared with children because activity (and energy costs) in adults vary enormously. Among adults, information about the relationships between low BMI and risks to health is very limited; no systematic study on the relationship between low BMI and ill health is available.

Four data sets from household surveys carried out by IFPRI and the World Bank between 1983 and 1988 are used to examine in the Philippines, Kenya, Pakistan and Ghana the low BMI-morbidity relationships. The first three were collected as part of the consumption and nutrition studies on commercialization of crops and food subsidies, and the Ghana survey was part of the Living Standard Surveys carried out by the World Bank.

Table 1. Body mass index (BMI) categories (15-59 years old): percentages

BMI trends

Table 1 provides data on individual BMI in the four countries broken down by the BMI categories suggested by James, Ferro-Luzzi & Waterlow (1988). The data are restricted to 15-59-year-old population. Excluded from the sample are pregnant and lactating women.

Using the categories suggested, 16.5% of women in Ghana are classified as having low BMI (<18.5) and in the* terminology, women with CED. This is significantly lower than Ghanaian men (23.4%). The trend across gender in the Philippines shows a different pattern. There are more women at the low end of BMI scale compared to men (22.2% vs 25.9%).

Kenyan women, on the other hand, show the lowest prevalence of low BMI among the sample countries, at only 6.4%. In the data reviewed here, Pakistani men showed the highest proportion of low BMI at 29.1%. It should be stressed, however, that samples in Pakistan, Philippines, and Kenya were drawn from rural areas, whereas in Ghana there was a national sample.

Figures 1 and 2 show low BMI trends across income classes. It is evident that the pattern is non-linear. The income effect is more dramatic in the Pakistan sample.

The prevalence of low BMI varied across income groups. The general pattern seems to be that the proportion of men and women that are likely to fall below the 18.5 BMI cut-off declines as income increases. It is most clearly shown in the case of Pakistani males. This pattern does not appear to hold in Kenyan women who show little variation over the income range. As noted in Kennedy & Garcia (1992) the time allocation patterns in Kenyan women tend to be different from that of say, Philippine women. Kenyan women spend nearly twice as much time in home production activities (getting fuelwood, fetching water, cleaning) as women in the Philippines.

It is also interesting to note that the incidence of low BMI across income groups in Ghana is also flat, perhaps reflecting similar conditions in Kenya. The prevalence of low BMI in Ghana, however, is about twice that in Kenyan women. In contrast, low BMI prevalence in Ghanaian men falls with increasing incomes.

The impact of age and ageing on low BMI is clearly shown in all the four countries. It is also shown to be non-linear. The prevalence of low BMI is higher in the 15-30 than in the 30-35 age band but increases thereafter.

Morbidity experience

This paper uses self-reported illness experience for all of the data sets analysed for this study. There are disadvantages in using self-reported diagnosis of illness. The possibility of systematic bias has been cited in some studies in the labour supply literature (King, Rozensweig & Wang, 1991). Unlike clinical diagnosis, self-reporting of illness suffers in terms of accuracy because the perception of being ill is, in many societies, a relative question. For example the poor may not report themselves as sick although they are not feeling well; and for some others, even mild symptoms are reported as illness.

While self-reported morbidity status is only a second best solution, there are features in the data used for this analysis that can internally check the validity of the morbidity reports. In three of the four data sets, measurements were done four times during the year (six times in Pakistan). With such a series of observations, it is possible to estimate the probability of illness and the predisposition of individuals to illnesses.

In comparing the results from various countries, one should note some differences in the definition of morbidity in each of the surveys. The recall period for the Philippines and Kenya is 2 weeks and for Ghana 4 weeks. For all three surveys, the morbidity variable is defined as 'days ill' in the past period. In the Pakistan survey, morbidity is defined as 'days unable to work due to illness' in the past 8 weeks.

Table 2. Illness pattern by gender and age group

Illness patterns in the sample are shown in Table 2 and Fig. 3A, by age groups and by gender. In the Philippines, the reported illness among older males is higher than in older females. Men under 20 years, however, do report lower rates of morbidity compared with the under-20 females. The differences in illness between men and women in Ghana, however, are not significant in either young or older cohorts. The Philippine sample includes sugarcane and maize farmers in rural areas whereas the Ghanaian sample covers both urban and rural populations.

In all the four countries examined, morbidity increases with age. Men in their forties in Pakistan for example are three times as likely to be sick as men in their late teens, while Philippine men in their thirties are three times more prone to illness than men in their late teens. The data also suggest that the rise in morbidity with age is much steeper amongst men than women in all the countries, suggesting a higher likelihood of illness among men in all the sample countries.

The rise in morbidity as populations get older is in line with the observed mortality patterns in both developed and developing countries. For example, in Mexico it was found that mortality rates of males from all causes in 1986 rise from 186 per 100000 for the 25-34 age group to 490 per 100000 for the 35-44 age group (ACC-SCN, 1992). It is also noted that deaths among females in those two age groups are half those of men. Epidemiological studies also indicate that the causes of death as a population gets older shifts from deaths due to infectious diseases to deaths from non-communicable chronic diseases (Lopez, 1990).

Contents - Previous - Next