Height and Body Mass Index in Relation to Total Mortality : Epidemiology

Secondary Logo

Journal Logo


Height and Body Mass Index in Relation to Total Mortality

Engeland, Anders1; Bjørge, Tone2; Selmer, Randi Marie1; Tverdal, Aage1

Author Information
doi: 10.1097/01.EDE.0000047889.30616.73



Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

Overweight and obesity represent a major and rapidly growing threat to the health of populations worldwide in both developing and developed countries. 1 The relation of height with total mortality, and the relation of weight and height combined (body mass index or BMI) with total mortality, has been explored in a number of studies. 2–6 However, the results of past studies are not easily compared because of differences in BMI cut points (some calculated within the different data sets), exclusion criteria, length of follow-up, study design and age groups. 1

It is not clear whether the relation between BMI and mortality is U-shaped, J-shaped or more linear. 5,7 Several studies have found excess mortality at both low and high levels of BMI. Another question is whether the “optimal” BMI, ie, the BMI giving the lowest death rate, varies with sex and age. 8–10

Other controversies in studies of the relation between BMI and mortality are whether and how adjustments for smoking habits, persistent diseases and physical activity should be performed. Some earlier studies on BMI and mortality that showed increased risk of death among very lean persons have been criticized for not excluding smokers and persons with persistent diseases. Because both smoking and illness are associated with leanness and mortality, this might explain the increased risk of death among lean persons. Some argue for restricting analyses of the relation between BMI and mortality to never-smokers, 8 whereas others have found no support for such restrictions. 6

The relation between height and total mortality has received less attention. The relationship seems to be inverse, 4,11ie, mortality decreases with increased height. Although the height of a person is strongly related to genetic factors, environmental factors during childhood and adolescence may influence the extent to which individuals reach their genetically determined height.

The present study was set up to explore the relation of height and BMI with total mortality in a cohort comprising two million Norwegian men and women age 20–74 years at measurement.


Study Population

Height and weight measurements were performed in the period 1963–1975 as part of a screening program aimed at detecting tuberculosis in the general Norwegian population. 2,12,13 This mass examination was compulsory for persons above the age of 15 and enrolled 1.7 million people. Previous reports have described the impact of height and weight on morbidity and mortality. 2,12 The present study is an extension, with regard to both follow-up and number of persons included (extended from 14.5 to 44 million person-years in persons aged 20–74 at measurement). In 1963 and from 1972 onwards, height and weight measurements were also collected in health surveys in various parts of Norway. 13–16 Hence, the present study cohort was collected from two main sources:2,13

  1. A compulsory population-based screening for tuberculosis. The attendance was about 85% in persons above the age of 15.
  2. Several population-based health surveys conducted in 1963 and 1972 and onward. The attendance in the mid 1970s was 85–90%, but decreased to around 75% in the mid 1990s.

In 1963–2000, 2,907,950 height and weight measurements were performed on more than 2.2 million Norwegians. The body weights (kilograms) were measured on scales that were regularly calibrated; weight was recorded to the nearest half-kilogram. Body height was measured and recorded to the nearest centimeter. Height was measured without shoes, and weight was measured with the subject wearing light clothing. If height and weight measurements were performed under irregular circumstances, this was noted (measured with shoes, the persons were pregnant, etc.).

The following exclusions were made in the present analysis:

  1. Year of measurement was missing (396 measurements).
  2. Measurements for which irregularities were noted (101,602 measurements).
  3. Measurements in persons younger than 20 years or above 74 years (327,364 measurements).
  4. Measurements with registered height below 120 cm or weight below 20 kg (171 measurements; 161 of these were lacking measurement of height or weight).

The earliest accepted measurement between 20 and 75 years of age was used for each person. Altogether, 1,978,999 persons with measurements were eligible for the analysis. BMI was defined as (weight in kilograms)/(height in meters). 2

All residents of Norway have a unique 11-digit individual identification number. By linkage to the Death Registry at Statistics Norway, it was possible to follow all persons in the present study from date of measurement until emigration, the age of 100, death or June 30, 2001, whichever occurred first. A relatively small number of persons (N = 561) were lost to follow-up. Furthermore, 485 persons were excluded because their measurements were taken after end of follow-up (when day and month of measurement were missing, the date was set to June 30, resulting in that some persons who emigrated or died between their actual date of measurement and June 30). After the exclusion, 1,977,953 people remained in the analysis sample. Three fourths of these were measured within the compulsory screening for tuberculosis (Table 1).

Number of Individuals Included in the Main Analyses by Source of Information

Most of the health surveys (but not the assessments performed in connection with screening for tuberculosis) included questions about smoking habits (self-administered questionnaire). Data on smoking habits were not included in the main analysis. However, a subanalysis was performed including only those who had answered questions on smoking habits (same follow-up procedure as above, average follow-up time 11.1 years). Some participants in the tuberculosis screening also participated in later health surveys (N = 189,013); they were included in the subanalyses if the smoking habits were known but with a later inclusion date. Analyses were performed separately for never-smokers to investigate whether smoking habits could explain differences in survival between groups with different body mass or height. The same exclusion criteria were used in the subanalysis as in the main analysis. Information on smoking habits was available for 622,308 persons after exclusions.

Analytic Methods

We fitted multivariate Cox proportional hazards regression models, with time since measurement as the time variable, to obtain relative risk (RR) estimates of dying. 17 That is, we assumed that the hazard function for a person with a covariate vector x = (x1, x2,…xp)’ could be expressed by h (t;x) =h0(t)exp(x’β), where h0(t) represents the hazard function for a person with covariate values all equal zero, and β = (β1, β2,…βp)’ is a vector of regression coefficients. In the analyses, the following categorized variables were included 1:

  1. Age at measurement: 20–24,..., 70–74.
  2. Year of birth: <1900, 1900–1909,..., 1950–59, ≥1960.
  3. BMI (ie, (weight in kilograms)/(height in meters)2: <17.50, 17.50–18.49,..., 22.50–23.49, 23.50–24.99, 25.00–27.49,..., 37.50–39.99, ≥40.00. This categorization is more detailed but consistent with the recommendations from a World Health Organization (WHO) Consultation on Obesity in 1997.
  4. Height:
  5. a. Men: <150, 150–154,..., 190–194, ≥195.
  6. b. Women: <140, 140–144,..., 180–184, ≥185.

Variables for area of living were first included in the models, but because the inclusion of these did not change the relation between BMI and height and mortality, they were omitted from the analyses. Analyses were also performed including observation time from 5 to 15 years after measurement to reduce the possible impact of persistent disease and to limit the length of follow-up. To find the “optimal BMI,”ie, the BMI value associated with the lowest mortality, we used a method described earlier. 6,18 This method assumes a quadratic relation between the “lean” BMI (1/BMI), LBMI and mortality. A model where parameters for linear, β1 and quadratic term, β2, of LBMI replace the parameters for the categorized BMI was fitted. The BMI connected to the lowest mortality was estimated by EQUATION

The variance of BMImin was estimated using the delta method:EQUATION

Analyses were performed separately for each sex. The results were presented as RR of dying with 95% confidence intervals (CI).


A total of 1,977,953 persons (952,873 men and 1,025,080 women) (mean age at measurement: 45 years) were followed for an average of 22.1 years (range 0–38 years), comprising 43,810,181 person-years (data available with the electronic version of this article at http://www.epidem.com). We identified 722,606 deaths in the cohort. Mean age at death was 72.9 years in men and 77.3 years in women. The overall mean BMI was 24.9 in men and 25.0 in women. The mean height increased with increasing birth years from 170 cm for men born around 1890 to 180 cm for those born around 1975 and from 157 to 167 cm for women (mean annual increase: 0.12 cm in both sexes).

Table 2 shows the RR of dying by categories of BMI and height from a multivariate analysis. The association between BMI and risk of dying seems to be J- or U-shaped with highest risk in the thinnest and most obese persons. The lowest risk of death is associated with BMI between 22.50 and 24.99 in both men and women. Persons with BMI below 17.50 had 74% and 96% higher risk of death, in men and women, respectively, than persons with BMI between 22.50 and 23.49 (reference group). Persons with BMI above 40 had 153% and 93% higher risk of death than the reference group in men and women, respectively.

Relative Risk (RR) of Dying, from Multivariate Analysis. Age at Measurement (11 Categories) and Birth Cohort (Eight Categories) Were Included in the Model, in Addition to Body Mass Index (BMI) and Height

In men, height was inversely related to the risk of death with an elevated risk among men below 150 cm (Table 2). The tallest men had the lowest mortality. In women, mortality decreased with increasing height to about 160–164 cm and increased thereafter. Women below 140 cm had 88% higher mortality than those 160–164 cm.

Figures 1 and 2 show the mortality rate for BMI categories by age at measurement. Men measured at the age of 20–29 seem to have a somewhat different risk profile by BMI than the other age groups. This was easier to notice when we fit Cox regression models in separate analyses for each age group and calculated RR. The elevated risk at low BMI levels seen in the other age groups was not present. Analyses were also performed with smaller age groups. The nadir of the mortality curve for the age 20–24 measurements was at a lower BMI than at older ages, and there were indications of a higher mortality in BMI levels below 21.50–22.49 than at this level (RR = 1.10, 95% CI = 1.00–1.21). However, among men measured at ages 20–24, there were relatively few deaths at the lowest BMI (14, 53 and 190 deaths at the three lowest BMI levels). The J- or U-shaped relation between BMI and mortality was also present when the analysis was performed stratified on birth cohort and height category.

Mortality rate by body mass index (BMI) and age (years) at measurement; men.
Mortality rate by body mass index (BMI) and age (years) at measurement; women.

The BMI value at measurement associated with lowest subsequent mortality is tabulated in Table 3. In men, the optimal BMI increased with age from 21.6 (at age 20–29) to 24.0 (at age 70–74). In women, the optimal BMI was higher than for men at all ages and increased from 22.2 (at age 20–29) to 25.7 (at age 70–74).

Body Mass Index (BMI) at Measurement Associated with Lowest Subsequent Mortality, from Multivariate Analysis with Adjustment for Birth Cohort (Eight Categories) and Height

We performed an analysis similar to that presented in Table 2, with an observation period from 5 to 15 years after measurement only (data available with the electronic version of this article at http://www.epidem.com). This analysis included 15,749,400 person-years. The overall patterns of the association between height and mortality and BMI and mortality did not change.

In the subcohort with known smoking habits, we performed an analysis restricted to never-smokers (Table 4), including 2,477,303 person-years (mean follow-up time 10.9 years). Roughly similar patterns were observed for BMI as in the main analysis. The tallest never-smoking women had an elevated risk of death, although in the two highest categories there were only 14 and 3 deaths, respectively. A nonlinear (quadratic) relationship was confirmed for both men and women (P < 0.001), with BMI as a continuous variable.

Relative Risk (RR) of Dying, from Multivariate Analysis for Those Who Were Never-Smokers at the Time of Measurement. Age at Measurement (11 Categories) and Birth Cohort (Eight Categories) Were Included in the Model, in Addition to Body Mass Index (BMI) and Height


In this study, we explored the relations of height and BMI with mortality in a cohort of two million Norwegian men and women. The RR of death by BMI was nonlinear, with the lowest rates of death in both men and women with BMI between 22.5 and 25.0. Mortality decreased with increasing height for men; for women mortality decreased with increasing height until about the height of 160–164 cm and increased among taller women.

Information on potential confounders other than age, area of residence, sex and birth year was lacking in the present study. Other potential confounders are smoking habits, prevalent diseases at measurement, weight changes during follow-up and physical activity. However, smoking habits and physical activity might be influenced by BMI and, hence, these factors might be intermediate variables not confounders. Although it can be hypothesized that smoking and prevalent diseases are confounders or effect modifiers in the relation between BMI and mortality, there are indications that the impact of these factors is limited in the general population. 19,20

The findings of an elevated risk of death in the thinnest persons have been attributed to inadequate adjustment for smoking. 1,8 Smokers tend to weigh less and have higher mortality rates than nonsmokers. 8 Some studies have adjusted for smoking 3,6,7 whereas others have questioned whether statistical adjustment for smoking is entirely satisfactory. 8 The alternative is to restrict analyses to never-smokers only. In an American study of nurses, a J-shaped relation between BMI and mortality was observed when the analyses were adjusted for smoking, but there was no evidence of a J-shaped curve in never-smokers. 3 In the present analysis, information on smoking status was available for a subcohort, and separate analyses were made for never-smokers in this subcohort. Elevated risks were observed in the thinnest persons and also in never-smokers. It should be noted, however, that the RR of death among the heaviest individuals was greater in never-smokers than in the entire population.

Elevated risk of death in the thinnest persons has been explained by the composition of the group, which presumably includes smokers and persons who have lost weight as a result of underlying disease. 8 Exclusion of people with preexisting and subclinical diseases, however, has not changed the pattern of the relation between BMI and mortality in other studies. 7,21 Because no information on persistent diseases at measurement was available, we could not exclude persons with such diseases. Instead, additional analyses were performed with exclusion of the first 5 years after measurement and final end of follow-up at 15 years after measurements, following the recommendations of Willett et al. 8 In the present study, these restrictions had no impact on the results.

The average follow-up time in the present study was 22 years. During follow-up, the persons were categorized according to their baseline BMI. Many of the study subjects may have gained or lost weight during follow-up. Weight change seems to be an important risk factor independent of initial BMI, 1,21 but to study the impact of weight changes on mortality, it is necessary to have many weight measurements for each person as well as information on reasons for eventual weight changes. 22 However, the analysis with follow-up restricted to 5 to 15 years after measurement should also decrease the eventual impact of changed BMI during follow-up on the relation between baseline BMI and mortality. 8

A subcohort (N = 22,000) of the population in the present study has been used to explore the effect of physical activity and smoking on the relation between BMI and mortality. 20 In this subcohort, more information on each subject (men 35–49 years old measured during 1974–1978) was available. The study suggested a J-shaped association between BMI and total mortality overall and also when stratified on smoking habits and physical activity. However, the J- or U-shaped association between BMI and mortality disappeared with increasing level of physical activity.

The study subjects in the present study were recruited from population-based studies with high attendance. However, these studies were performed over a long period of time, from 1963 during 2000, and therefore we cannot generalize the study results to the Norwegian population at a certain time. However, unlike many other studies on this issue, eg, American Nurses, 3 our study population was recruited from the general population rather than selected groups.

In contrast with many other studies, we can rely on measured height and weight instead of self-reported measures. Both in the compulsory tuberculosis-screening program and in the later health surveys, the measurements were performed in a standardized way. In 1984, a large subset of the subjects in the present study (N= 1,537,000) was analyzed in a similar study. 2 However, the inclusion of 440,000 “new” individuals and a much longer follow-up (until June 2001, compared with December 1979) increases the power of the present study markedly.

Other than previous publications from subcohorts of the present study, 2,20,23 the largest previous studies (such as the American Cancer Prevention Study I and II 24 and the Nurses’ Health Study 3) have relied on self-reported height and weight. Self-reported height and weight have shown to vary systematically with BMI, 25 although views differ on how important these biases are. 8 Limitations in self-reporting of heights and weights have also been observed in older adults (>60 years). 26 In a subsample of American nurses, self-reported weights were highly correlated with measured weights, but the measured weights were on average 1.5 kg higher. 3,27 If a woman reports her height and weight to be 1.65 meter and 55 kg, while her weight actually is 56.5, the BMI is calculated to be 20.20 instead of 20.75. Consequently, the relationship between BMI and mortality should not be materially different in studies with self-reported and measured weights. However, the BMI may be shifted to the left in studies with self-reported weights, depending on the pattern of underreporting, resulting in a too low “optimal” BMI. In addition to underreporting of weight, height is often over reported, especially in obese persons and in person above 60 years. 26,28

Because of the very large number of study participants, we were able to create a finer categorization of BMI than in earlier studies and to analyze many categories for both BMI and height instead of using these variables as continuous variables. The effects of different levels of the variables were then not “forced” into a specific pattern. It was also possible to estimate the mortality in groups with very low BMI (<17.50) and with very high BMI (≥40.00). Bender et al. 29 claimed that their study was the largest mortality follow-up of a cohort of obese patients. Their study included 344 men and 1,191 women with BMI above 40. In the present study, 713 men and 6,050 women with BMI above 40 were included.

The detailed categorization of BMI used in the present study is consistent with the recommendations from a WHO Consultation on Obesity in 1997. 1 Stevens et al. 24 recently evaluated the categorization in the WHO report by the use of mortality rates in a large American cohort of never-smokers. This study supported the appropriateness of the suggested lower BMI cutoff point for overweight of 25.0. In the present study, we used an even more detailed categorization than suggested by WHO for “underweight,” “normal” weight, and for “overweight.” The results suggest that the lower categories within the “normal range” suggested in the WHO report (BMI 18.5–24.9) have higher mortality than those with BMI from 25.00 to 27.49.

Although our estimates varied by age and sex, there was a large interval of BMI values in all age and sex groups where the risk of death was only moderately different from the nadir of the curves. The higher “optimal” BMI at the older age may not mean that it is healthier for older people to be heavier. Low weight measured at younger age is more likely to represent healthy leanness whereas low weight measured at older age is more likely to represent leanness because of illness.

In the present study, we have used RR as the measure of effect. This must be taken into consideration when comparing the relative effect in different age groups. Even though the RR of dying in different BMI groups is smaller in elderly than in younger persons, the difference in absolute mortality rates between the BMI groups might be much higher in the elderly than in younger persons. 9

A 10% decrease in mortality per 5 cm increase in height was observed in a cohort of Finnish men and women when adjusting for age and birth cohort. 4 In the present study, a monotonic decreasing mortality by increasing height was observed in men; the mean decrease in mortality per 5 cm increase in height was 3.5% (CI = 3.3–3.8). In women, the mean decrease in mortality per 5 cm increase in height was 1.6% (CI = 1.3–1.9). However, restricting the analysis to women below 165 cm, the mean decrease in mortality per 5 cm increase in height was 3.8% (CI = 3.4–4.3).

In conclusion, the overall relationship between BMI and mortality was J- or U-shaped; the highest mortality was observed in persons with low and high BMI. Neither the limitation of follow-up to the period from 5 to 15 years after measurement nor the analysis of never-smokers only changed this relation. The results suggest that the “normal range” of BMI, suggested in the WHO report, 1 should be shifted upwards because higher mortality rates were observed in the lower “normal range” than in the lower “pre-obese” category.


We express gratitude to those who during almost 40 years have collected the data used in the present study. These are persons connected to the former National Health Screening Service, The Nord-Trøndelag health survey (HUNT), The Hordaland health survey (HUSK) and The Tromsø study.


1. World Health Organization Consultation on Obesity. Preventing and Managing the Global Epidemic: Report of a WHO Consultation on Obesity, Geneva, 3–5 June 1997. Geneva, Switzerland: World Health Organization, 1998.
2. Waaler HT. Height, weight and mortality. The Norwegian experiment. Acta Med Scand 1984;suppl 679:1–56.
3. Manson JE, Willett WC, Stampfer MJ, et al. Body weight and mortality among women. N Engl J Med 1995; 333: 677–685.
4. Jousilahti P, Tuomilehto J, Vartiainen E, Eriksson J, Puska P. Relation of adult height to cause-specific and total mortality: a prospective follow-up study of 31,199 middle-aged men and women in Finland. Am J Epidemiol 2000; 151: 1112–1120.
5. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW Jr. Body-mass index and mortality in a prospective cohort of U.S. adults. N Engl J Med 1999; 341: 1097–1105.
6. The BMI in Diverse Populations Collaborative Group. Effect of smoking on the body mass index-mortality relation: empirical evidence from 15 studies. Am J Epidemiol 1999; 150: 1297–1308.
7. Song YM, Sung J. Body mass index and mortality: a twelve-year prospective study in Korea. Epidemiology 2001; 12: 173–179.
8. Willett WC, Dietz WH, Colditz GA. Guidelines for healthy weight. N Engl J Med 1999; 341: 427–434.
9. Stevens J, Cai J, Pamuk ER, Williamson DF, Thun MJ, Wood JL. The effect of age on the association between body-mass index and mortality. N Engl J Med 1998; 338: 1–7.
10. Stevens J, Cai J, Juhaeri, Thun MJ, Williamson DF, Wood JL. Consequences of the use of different measures of effect to determine the impact of age on the association between obesity and mortality. Am J Epidemiol 1999;150:399–407.
11. Davey Smith G, Hart C, Upton M, et al. Height and risk of death among men and women: aetiological implications of associations with cardiorespiratory disease and cancer mortality. J Epidemiol Community Health 2000; 54: 97–103.
12. Tverdal A. Body mass index and incidence of tuberculosis. Eur J Respir Dis 1986; 69: 355–362.
13. Bjartveit K. [The National Health Screening Service: From fight against tuberculosis to many-sided epidemiological activities]. Nor Epidemiol 1997; 7: 157–174.
14. Bjartveit K, Foss OP, Gjervig T, Lund-Larsen PG. The cardiovascular disease study in Norwegian counties. Background and organization. Acta Med Scand Suppl 1979; 634: 1–70.
15. Stensvold I, Tverdal A, Foss OP. The effect of coffee on blood lipids and blood pressure. Results from a Norwegian cross-sectional study, men and women, 40–42 years. J Clin Epidemiol 1989; 42: 877–884.
16. Tverdal A. [Prevalence of obesity among persons aged 40–42 years in two periods]. Tidsskr Nor Laegeforen 2001; 121: 667–672.
17. Cox DR, Oakes D. Analysis of Survival Data. London: Chapman and Hall Ltd, 1984; 1–201.
18. Durazo-Arvizu R, McGee D, Li Z, Cooper R. Establishing the nadir of the body mass index-mortality relationship: a case study. J Am Stat Assoc 1997; 92: 1312–1319.
19. Durazo-Arvizu RA, McGee DL, Cooper RS, Liao Y, Luke A. Mortality and optimal body mass index in a sample of the U.S. population. Am J Epidemiol 1998; 147: 739–749.
20. Meyer HE, Søgaard AJ, Tverdal A, Selmer RM. Body mass index and mortality: the influence of physical activity and smoking. Med Sci Sports Exerc 2002; 34: 1065–1070.
21. Mikkelsen KL, Heitmann BL, Keiding N, Sørensen TIA. Independent effects of stable and changing body weight on total mortality. Epidemiology 1999; 10: 671–678.
22. Byers T. The observational epidemiology of changing weight: an appeal for reasons. Epidemiology 1999; 10: 662–664.
23. Tverdal A, Thelle DS, Stensvold I, Leren P, Bjartveit K. Mortality in relation to smoking history: 13 years’ follow-up of 68,000 Norwegian men and women 35–49 years. J Clin Epidemiol 1993; 46: 475–487.
24. Stevens J, Cai J, Juhaeri, Thun MJ, Wood JL. Evaluation of WHO and NHANES II standards for overweight using mortality rates. J Am Diet Assoc 2000;100:825–827.
25. Plankey MW, Stevens J, Flegal KM, Rust PF. Prediction equations do not eliminate systematic error in self-reported body mass index. Obes Res 1997; 5: 308–314.
26. Kuczmarski MF, Kuczmarski RJ, Najjar M. Effects of age on validity of self-reported height, weight, and body mass index: findings from the Third National Health and Nutrition Examination Survey, 1988–1994. J Am Diet Assoc 2001; 101: 28–34.
27. Willett W, Stampfer MJ, Bain C, et al. Cigarette smoking, relative weight, and menopause. Am J Epidemiol 1983; 117: 651–658.
28. Nawaz H, Chan W, Abdulrahman M, Larson D, Katz DL. Self-reported weight and height: implications for obesity research. Am J Prev Med 2001; 20: 294–298.
29. Bender R, Jockel KH, Trautner C, Spraul M, Berger M. Effect of age on excess mortality in obesity. JAMA 1999; 281: 1498–1504.

body mass index; BMI; height; mortality; cohort study; Norway

Supplemental Digital Content

© 2003 Lippincott Williams & Wilkins, Inc.