Skip Navigation LinksHome > November 1999 - Volume 31 - Issue 11 > Overweight and obesity in the mortality rate data: current e...
Medicine & Science in Sports & Exercise:
Roundtable Consensus Statement

Overweight and obesity in the mortality rate data: current evidence and research issues


Free Access
Article Outline
Collapse Box

Author Information

Department of Chronic Diseases Epidemiology, National Institute of Public Health and the Environment, 3720 BA Bilthoven, THE NETHERLANDS

Address for correspondence: Jacob C. Seidell, Ph.D., Department of Chronic Diseases Epidemiology, National Institute of Public Health and the Environment, P.O. Box 1, 3720 BA Bilthoven, The Netherlands; E-mail: J.Seidell@RIVM.NL.

Roundtable held February 4–7, 1999, Indianapolis, IN.

Collapse Box


SEIDELL, J. C., T. L. S. VISSCHER, and R. T. HOOGEVEEN. Overweight and obesity in the mortality rate data: current evidence and research issues. Med. Sci. Sports Exerc., Vol. 31, No. 11, Suppl., pp. S597–S601, 1999.

Purpose: The relation between indicators of overweight (body mass index (BMI)) and all-cause mortality and factors that potentially affect such a relationship were reviewed.

Methods: The literature was reviewed.

Results: Although there are many reports on the relationship between indicators of overweight (such as BMI) and all-cause mortality, there are no two studies that have been analyzed identically. It is now usually assumed that there is a U- or J-shaped association between BMI and mortality, but there are many issues that remain unsolved until today. These issues include the effects of: adequate control for cigarette smoking; adequate control for (sub)clinical disease at baseline; adequate control for intermediate risk factors; adequate measures for exposure to obesity; age, period, and cohort effects; adequate control for underlying lifestyle factors; adequate control or stratification for ethnicity and socioeconomic status; effects of sample size and duration of follow-up; and reliance on self-reported body weight and height.

Conclusion: The literature is dominated by studies in young adult and middle-aged white inhabitants of North America and Europe. In those populations, it seems well accepted that lowest mortality is in the range of BMI between 18.5 and 25 kg·m−2. When BMI reached values of 30 kg·m−2 or more, mortality is substantially elevated by about 50–150%. These results may not be generalizable to other populations, and more studies are needed. All evidence is of category C (observational studies).

In this short review, we will limit our discussion to the nature of the relationship between body mass index (BMI) and all-cause mortality. All of the evidence cited here is from observational studies (evidence category C). It is unlikely that the issue of body weight and mortality will ever be resolved from randomized controlled trials. The relation between indicators of overweight and mortality has been the subject of many studies over the last 40 years. Many of the important studies have recently been reviewed by Stevens (27). It is often stated that the relationship between BMI and all-cause mortality is U- or J-shaped (15,30). It has been argued that the high mortality at low BMI is due to confounding of smoking and presence of disease and that, in fact, mortality rates are progressively increasing throughout the range of BMI (11,12,16,33). The relation between BMI and mortality is still subject to debate in the literature partly because most studies suffer from methodological drawbacks. Often cited are those proposed by Manson et al. in 1987 (15):

i. failure to control for cigarette smoking;

ii. failure to eliminate effects of clinical or subclinical illness present at baseline; and

iii. inappropriate adjustment for intermediate risk factors in the relation between BMI and mortality such as hypertension, hyperlipidemia, and diabetes.

Other problems are:

iv. the use of one single estimate of overweight by the BMI (kg·m2) as an indicator of exposure of individual to excess fat mass;

v. age, period, and cohort effects that may modify the association between BMI and mortality;

vi. failure to control for underlying lifestyle factors contributing to variation in BMI in the population;

vii. failure to adequately control for effects of socioeconomic status and ethnicity;

viii. insufficient sample size and duration of follow-up both affecting the strength of the association between BMI and mortality; and

ix. reliance on self-reported height and weight.

We will briefly discuss some of these methodological issues below:

i. The potential role of smoking habits concerns the possibility that this is a confounder or an effect modifier in the relationship between overweight and death. The role of smoking as a confounder can be explained by the observation that smokers have lower BMI compared with nonsmokers as well as higher mortality rates. This could attenuate the relationship between BMI and mortality. The issue of smoking being an effect modifier of the relationship between BMI and mortality was first raised by Garrison et al. in 1983 who reported that there appeared to be a J- or U-shaped relationship between “Metropolitan Relative Weight” (MRW) and death in male smokers but a direct linear relationship in nonsmokers (7). These data have been recently reanalyzed this time with the conclusion that significant U- or J-shaped associations between MRW and mortality could be found in female and male smokers and nonsmokers (24). The authors advocate caution in interpreting relationships by visual inspection only especially when data are thin. In addition, it became clear that Garrison excluded nonsmokers at the lowest level of relative weight.

Also in other studies it has been proposed that the relationship between BMI and mortality in healthy nonsmokers is linear (11,16), but in many studies a curvilinear relationship remains.

Although the issue whether or not smoking should be considered to be a confounder or an effect modifier of the overweight-mortality relationship is not settled, it is safe to recommend to always stratify the analyses by smoking habits if only because this allows more direct estimates of absolute risk.

ii. The issue of preexisting clinical and subclinical disease is controversial. Although not many dispute the possibility that, especially at the lower ranges of BMI, some diseases may lead to dramatic weight loss as well as a high risk of dying, the methods to overcome these have not been clearly established. Because information about subclinical disease is usually unavailable, most researchers follow the advice by Manson et al. “to disregard mortality within the first few years of follow-up, based on the assumption that such deaths are largely due to disease present at entry” (15). Allison et al. recently completed a meta-analyses on this issue and concluded that “the effect of eliminating early deaths was statistically significant but minuscule in magnitude” and that “either preexisting disease does not confound the BMI-mortality association or that eliminating early deaths is inefficient for reducing that confounding” (1). This issue clearly needs further research.

iii. In early studies like the Framingham Heart Study, the association between metropolitan relative weight and mortality was evaluated from statistical models that incorporated other risk factors at the same time, such as blood pressure, serum cholesterol, and blood glucose (31) From such analyses, MRW did not appear to be independently related to mortality, which was interpreted that obesity per se was not a risk factor and that obesity is benign when it exists without other major risk factors for cardiovascular disease. Similar conclusions were also drawn by Keys from analyses of the Seven-Countries Study (8). Although these authors did acknowledge that other cardiovascular disease risk factors are likely to be intermediates in the causal chain linking obesity to mortality, the emphasis in their interpretation of the data was to focus on treatment of other risk factors than obesity. In the last 10 years or so, more authors tend to concentrate on age- and smoking-adjusted results, and emphasis has shifted toward treatment of obesity as a primary target. Another issue is that the coexistence of obesity with other risk factors for mortality such as diabetes mellitus, hypertension, and hyperlipidemia may potentiate the risk of dying in obese subjects (16,22). For some time, it was proposed that lean hypertensives have higher mortality compared with obese hypertensives, but this has not been confirmed in later studies (22,26). Stratification by risk factor status is not done often but should be considered when possible.

iv. It is well accepted that there are limitations in the use of BMI as a measure of fatness. Especially in the middle-range of BMI values, contributions of lean body mass and body fat mass are both relatively large. It is likely that body fat and lean body mass have different associations with mortality. One alternative interpretation for this curvilinear association is that it is the result of the combination of two linear functions, namely that increasing fat mass is directly associated with increasing risk and increasing lean mass is inversely associated with decreasing risk (2).

The issue of the limitation of BMI as a measure of fatness becomes an increasingly important one at older ages. For a given BMI, an older woman is likely to have more fat mass but less muscle and lean body mass than a younger woman.

Most epidemiological studies on relationships between exposures and outcomes have a limitation in common that they regard one measurement of exposure as indicative of exposure. This implies that investigators have to ignore effects of body weight changes before and after the measurement. Several reports suggest that weight change and weight fluctuations are independent determinants of mortality (14,17,34,35). Weight changes and fluctuations may be more common toward extremes of the distribution of BMI in the populations and may affect the risk estimates at these levels.

v. Age, period, and cohort effects are general issues that are not unique to the BMI-mortality relationship but may be relevant here. Age, period (calendar time), and cohort (year of birth). When analyses of the impact of age on the BMI-mortality relationship is evaluated (3,28), usually cohorts of different age-groups are followed for an identical period, and mortality rates and ratios are calculated in categories of BMI.

Not only do these groups have different ages, they also have different birth-years and may or may not share time periods that affect both mortality and body weight.

For example, one of the largest studies on BMI and mortality was performed in Norway where 1.7 million Norwegians aged 20–90 yr were recruited between 1963 and 1979 and followed for an average of 10 years (32). Ninety-year-olds recruited in 1963 were born in 1873 and experienced as youngsters the large tuberculosis epidemics at the turn of the century and also experienced World War I and II as adults. During most of their life up to 1963, infectious diseases were the major cause of death, cardiovascular diseases were rare, and life expectancy was low. Those recruited in their twenties were born in the period around World War II when food restriction was common. Thus, besides an effect of age, we may also be looking at effects of different past and unshared exposures affecting body weight development and mortality and generizability to people of different ages living today may not be appropriate.

vi. The issue of control for underlying lifestyle factors is one that has not been dealt with extensively in the past. Variation in body weight is partly the result of variation in energy expenditure and energy intake. It is thought that obese subjects are less active but have higher metabolic rates compared with nonobese subjects. Cause and effect are difficult to ascertain in the activity-obesity relationship. Physical inactivity may lead to increased obesity (21). Alternatively, obese subjects have often physical limitations caused by impaired respiratory function and musculoskeletal problems (9), which may lead to reduced physical activity. Blair et al. have proposed that high levels of obesity may be relatively benign when accompanied by high levels of fitness (10). The coexistence of high levels of fitness and obesity, however, may be rather low in most sedentary societies.

Also with respect to dietary intake data there are problems in methodology and interpretation of findings. High fat intakes have been implicated in the development and sustaining of high body weights, although the evidence is not as strong as often proposed. With regard to food intake data, there is the additional problem of underreporting of intake and intermittent dieting (4).

The risk of obesity may, however, be very well dependent upon underlying behavioral characteristics. Obesity produced in Finnish male subjects by high intakes of beer, sedentary lifestyle, and high-fat diets (19) may be associated with very different health risks compared with obesity in traditional Mediterranean diets and heavy labor. In a study comparing measures of obesity and other risk factors among European women aged 38 yr, we observed that Dutch women had an average BMI of 23 kg·m2 and women in a village outside Naples, Italy, had a mean BMI of 28 kg·m2 (23). Nevertheless, cholesterol levels were higher in Dutch women (mean: 5.6 mmol·L1 vs 4.9 mmol·L1 in Italian women). This may reflect underlying differences in diet and/or genetic factors. Alcohol consumption may also be important in this respect. In one study, it was observed that moderate and high alcohol consumption were associated with increased mortality especially at low levels of BMI (5).

vii. Effects of socioeconomic status and ethnicity: low socioeconomic status is related to increased prevalence of obesity as well as increased mortality and this may therefore confound the relationship between obesity and mortality. Many of these confounding effects may be due to underlying differences in lifestyle, but it is also likely that high educational levels reflect other reasons for low mortality and low BMI (like better access to adequate medical interventions and prevention programs). Few studies have adequately dealt with the influence of components of socioeconomic status (education, income, and profession) on the BMI-mortality relationship. Tayback et al. (29), for instance, observed in elderly women that low BMI was associated with increased mortality only in women with poor socioeconomic status. Ethnicity is in many societies strongly associated with socioeconomic status. Without trying to be complete, it is clear that the relations between BMI and mortality vary across ethnic groups. This partly reflects limited appropriateness of BMI as an indicator of body fatness (6), but there may be other reasons. In Pima Indians, for instance, relative excess mortality is not observed until BMI reach values well over 40 kg·m2 (18). At BMI levels below 40 kg·m2, there is no relationship between BMI and mortality in this particular subgroup.

viii. Sample size and duration of follow-up: Sjöström (25) showed in a review of 51 cohorts (40 studies) that studies in which no relationship between BMI and mortality was found were clustered among small and/or short-term studies. All studies with more than 20,000 subjects and 20 of 21 studies with more than 7000 subjects showed a positive association between BMI and mortality. He estimated that follow-up needs to be longer than about 5 years (but preferably much longer) to be able to demonstrate effects of obesity and mortality. It seems obvious that duration of follow-up is important (in young women, a short follow-up will not be sufficient to see the long-term impact of obesity on chronic diseases and mortality), but very little systematic work is done in this area. For instance, in Seventh-Day Adventist women followed for 26 yr, it was observed that in those aged 30–54 a weak linear association between BMI and mortality was observed during the first 8 yr of follow-up, a significant linear relation during years 9–14 of follow-up, and a U-shaped association during 15–16 yr of follow-up (13).

ix. Self-report of weight and height. Many of the largest studies use self-reported height and weight. Although the validity is reasonably adequate for ranking individuals, there is evidence that with increasing degrees of obesity subjects tend to underestimate their weight (20).

Measures of effect: Usually the relation between BMI and mortality is expressed in terms of relative risk in categories of BMI as compared with an arbitrary reference category. The choice of reference categories and cut-off points for other categories will greatly affect the estimates of relative risk. Absolute risks or population attributable risks are alternatives less often used but these suffer from the same drawbacks with respect to choice of classification of BMI.

Differences in life expectancy are also very illustrative, but we know of no studies in which this was expressed with respect to overweight and obesity. From the data of the Nurses’ Health study, we estimated differences in life expectancy by degree of overweight taking the leanest subgroup (average BMI 18.5 kg·m2 as the reference). Only data of women who had never smoked were used. From our calculations, we estimate that women with BMI over 30 kg·m2 live about 5–6 yr fewer than very lean women (Fig. 1).

Figure 1
Figure 1
Image Tools

This is a compelling illustration of the importance of preventing and treating obesity adequately and vigorously.

Back to Top | Article Outline


Although many studies have been published on the relation between BMI and all-cause mortality, there are still many issues that need to be resolved before definite conclusions can be made (see Table 1). Most of the evidence points toward a U- or J-shaped association. The extent to which the increased mortality is attributable to smoking and (sub)clinical illness remains not totally clear, but both do not seem to play a major role at high BMI levels. Among the research priorities are the development of more appropriate measures of body fatness especially in the elderly and nonwhite populations. The effects of underlying lifestyles leading to obesity and leanness as well as to mortality may be among the most important research topics. This is not an easy task and requires the development of accurate methodology that allows quantification of energy intake and expenditure in population studies.

Table 1
Table 1
Image Tools
Back to Top | Article Outline


1. Allison, D., M. Faith, K. Carpenter, D. Flanders, and D. Williamson. An evaluation of the strategy of eliminating early deaths as a means of controlling for confounding due to preexisting disease. Obes. Res. (in press).

2. Allison, D. B., M. S. Faith, M. Heo, and D. P. Kotler. Hypothesis concerning the u-shaped relation between body mass index and mortality. Am. J. Epidemiol. 146: 339–349, 1997.

3. Andres, R, D. Elahi, J. Tobin, D. C. Muller, and L. Brant. Impact of age on weight goals. Ann. Intern. Med. 103: 1030–1033, 1985.

4. Braam, Lajlm, M. C. Ocké, H. B. Bueno de Mesquita, and J. C. Seidell. Determinants of obesity-related underreporting of energy intake. Am. J. Epidemiol. 147:1081–1086, 1998.

5. Chyou, P-H., C. M. Burchfiel, K. Yano, et al. Obesity, alcohol consumption, smoking, and mortality. Ann. Epidemiol. 7: 311–317, 1997.

6. Deurenberg, P., M. Yap, and W. A. van Staveren. Body mass index and percent body fat: a meta analysis among different ethnic groups. Int. J. Obes. 22: 1164–1171, 1998.

7. Garrison, R, M. Feinleib, W. Castelli, and P. McNamara. Cigarette smoking as a confounder of the relationship between relative weight and long-term mortality. JAMA 249: 2199–2203, 1983.

8. Keys, A. Longevity of man: relative weight and fatness in middle age. Ann. Med. 21: 163–168, 1989.

9. Lean, M. E. J., T. S. Han, and J. C. Seidell. Impairment of health and quality of life using new US Federal guidelines for the identification of obesity. Arch. Intern. Med. 159: 837–843, 1999.

10. Lee, C. D., A. S. Jackson, and S. N. Blair. US weight guidelines: is it also important to consider cardiorespiratory fitness? Int. J. Obes. 22(Suppl. 2): S2–S7, 1998.

11. Lee, I-M., J. E. Manson, C. H. Hennekens, and R. S. Paffenbarger. Body weight and mortality: a 27-year follow-up of middle-aged men. JAMA 270: 2823–2828, 1993.

12. Lindsted, K, S. Tonstad, and J. W. Kuzma. Body mass index and patterns of mortality among Seven Day Adventist men. Int. J. Obes. 15: 397–406, 1991.

13. Lindstedt, K. D., and P. N. Singh. Body mass and 26-year risk of mortality among women who never smoked: finding from the Adventist mortality study. Am. J. Epidemiol. 146: 11, 1997.

14. Lissner, L., P. M. Odell, R. B. d’Agostino, et al. Variability of body weight and health outcomes in the Framingham population. N. Engl. J. Med. 27:324, 1839–1844, 1991.

15. Manson, J. E., M. J. Stampfer, C. H. Hennekens, and W. C. Willett. Body weight and longevity: a reassessment. JAMA 257: 353–358, 1987.

16. Manson, J. E., W. C. Willett, M. J. Stampfer, et al. Body weight and mortality among women. N. Engl. J. Med. 333: 677–685, 1995.

17. Peters, E. Th. J., J. C. Seidell, A. Menotti, et al. Changes in body weight in relation to mortality in 6,441 European middle-aged men: the Seven Countries Study. Int. J. Obes. 19: 862–868, 1995.

18. Pettit, D. J., J. R. Lisse, W. C. Knowler, and P. H. Bennet. Mortality as a function of obesity and diabetes mellitus. Am. J. Epidemiol. 115: 359–366, 1982.

19. Rissanen, A. M., M. Heliovaara, P. Knekt, A. Reunanen, and A. Aromaa. Determinants of weight gain and overweight in adult Finns. Eur. J. Clin. Nutr. 45: 419–430, 1991.

20. Rowland, M. Self-reported weight and height. Am. J. Clin. Nutr 52: 1125–1133, 1990.

21. Saris, W. H. M. Fit, fat and fat free: the metabolic aspects of weight control. Int. J. Obes. 22(Suppl. 2): S15–S21, 1998.

22. Seidell, J. C., W. M. M. Verschuren, E. M. van Leer, and D. Kromhout. Overweight, underweight, and mortality: a prospective study in 48,287 men and women. Arch. Intern. Med. 156: 958–963, 1996.

23. Seidell, J. C., M. Cigolini, J. Charzewska, et al. Indicators of fat distribution, serum lipids, and blood pressure in European women born in 1948: the European fat distribution study. Am. J. Epidemiol. 130: 53–66, 1989.

24. Sempos, C. T., R. Durazo-Arvizu, D. L. McGee, R. S. Coopers, and T. E. Prewitt. The influence of cigarette smoking on the association between body weight and mortality: the Framingham Heart Study revisited. Ann. Epidemiol. 8: 289–300, 1998.

25. Sjöström, L. V. Mortality of severely obese subjects. Am. J. Clin. Nutr 55: 516, 1992.

26. Stamler, R., C. E. Ford, and J. Stamler. Why do lean hypertensives have higher mortality rates than other hypertensives? Findings of the Hypertension Detection and Follow-up Program. Hypertension 17: 553–564, 1991.

27. Stevens, J. Studies of the impact of age on optimal body weight. J. Nutr. Biochem. 9: 501–10, 1998.

28. Stevens, J., J. Cai, E. R. Pamuk, D. F. Williamson, M. J. Thun, and J. L. Wood. The effect of age on the association between body-mass index and mortality. N. Engl. J. Med. 338: 1–7, 1998.

29. Tayback, M., S. Kumanyika, and E. Chee. Body weight as a risk factor in the elderly. Arch. Intern. Med. 150: 1065–1072, 1990.

30. Troiano, R. P., E. A. Frongillo, and D. A. Levitsky. The relationship between relative body weight and all-cause mortality, increased risk at both ends of the scale. Int. J. Obes. (in press).

31. Truett, J., J. Cornfield, and W. B. Kannel. A multivariate analyses of the risk of coronary heart disease in Framingham. J. Chron. Dis. 20: 511–20, 1967.

32. Waaler, H. Height, weight, and mortality: the Norwegian experience. Acta Med. Scand.Suppl. 679: 1–56, 1984.

33. Wannamethee, G., and A. G. Shaper. Body weight and mortality in middle aged British men: impact of smoking. Br. Med. J. 299: 1497–1502, 1989.

34. Willett, W. C., J. E. Manson, M. J. Stampfer, et al. Weight, weight change, and CHD in women: risk within the “normal” weight range. JAMA 273: 461–465, 1995.

35. Williamson, D., E. Pamuk, M. Thun, D. Flanders, T. Byers, and C. Heath. Prospective study of intentional weight loss and mortality in never-smoking US white women aged 40–64 years. Am. J. Epidemiol. 141: 1128–1141, 1995.



© 1999 Lippincott Williams & Wilkins, Inc.


Article Tools



Search for Similar Articles
You may search for similar articles that contain these same keywords or you may modify the keyword list to augment your search.

Connect With Us