¹Danish Epidemiology Science Centre at the Institute of Preventive Medicine, Copenhagen University Hospital, Copenhagen, Denmark

²Department of Medicine, Sahlgrenska University Hospital, Ôstra, Göteborg, Sweden

³Department of Medicine, Sahlgrenska University Hospital, Sahlgrenska, Göteborg, Sweden

Correspondence to: B L Heitmann, Unit for Dietary Studies at the Centre for Preventive Medicine, Medical Department F/C, Glostrup University Hospital, DK-2600 Glostrup, Denmark. BeHe@Glostruphosp.KBHamt.dk.

OBJECTIVE: To describe differences in the 22 y mortality risk associated with body mass index (BMI), body fat or fat-free mass, in order to examine if the differential health consequences of fat and fat-free mass may be responsible for elevated mortality rates at both high and low BMI.

DESIGN: Prospective cohort study, a 22 y follow-up.

SETTING: General community. The study of men born in 1913, Gothenburg.

SUBJECTS: 787 men aged 60 y.

MAIN OUTCOME MEASURES: Number and time of total deaths from 1973 to 1995.

RESULTS: The risk of dying was a linear function of percentage fat and fat-free mass, and increased from a relative risk of 1.00 in men belonging to the lowest fifth to 1.4 (95% confidence interval 1.11-1.99) in men in the highest fifth of percentage fat mass. For BMI the lowest risk was observed for men belonging to the middle fifth of BMI. When the relative risk was set at 1.00 for subjects belonging to the middle fifth of BMI the risk associated with the low BMI fifth was 1.3 (95% confidence interval 0.94-1.68) and that with the highest fifth was 1.5 (95% confidence interval 1.09-1.96). Analyses including both body fat and fat-free mass showed that total mortality was a linear increasing function of high fat and low fat-free mass.

CONCLUSION: The apparent U-shaped association between BMI and total mortality may be the result of compound risk functions from body fat and fat-free mass.

International Journal of Obesity (2000)24, 33-37

total mortality; fat-free mass; body fat; body mass index; waist circumference

Introduction

Obesity is defined as an abnormally high proportion of fat.¹ However, traditionally, overweight estimated from relative weight indices, such as the body mass index (BMI), is used to describe obesity. Simple measures like skinfolds have been used to assess the body fat morbidity association,^2,3 but only a few studies have measured body fat by advanced methodology, and these studies have not included a sufficient number of participants to attribute mortality to obesity. Therefore, all public health recommendations regarding obesity are based on studies relating relative weight to health.^1,4,5 A large number of studies have suggested that the association between BMI and mortality is U-shaped,⁶ indicating that not only is a high BMI associated with increased total mortality, but that even a low BMI seems to carry an increased risk of premature death, a relation not easy to understand. A number of studies have suggested that the increased risk of total mortality related to low BMI, can, in part, be attributed to confounding by smoking or pre-existing disease.^7,8 However, other studies have found excess mortality related to low BMI even after control for these factors.^6,9,10 In this regard, it has been speculated that differential health consequences of fat and fat-free mass can be masked by the use of BMI when studied in relation to mortality.^11,12

The present study describes differences in the 22 y mortality risk associated with BMI, body fat or fat-free mass, with the purpose of examining whether the U-shaped relationship found between BMI and premature death may depend on a mixed risk function related to body composition, for example, whether differential health consequences of fat and fat-free mass may be responsible for elevated mortality rates at both high and low BMI.

Methods and subjects

A sample, including all men born in 1913 on a date divisible by three (for example born 3rd, 6th, 9th,¼, 30th in a given month) and living in the city of Gothenburg in November 1972, was selected. All residents of Sweden have a registration number that includes their date of birth and other vital statistics. Names, addresses and identification numbers are registered by the County Census Bureau in a population register that, by law, must be kept up to date. In total, 945 men met these criteria, and 83.3% (787) agreed to participate in a health examination. Non-participation has been described earlier.¹³

Body composition

Total body potassium was measured in a whole body counter,¹⁴ and lean body mass calculated using the assumption of a potassium content of 68.1 mEq/kg lean body mass. Fat mass was calculated by subtracting the lean body mass from body weight.

Anthropometric measures

Body weight was measured to nearest 0.1 kg with subjects dressed in underpants only, using a lever balance. Height was measured to the nearest 1 cm, subjects being without shoes. Waist circumference was measured at the level of the umbilicus, subjects standing, using a tape measure to the nearest 1 cm.

Questionnaire data

Leisure time activity was recorded on a four-point scale ranging from (1) almost completely inactive (reading, TV-viewing, cinema); (2) some physical activity (at least 4 h per week); (3) regular activity; and (4) regular hard physical training for competition.¹⁵ Subjects recorded their present and previous smoking habits, and were classified as current, former or non-smokers.

Follow-up

The vital statistics of subjects were followed until 31 December 1995, using their individual identity number. All 787 men were observed from the initial examination in 1973 until death or 31 December 1995. The follow-up rate for mortality was 100%.

Statistical analyses

Data were analysed, using Cox regression models. Measures of BMI, body fat or fat-free mass were recoded in fifths. A first series of models included only the risk factor. In a second series of models, the covariate smoking habit (dummy coded as non-smokers, ex-smokers and current smokers) was included, and in a third series the covariate level of physical activity was included. Due to an insufficient number of active subjects (activity levels 3-4) in the present analysis, subjects were classified as either sedentary (activity level 1) or active (activity levels 2-4). A fourth series added first-order interactions between these variables. It was tested whether a continuous modelling of these variables was not significantly different from the categorical representation by comparing the likelihoods of the two nested models for each variable.

Finally, as the division of a continuous distribution into ordered categories may result in loss of power or flawed results, data were returned to a continuous form for a final analysis using fractional polynomial regression.¹⁶ All analyses were performed using the Stata Statistical Package.¹⁷

Results

Out of a total of 787 men, 52 were excluded due to incomplete information on total body potassium. Hence, the present analysis included 735 men with complete data. During the 22 y follow-up, a total of 460 men died. On average, this group of 60 y old men were slightly overweight (BMI=25.5 kg/m²) and had a mean percentage body fat of 30.3%. In total, 44% (n=321) were current, and 35% (n=257) ex-smokers, and 76% (n=561) were sedentary. Table 1 gives the characteristics of the men by fifths of BMI.

Cox-regression analyses

All risk functions were essentially similar before and after adjustment for baseline smoking and physical activity. In addition, the inclusion of product terms suggested that there were no interactions between either smoking or physical activity, or any of the obesity mortality risk functions, or between body mass index, body fat or fat-free mass. In the following, therefore, only analyses adjusted for main effects of smoking and physical activity are presented.

Waist circumference and total mortality

Independent of whether BMI was included in the model, or whether waist was included as a linear or quadratic function, waist circumference was not statistically associated with total mortality. The highest risk was observed for men in the highest fifth of waist circumference with RR=1.2 (95% confidence interval 0.90-1.60), compared with men belonging to the lowest fifth of waist circumference.

Percentage body fat and total mortality

The highest risk was observed for men in the highest fifth of percentage body fat, with a significantly increased RR=1.5 (95% confidence interval 1.11-2.00), compared with men belonging to the lowest fifth of percentage body fat. The risk function of percentage body fat mass and total mortality was compatible with linearity (chi-square test=3.6, P=0.31).

BMI and total mortality

The lowest risk was observed for men belonging to the middle fifth of BMI. The highest risk was observed for men in the highest fifth of BMI, and this risk was not significantly different from that associated with the lowest fifth of BMI (P=0.29), in that there was a U-shaped association between BMI and mortality. The U-shaped association was confirmed in an additional analysis in which the term BMI² was included, and found to be significant (P=0.02). When the relative risk was set at 1.00 for subjects belonging to the middle fifth of BMI, the risk associated with the lowest BMI fifth was 1.3 (95% confidence interval 0.94-1.68) and that associated with the highest fifth was 1.5 (95% confidence interval 1.09-1.96). The U-shaped association persisted when deaths occurring during the first 5 y (57 deaths) or 10 y (140 deaths) of follow-up were excluded (data not shown).

Body composition, BMI and total mortality

A comparison of the total mortality risk associated with percentage body fat and BMI, using the middle fifth as reference, is given in Figure 1. In a fifth set of models we included both fat-free mass and body fat to predict total mortality. From these analyses there was no evidence of a lower limit for body fat below which total mortality risk is increased (linearity test: chi-square=3.2, P=0.36). For fat-free mass, on the other hand, men belonging to the lowest fifth generally had a higher total mortality risk than men from higher fifths, in whom mortality risk did not differ (linearity test: chi-square=4.5, P=0.21). Figure 2 shows the association between mortality and BMI, body fat and fat-free mass. Expressing body fat and fat-free mass as fractions of BMI (in that as functions of height-squared), as proposed by VanItallie et al,¹⁸ gave similar results and suggested a lower absolute limit, corresponding to the upper range for height-adjusted fat-free mass in the lowest fifth, of 16 kg/m². Fractional polynomial regression analysis, including either body fat or fat-free mass, confirmed the categorical analysis (Figure 3).

Discussion

The present study shows that the apparent U-shaped association between BMI and total mortality may be a result of compound risk functions from body fat and fat-free mass. While this, and several other studies, have found that both a high and a low BMI is associated with increased mortality,⁶ the present study demonstrates that percentage fat and fat-free mass are a positive and a negative linear function of mortality, respectively. This has been suggested earlier by Allison et al,¹² who, on a theoretical basis, argued that the U-shaped function is made up by an ascending and a descending function. The present study demonstrates, based on data on fat and fat-free mass estimated from total body potassium counting, that this may be the case.

Although it is apparent that misclassification with respect to obesity occurs when BMI is used as a proxy variable for body fat, the present study is the first to demonstrate the different mortality risk functions described by overweight, using BMI, and by body composition, using fat or fat-free mass. Furthermore, we have shown that the use of body composition measures offers a more plausible explanation regarding the association between adiposity and mortality than that found for BMI. It is well recognized that, at the high end of BMI, muscular subjects may well be misclassified as overweight when, in fact, they are lean. However, more surprisingly, less than half of the subjects with a low BMI (66/147) were actually lean in terms of body fat. This is in agreement with other studies finding that the sensitivity of classifying a high body fat mass from a high BMI may be as low as 20-50%.¹¹ A low sensitivity has been reported particularly among older subjects. Therefore, it cannot be excluded that the noted discrepancies in the mortality risks associated with BMI and with the measures of body composition may be less in a younger population group. Also, differences dependent on gender could not be examined in the present study, where only men had measurements of total body potassium.

The present study has also demonstrated that body fat and fat-free mass, but not BMI, may be used to rank individuals with respect to total mortality risk. In this regard, a high percentage body fat was significantly associated with a 40% increase in total mortality, compared with a low percentage body fat. The corresponding figure for BMI was 20%, and was not significant. Similar findings have been reported for Gothenburg women,³ where the 12 y incidence of diabetes was almost three times higher when comparing the high and low fifths of BMI. Also, in a clinical experiment, Segal et al¹⁹ demonstrated that adverse risk factor levels (for instance a high blood pressure, cholesterol level or blood glucose) were associated with a high body fat mass rather than with a high BMI.

In a recent report from the American Institute of Nutrition Committee (AIN) on Healthy Weight,²⁰ it was concluded that the lowest risk associated with body mass index was in the range from 18-23 kg/m², and noted that 'many adults with a BMI less than 18 may be lean, fit and healthy¼[although] with minimal reserves in the event of unrelated illness and, particularly with a BMI less than 16, may be an indication of an eating disorder or underlying illness'. From the present analyses, where both body fat and fat-free mass were expressed as functions of body weight, we found that total mortality was a linear function of body fat mass, suggesting that there is no lower critical fat mass below which total mortality is increased. On the other hand, for fat-free mass, it seems necessary that a certain minimum mass is present. For these 60 y old Swedish men that mass was critical when their fat-free mass/height², and hence their BMI, was less than 16.0 kg/m², a figure in accordance with that given by AIN.

The present study underscores the importance of understanding the nature of how body fatness varies with BMI, in order to fully understand the health risks associated with obesity. Although, in observational studies, BMI is a simple and easy measure for obesity, we argue that this measure may be inappropriate for measuring body fatness. Indeed, with the greater availability of simpler methods for measuring body composition, like bioelectrical impedance, a better understanding of the risk associated with obesity may be achieved.

In the present study, we could not demonstrate that regional fat distribution, as measured by the waist circumference, was associated with total mortality, and hence that fat distribution influenced associations between mortality and any of the measures of general fatness. Although other studies have indicated that fat distribution measured by waist circumference alone is an important determinant of mortality,²¹ this has never been demonstrated in the present population sample.²² Rather, associations with mortality have been found only for the waist-to-hip ratio.²² However, at the examination in 1973, where the measurements on total body potassium were taken, only measurement of waist, and not hip circumference, was performed.

Finally, it should be noted that, compared with other methods like underwater weighing or indicator dilution (for body water), fat-free mass may be underestimated in obese or elderly by potassium counting.^23,24 However, whereas underestimation may be substantial, it would not be expected to affect the ranking of the men into fifths of fat-free mass, and consequently does not offer an explanation for the present findings of a stronger association between body fat and mortality than between BMI and mortality, or for the possibility that the BMI mortality risk function be made up of compound risk functions from body fat and fat-free mass.

Conclusion

The present study suggests that the U-shaped association between overweight and total mortality may be a result of differential health consequences for body fat and fat-free mass, and furthermore that, compared with low levels, a high fat mass is more strongly associated with mortality risk than BMI. The present findings have implications for the understanding of obesity-mortality relations, and for further public health recommendations.

Acknowledgements

This study was supported by grants from the Swedish Medical Research Council (B98-27X-0626-17), King Gustaf V and Queen Victoria's Foundation, the Gothenburg Medical Association, the Danish National Research Foundation, the Danish Health Insurance Foundation and the Danish Heart Association.

1 Seidell JC, Deurenberg P, Hautvast JGAJ. Obesity and fat distribution in relation to health¾current insights and recommendations. Wld Rev Nutr Diet 1987; 50: 57-91.

2 Yao C-H, Slattery ML, Jacobs DR, Folsom AR, Nelson ET. Anthropometric predictors of coronary heart disease and total mortality: Findings from the US railroad study. Am J Epidemiol 1991; 134: 1278-1289. MEDLINE

3 Bengtsson C, Lapidus L, Stendahl C, Waldenstrom J. Hyperuricaemia and risk of cardiovascular disease and overall death. A 12-year follow-up of participants in the population study of women in Gothenburg, Sweden. Acta Med Scand 1988; 224: 549-555. MEDLINE

4 US Department of Agriculture, US Department of Health and Human Services. Nutrition and your health: Dietary guidelines for Americans, 3rd edn. US Government Printing Office: Washington, DC, 1990.

5 US Department of Agriculture, US Department of Health and Human Services. Nutrition and your health: Dietary guidelines for Americans. Homes and Garden Bulletin no. 232). US Government Printing Office: Washington, DC, 1995.

6 Troiano RP, Frongillo EA, Sobal J, Levitsky DA. The relationship between body weight and mortality: A quantitative analysis of combined information from existing studies. Int J Obes 1996; 20: 63-75.

7 Manson JE, Willet WC, Stampfer MJ, Coldtiz GA, Hunter DJ, Hankinson SE, Hennekens CH, Speizer FE. Body weight and mortality among women. N Engl J Med 1995; 333: 677-685. MEDLINE

8 Losonczy KG, Harris TB, Cornoni-Huntley J, Simonsick EM, Wallace RB, Cook NR, Ostfeld AM, Blazer DG. Does weight loss from middle age to old age explain the inverse weight mortality relation in old age? Am J Epidemiol 1995; 141: 312-321. MEDLINE

9 Waaler HT. Height, weight and mortality¾the Norwegian experience. Acta Med Scand 1984; : (suppl) 679 56.

10 Harris TB, Ballard-Barbasch R, Madan J, Makuc DM, Feldman JJ. Overweight, weight loss, and risk of coronary heart disease in older women. The NHANES I Follow-up Study. Am J Epidemiol 1993; 137: 1318-1327. MEDLINE

11 Baumgartner RN, Heymsfield SB, Roche AF. Human body composition and the epidemiology of chronic disease. Obes Res 1995; 3: 73-95. MEDLINE

12 Allison DB, Faith MS, Heo M, Kotler DP. Hypothesis concerning the U-shaped relation between body mass index and mortality. Am J Epidemiol 1997; 146: 339-349. MEDLINE

13 Svärdsuud K, Tibblin G. A longitudinal study of blood pressure. Change of blood pressure during ten years in relation to individual values. The study of men born in 1913. J Chron Dis 1980; 33: 627-633.

14 Forbes GB, Gallup J, Hurch JB. Estimation of total body fat from potassium-40 content. Science 1961; 133: 101.

15 Grimby G, Wilhelmsen L, Björntorp P, Saltin B, Tibblin G. Habitual physical activity. In Saltin B (ed.). Muscle metabolism during exercise. Plenum Press: New York., 1971, 469-481.

16 Royston P, Altman DG. Regression using fractional polynomials of continuous covariates: Parsimonious parametric modelling. Appl Statist 1994; 43: 429-467.

17 StataCorp. Stata Statistical Software, Release 4.0, 1995.

18 VanItallie TB, Yang M-U, Heymsfield SB, Funk RC, Boileau RA. Height-normalized indices of the body's fat-free mass and fat mass: Potentially useful indicators of nutritional status. Am J Clin Nutr 1990; 52: 953-959. MEDLINE

19 Segal KR, Dunaif A, Gutin B, Albu J, Nyman A, Pi-Sunyer FX. Body composition, not body weight, is related to cardiovascular disease risk factors and sex hormone levels in men. J Clin Invest 1987; 80: 1050-1055. MEDLINE

20 AIN Healthy Weight Steering Committee. Report of the American Institute of Nutrition (AIN) Steering Committee on Healthy Weight. J Nutr 1994; 124: 2240-2243. MEDLINE

21 Lean ME, Han TS, Seidell JC. Impairment of health and quality of life in people with large waist circumference. Lancet 1998; 351: 853-856. Article MEDLINE

22 Larsson B, Svardsudd K, Welin L, Wilhelmsen L, Björntorp P, Tibblin G. Abdominal adipose tissue distribution, obesity, and risk of cardiovascular disease and death: 13 year follow up of participants in the study of men born in 1913. Br Med J 1984; 288: 1401-1404.

23 Sjöström L, Kvist H, Cederblad Å, Tylén U. Determination of total adipose tissue and body fat in women by computed tomography, ⁴⁰K and tritium. Am J Physiol 1986; 250: E736-E745. MEDLINE

24 Heitmann BL. Methods for estimating body fat and fat free mass. In Romsos DR et al (eds). Obesity: Dietary factor and control. Japan Science Society Press: Tokyo/Karger: Basel, 1991, 227-236.

Figure 1 Relative risks of total mortality associated with either BMI or percentage body fat mass, using the middle fifth as reference.

Figure 2 Relative risks of total mortality associated with either body fat or fat-free mass.

Figure 3 Fitted relative risks: functional polynomial curves for body fat and fat-free mass.

Table 1 Characteristics of 60-y-old Swedish men by fifths of body mass index (n=147 in each fifth; results given as mean±s.d.)