Contents - Previous - Next


The choice of cut-off point for distinguishing normal body weights from underweight or 'chronic energy deficiency' in adults


Introduction
Optimum population BMIs in men
Optimum BMIs in women
Age-related changes in BMI and its implications for the CED classification
Conclusions
References



W. P. T. James1 and P. J. François2

1Rowett Research Institute, Greenburn Road, Bucksburn, Aberdeen AB2 9SB, UK; 2Fomerly of FAO Nutrition Division; current address: 12 Square pont de Sevres, Paris 92100, France

Correspondence to: Professor W. P. T. James.

Introduction


When IDECG commissioned three of us (James, Ferro-Luzzi & Waterlow, 1988) to develop criteria for specifying adult 'chronic energy deficiency' (CED) we were faced with the problem of how best to choose a population with a defined level of health so that they could act as a reference group. Although US data had been used in developing the reference NCHS values for normal child growth, we were conscious of the remarkably high rates of obesity in the adult US population. To take NHANES data on adult weights would therefore distort the picture, particularly if we were to define a lower cut-off point on the basis, for example, of the mean -2 SD, a limit which is frequently chosen in nutritional and anthropometric studies. This led us to use the data from British army recruits who had been not only screened medically for disability and for substantial obesity but also had undergone physical training so that they could properly be considered both adequately nourished and healthy. We recognized that the men tended to increase their weight from the age of 18 to 23 or 24 years (James, 1976) by about 2 BMI (body mass index: kg/m2) units and this was construed as reflecting the final stages of muscular development in individuals who were physically fit throughout their army career. Nevertheless, the choice of 18.5 was pragmatic and did not reflect the third percentile. In the army women the BMI range was less, but again we originally decided to apply the same cut-off point to women as to men because we were aware that both sexes increased their morbidity when their weights were below a BMI of 17.0 and that women performed arduous work in Third World countries and therefore needed energy reserves. We were also aware of François' data on the better work performance in women who had a BMI 18.7 (Shetty & James, 1994).

Optimum population BMIs in men


The problem of how best to choose an appropriate cut-off point appropriate to Third World studies is illustrated in Fig. 1 which shows the cumulative distribution curves of BMIs for four population groups. Included in the display is the third percentile line. If we take the third percentile as the appropriate cut-off point, then only the Tunisian data suggest that the choice of 18.5 as the lower limit of a healthy weight is reasonable for men. The median BMI of this male Tunisian population is 23.1 but, as is also evident on this graph, about 22.5% of them are overweight with a BMI >25.0. Extending the analysis of the third percentile to a wider range of populations proved possible because one of us (P.F.) had accumulated representative national data on the anthropometry of adults over several years. Figure 2 shows the relationship in men between the median BMI of different populations and the third percentile BMI value. The scatter of the data is surprisingly consistent and suggests that a median BMI of 23-24 is needed if one chooses to have a third percentile of 18.5. The issue is therefore how best to choose either a suitable lower cutoff point and then an optimum population BMI or vice versa.

Fig. 1. Fitted cumulative distribution curves of body mass index (BMI) in four countries, for adult males. Median BMIs: India (O) 18.6; Rural Togo () 20.2; Brazil (m) 21.9; Tunisia (•) 23.1.

We have, of course, no direct data on the health experience of these different populations to justify a particular choice of cut-off point, but individual studies had clearly demonstrated that morbidity increased once the BMIs fell below 17.0 in Bangladeshi men (James et al., 1988). Figure 2 also implies that the minimum median male BMI value, corresponding to a third percentile of 17.0, would be about 21.0. In practice we chose 18.5 as a cut-off point for three reasons. First, a greater body weight, e.g. when the BMI is >18.5, enhances work capacity (Shetty & James, 1994). Secondly, there are very few healthy adult men in the 17.0-18.5 bracket in affluent societies and, thirdly, there is some concern that the J-shaped curve of Western morbidity/mortality studies in relation to BMI does imply higher death rates in thin adults. Although much of this enhanced mortality relates both to the enhanced risk of smokers who tend to be thin and to pre-existing disease, we cannot be sure that thinness is not a health problem. Thus, the combination of factors favoured the higher cut-off point of 18.5 rather than a more rigorous choice of 17.0 as a cut-off point. This then implies a median BMI for a male population of about 23.0.

Rose (1991) emphasized the close link between the median BMI of adult communities and the proportion of a population which is overweight by re-analysing the BMI distribution curves of the population groups involved in his INTERSALT study. As widely recognized, as the distribution of BMI shifts to the right so the skewness of the data increases markedly. Rose then displayed data on 53 population groups where the mean BMI was related to the proportion of a population with a BMI >30.0. There is a clear cut-off point of 23.0 for the mean BMI where the probability of an increasing prevalence of obesity (BMI >30.0) increases markedly. These data, together with Figs 1 and 2, imply that both the maximum and the minimum value for the average BMI of a male population should be (23.0 if both obesity (BMI >30) and underweight (BMI <18.5) are to be avoided.

Rose's data were based on adults aged 20-59 years of age, so this is very comparable with the present François data on adults shown in Fig. 2. Rose's data included some Third World communities, but the developed countries dominated his INTERSALT study. The choice of a median BMI of 23.0 for the weights of all adults to eliminate the prevalence of obesity (i.e. BMI 230) may not apply at all ages and the choice could vary depending on environmental factors, e.g. the demand for work and the possibility of famine. All of these factors differ markedly between First and Third World countries.

Fig. 2. Scatter diagram relating median body mass index (BMI) and BMI values at percentile 3. Data on adult males from 35 large, territorially representative samples.

The validity of the choice of a median BMI of 23.0 depends critically on having confidence that a BMI 230.0 is hazardous to health. This issue has been discussed extensively (Royal College of Physicians, 1983; James, 1987) and seems true whether or not one removes the issue of smoking from the analysis. It is especially true if the long-term impact of being obese as a young adult is considered. Although the increment in risk as weights rise from a BMI of 25.0 to 27 or 28 is modest, the rise in risk for those with a BMI >30 is widely accepted. There is, however, the same dilemma in using BMIs of 25-30 in the obesity classification as a hazardous category as that now encountered in specifying a health hazard of having a body weight below a BMI of 18.5 rather than below 18.0 or below 17.5. If health issues are discounted and reliance is placed only on using the third percentile as a cut-off point, then the choice of a value of 18.5 is reasonable given a median BMI value of 23.0. If, however, we seek to limit the extent of overweight by minimizing the proportion of a population with a BMI 25, then the only population we have so far discovered to have only small proportions of both underweight, i.e. <18.5 BMI, and overweight (BMI >25.0) is the Chinese population (Shetty & James, 1994). In the Chinese the median BMI was 20.7 in men and 21.2 in women. The small distribution of weights in this population is very unusual and presumably reflects the equity of distribution of resources and the similarity of lifestyles throughout the Chinese community during the early 1980s. With median BMIs of 21.0, however, most societies will have a third percentile of about 17.0 (Fig. 2). Thus, a median BMI of 23.0 provides a third percentile of 18.5 whereas a median BMI of 21.0 minimizes the overweight. Perhaps with the modest effects of both underweight (BMI 17.0-18.5) and of overweight (BMI 25-30) we should designate pragmatically a median population BMI of 21-23 as acceptable in men.

Optimum BMIs in women


As frequently noted in epidemiological studies on health in relation to BMI, most of the insurance data and the prospective data on health and chronic disease relate to men and not women. The Metropolitan Life Insurance analyses did, however, specify the optimum range for women and these were recalculated by one of us (W.P.T.J.) to provide BMI indices based on nude weights and heights corresponding to a BMI of 18.7-23.8 in women and to 20.1-25 in men (Royal College of Physicians, 1983). These figures were then rounded up by Garrow & Webster (1985) to give the universally accepted simplified range of 20-25 for both sexes as optimum in Western societies. Since then Willett, in prospective studies on US nurses, has found that women have the lowest risk of ill health and coronary artery disease at rather low BMIs, i.e. in the 19-21 range (Manson et al., 1990). Therefore, the optimum BMI of women may in fact be somewhat less than that of men, but the choice may depend on whether we are considering First or Third World conditions.

Figure 3 suggests that in practice, in the Third World, women's weights are greater than men's but that the distribution is also greater with more women than men having a BMI <18.5 at equivalent median BMIs for the population. This is very consistent in data from the selected 27 regions of the world. This does not, of course, prove that women benefit from having a different range of BMIs. Good data are needed on the health of men and women at different BMIs in Third World as well as First World countries. François has shown that below a BMI of 17.0 African women have a greater likelihood of illness and are more frequently confined to bed. Their ability or willingness to engage in heavy work is also greater if they have a BMI of 18.7 or more (Shetty & James, 1994).

Women's work capacity is less than that of men at low BMIs even allowing for their smaller muscle mass. This simply reflects the fact that women tend to be shorter than men (Berio, François & Perisse, 1985) and therefore at equivalent BMIs the women will be of a lower body weight. Work requiring the leverage of the body's weight as a power function gives the heavier women (or men) an advantage. Similarly, heavier women tend to have a greater maximum oxygen uptake (VO2 max) than thin women. The ability to maintain high work outputs is directly related to the proportion of VO2 max being demanded. Thus the higher VO2 max of an individual the lower is the 'stress' of a particular task and the greater the ability to sustain high energy ouputs over a prolonged period of time. Men do better at power-related work, e.g. cane-cutting, carrying weights uphill etc., even if women are capable of enduring modest activities for prolonged periods of time. The data on heavy working by African women (Shetty & James, 1994) do indeed suggest that there are functional benefits to having a body weight of a 18.7 and one could argue that women, being shorter than men, might gain some power advantage from having a higher BMI.

Fig. 3. Scatter diagram relating median body mass index (BMI) and the proportion of population <=BMI 18.4. Data from 27 large, territorially representative samples.

Figure 3 would seem to require a median BMI of 24 for women to retain the third percentile as 18.5, but this median BMI exceeds the upper individual limit for health of 23.8 proposed on the basis of Western data (Royal College of Physicians, 1983). This value also substantially exceeds that proposed by Manson et al. (1990). Thus we either reduce the optimum BMI of women and specify a lower cut-off point of, say, 17.0 below which health risks increase markedly or accept that, in Third World countries where women engage in very demanding agricultural work, the weigh-trelated capacity for work does demand a cut-off point for women equivalent to that of men. Again we see a trade-off between the potential hazards of being underweight and those of being overweight. A median BMI range of 21-23.0 seems reasonable with the Third World women gaining greater benefit from a median BMI of 23.0 whereas the affluent women may be better with a BMI of 21.0.

Age-related changes in BMI and its implications for the CED classification


We have seen in different population groups a clear increase in BMI with age: this affects affluent societies to a much greater extent than Third World communities but as those communities become affluent first the middle-aged women become heavier and then, with further increases in income, middle-aged men also increase in weight. Berio et al (1985) assessed the BMIs of men and women aged 20-24 years of age in 58 countries and when these data are redisplayed as the BMIs themselves it is evident that young men have a consistently lower BMI than women at this age. The recalculated data are displayed in Fig. 4. Clearly there is an overlap of First, Second and Third World weights in young men with BMIs of 20-22 being common. This in turn might imply (Fig. 3) that a third percentile of these young groups could be as low as 16.0 with 22% of the population being <18.5. The distribution data in Fig. 3, however, relate to the whole population rather than selectively to the young displayed in Fig. 4. Clearly we must be cautious in interpreting low BMI data in young men.

The data on young women provide a similar cautionary tale with women in affluent societies often having a median BMI of 21.0. At this BMI the distribution curves of men and women are closer to each other. This lends credence to the cut-off point of 17.0 (James et al., 1988) in this affluent group. If power/work-related functions are important between a BMI of 17.0 and 18.5 then this issue is rarely of relevance to women in Western societies but could be crucial to those in developing countries.

Fig. 1. Fitted cumulative distribution curves of body mass index (BMI) in four countries, for adult males. Median BMIs: India (O) 18.6; Rural Togo () 20.2; Brazil (m) 21.9; Tunisia (•) 23.1.



Conclusions


These systematic analyses reveal that the present BMI cut-off point of 18.5 in both men and women must be seen as the outcome of a series of compromises and it accords with the expected third percentile for a male population with a median value of 23.0 and in women of about 24.0. However, these median values imply that one accepts an appreciable proportion of overweight adults with a BMI of 25-20. We may conclude that a BMI of 23.0 and 24.0 is optimum in men and women in Third World societies because values below this lead to a higher proportion of adults who are less able to sustain high work outputs.

The distribution of BMIs in the First World and Western data implies that a median BMI between 20 and 21 is consistent with excellent long-term health in groups of adults but a population median of 20 implies that the third percentile is 16.0. This is clearly associated with marked morbidity in the developing world and the same is probably true in the West. It is therefore proposed to retain the optimum median at 21-23 with a cut-off point of 18.5 pending further studies. The avoidance of overweight may be more important in Western societies, but for the simplicity of handling data it would seem reasonable for the present to retain the BMI range of 18.5-25.0 as acceptable for individuals and a median population BMI as best within the 21-23 range.

In both men and women aged 20-24 the choice of 18.5 as a cut-off point is likely to categorize as underweight an appreciable proportion of the supposedly normal population. The significance of a BMI of 17.0-18.5 may therefore be of less importance in this age group.

References


Berio AJ, François P & Perisse J (1985): New insights into human energy requirements. Food Nutr. 11, 21-35.

Garrow JS & Webster J (1985): Ouetelet's index (W/H2) as a measure of fatness. Int. J. Obes. 9, 147-153.

James WPT (1976): DHSS/MRC report on research on obesity. London: HMSO.

James WPT (1987): Being overweight: working definition. In Body weight control: the physiology, clinical treatment and prevention of obesity, eds AE Bender & LJ Brookes, p. 90-106. London: Churchill Livingstone.

James WPT, Ferro-Luzzi A &: Waterlow JC (1988): Definition of chronic energy deficiency in adults. Eur. J. Clin. Nutr 42, 969-981.

Manson JE, Colditz GA, Stampfer MJ et al (1990): A prospective study of obesity and risk of coronary heart disease in women. N. Engl J. Med. 322, 882-889.

Rose G (1991): Population distributions of risk of disease. Nutr Metab. Cardiovasc. Dis. 1, 37-40.

Royal College of Physicians (1983): Obesity. J. R. Coll Physicians 17 (1).

Shetty PS & James WPT (1994): Body mass index - a measure of chronic energy deficiency in adults. Rome: FAO.


Contents - Previous - Next