Skip Navigation


American Journal of Epidemiology Advance Access originally published online on August 1, 2007
American Journal of Epidemiology 2007 166(8):975-982; doi:10.1093/aje/kwm152
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
166/8/975    most recent
kwm152v2
kwm152v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (5)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Flegal, K. M.
Right arrow Articles by Gail, M. H.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Flegal, K. M.
Right arrow Articles by Gail, M. H.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

American Journal of Epidemiology Published by the Johns Hopkins Bloomberg School of Public Health 2007.

PRACTICE OF EPIDEMIOLOGY

Impact of Smoking and Preexisting Illness on Estimates of the Fractions of Deaths Associated with Underweight, Overweight, and Obesity in the US Population

Katherine M. Flegal1, Barry I. Graubard2, David F. Williamson3 and Mitchell H. Gail2

1 National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, MD
2 Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD
3 Division of Diabetes Translation, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, Atlanta, GA

Correspondence to Dr. Katherine Flegal, National Center for Health Statistics, 3311 Toledo Road, Hyattsville, MD 20782 (e-mail: kflegal{at}cdc.gov).

Received for publication October 20, 2006. Accepted for publication April 23, 2007.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Studies of body weight and mortality sometimes exclude participants who have ever smoked or who may have had preexisting illness at baseline. This exclusionary approach was applied to data from the National Health and Nutrition Examination Surveys to investigate the potential effects of smoking and preexisting illness on estimates of the attributable fractions of US deaths in 2000 that were associated with different levels of body mass index (BMI; weight (kg)/height (m)2). Synthetic estimates were calculated by using postexclusion relative risks for BMI categories in place of BMI relative risks from the full sample, holding the relative risks for all other covariates constant. When the postexclusion relative risks were used, the attributable fractions of deaths associated with underweight and with higher levels of obesity increased slightly and the attributable fractions of deaths associated with overweight and with grade 1 obesity decreased slightly. The relative risks for BMI categories did not show large or systematic changes after simultaneous exclusion of ever smokers, persons with a history of cancer or cardiovascular disease, and persons who died early in the follow-up period or had their heights and weights measured at older ages. These analyses suggest that residual confounding by smoking or preexisting illness had little effect on previous estimates of attributable fractions from nationally representative data with measured heights and weights.

body mass index; body weight; confounding factors (epidemiology); epidemiologic methods; health surveys; mortality; overweight; thinness

Abbreviations: BMI, body mass index; NHANES, National Health and Nutrition Examination Survey


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
In population studies of weight and mortality, weight categories are usually defined using body mass index (BMI; weight (kg)/height (m)2) (1). In many studies, a plot of mortality relative risk against BMI is U- or J-shaped, with a nadir around a BMI of 25 (2–6). Mortality risk, then, tends to increase not only as BMI increases above 25 but also as BMI decreases below 25. Thus, the mortality risk in the overweight category (BMI 25–<30) may be similar to or even below that in the normal weight category (BMI 18.5–<25).

It is sometimes suggested that this observed U- or J-shaped weight-mortality association may be an artifact due in part to the operation of "reverse causation." Reverse causation in general refers to a situation where the outcome affects the exposure, rather than the exposure's affecting the outcome. For example, in the epidemiologic literature, reverse causation has been invoked as a possible explanation for a number of apparent effects: effects of poverty on health (poor health may instead lead to poverty) (7), antibiotics on asthma (asthmatics may instead be more prone to infections) (8), coffee drinking on fetal death (unrecognized early fetal death may instead be associated with higher coffee consumption) (9), Parkinson's disease on smoking (Parkinson's disease may instead reduce the likelihood of smoking) (10), and exhaustion on heart disease (unrecognized heart disease may instead lead to exhaustion) (11).

In the case of weight and mortality, reverse causation cannot strictly operate, as mortality does not affect prior weight. In this context, "reverse causation" is often used to refer to the effect of illness on both weight and mortality. It is hypothesized that preexisting illness at baseline, possibly undetected, is associated both with prior weight loss and with increased probability of death. This might more properly be considered an issue of confounding or effect modification by preexisting illness or by illness-induced weight loss rather than true reverse causation. According to Manson et al. (12, p. 355), "Clinical or subclinical illness can reduce body weight, thereby attenuating the weight-mortality association... [T]he weight loss associated with illness will lead to an underestimate of the weight-mortality slope throughout the weight range." According to Willett et al. (13, p. 428), "Reverse causation is the most serious problem associated with using total mortality as an outcome; people frequently lose weight as a result of an illness that is ultimately fatal, a situation that creates the appearance of higher mortality among those with lower weights."

Smoking is another factor associated with both lower weight and increased probability of death. According to Willett et al. (13, p. 428), "A second major concern is that confounding factors may distort the association between body weight and mortality. Smoking is particularly important, because smokers tend to weigh less and to have much higher mortality rates than nonsmokers."

Controlling for confounding by smoking and preexisting illness through exclusion rather than statistical adjustment is sometimes recommended. For example, according to Stampfer (14, p. 476), "The best way to assess the impact of overweight on risk of mortality is simply to exclude current and past smokers." This exclusionary approach has been applied in a number of large prospective studies (3–5). The exclusionary approach is designed to adjust the weight-mortality relative risks for confounding by factors such as smoking, but it does not supply information on the effects of the confounding factors themselves on mortality. Because of the lack of information on confounding factors, the exclusionary approach is not appropriate for estimating attributable fractions for the entire population; it is appropriate only within the subgroup of the population that remains after the exclusions.

Previous research using national survey data with measured heights and weights found that overweight was associated with slightly decreased mortality relative to the normal weight category and that underweight (BMI <18.5) was associated with slightly increased mortality (15). Although those results were adjusted for smoking status, there has been some concern that bias due to confounding or effect modification by smoking and preexisting illness might have affected the results (16, 17). Some previous results have already been reported to address these issues (15, 18). The purpose of this study was to apply the exclusionary approach to national survey data in order to investigate the potential effects of confounding or effect modification by baseline illness and smoking on the estimates of attributable fractions of deaths associated with different BMI levels.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
All data in this report came from the National Health and Nutrition Examination Survey (NHANES), a series of surveys conducted by the National Center for Health Statistics. In each survey, a different nationally representative cross-sectional sample of the US population was interviewed and examined. To estimate relative risks, we used baseline data from NHANES I (1971–1975), NHANES II (1976–1980), and NHANES III (1988–1994) and subsequent mortality data running through 1992 for NHANES I and NHANES II and through 2000 for NHANES III (18–25). In each survey, height and weight were measured using standardized procedures. BMI was calculated as weight (kg) divided by height (m) squared.

We calculated relative risks (hazard ratios) using Cox proportional hazards models with age as the time scale (26). Because the proportional hazards assumption was not met across the range of age, we divided the age data into three strata (25–<60, 60–<70, and ≥70 years) and fitted models separately within each age stratum. Because age was the time scale, these age strata refer to attained age rather than age at baseline. For analysis, we grouped BMI as follows: <18.5 (underweight), 18.5–<25 (normal weight; reference category), 25–<30 (overweight), 30–<35 (grade 1 obesity), and ≥35 (grade 2–3 obesity). The model included BMI categories, sex, smoking status (never, former, or current smoker), race/ethnicity (White, Black, or other), and alcohol consumption categories in g/day (none, <0.07, 0.07–<0.35, or ≥0.35).

Attributable fractions were calculated as previously described by applying the relative risks for all covariates from these models to the NHANES 1999–2002 data (15). Within each age stratum, we first calculated the relative risks from a data set with combined data from NHANES I, NHANES II, and NHANES III. We then applied each set of relative risks in turn to the current distribution of the covariates (BMI category, sex, smoking status, race/ethnicity, and alcohol consumption) in the general population, which was estimated from the NHANES 1999–2002 cross-sectional survey data. Within each age group, we calculated the relative risk ri corresponding to each combination, i, of BMI level and the levels of the other covariates. The relative risk ri for a given combination was calculated as the product of the adjusted relative risks for BMI and the other covariates in that combination. From the NHANES 1999–2002 cross-sectional survey data, we estimated the corresponding prevalence of the risk factor combination, pi. The mortality rate for a given age group was calculated as R = I {sum} ri pi, where I was the mortality rate for persons who were at the reference level for BMI and all other covariates and the sum was the sum over all i risk factor combinations. We calculated ri* as the "counterfactual" relative risk, in which BMI is set to the reference level but all other risk factors for each participant are left unchanged. The hypothetical counterfactual mortality rate from moving all participants to the reference weight category was R* = I {sum}ri*pi. The proportion of deaths attributable to nonreference weight categories was calculated as (R – R*)/R. Because the factor I cancels out, the attributable fraction depends only on the relative risks and prevalences of the covariates. This approach accounts for confounding by all covariates in the model.

The estimated number of excess deaths associated with a given BMI level and age group was then calculated by multiplying the total number of deaths for that age group in the year 2000 by the attributable fraction for that BMI level. The attributable fraction for the whole sample was calculated by summing estimated excess deaths across age groups and dividing by the total number of deaths for all age groups.

We then applied to the NHANES I, NHANES II, and NHANES III data a set of sequential exclusions taken from published recommendations to exclude people with conditions (such as preexisting illness or smoking) that could lead to both weight loss and increased mortality risk. First, participants with cancer or cardiovascular disease at baseline were excluded. Second, participants who had ever smoked (both current and former smokers) were also excluded. Third, participants who died within the first 3 years of follow-up were excluded. In addition, because of concerns about weight measured at older ages, participants whose weight and height measurements were taken at age 70 years or above were excluded. These exclusions were applied sequentially; therefore, the group of participants measured before age 70 years consisted of never smokers who were free of cancer or cardiovascular disease at baseline, who had not died in the first 3 years of follow-up, and who had had their weight and height measured before age 70 years.

We calculated relative risks for each BMI category using the full sample of combined NHANES I, II, and III follow-up data and then after applying the exclusions. However, our method of estimating the attributable fraction of deaths also uses relative risks for the other factors in the model and applies all relative risks to the current joint distribution of BMI and the other factors. To evaluate the impact of exclusions, we maintained the original relative risks associated with the other factors in the model (sex, smoking, racial/ethnic group, and alcohol consumption) but substituted the relative risks for BMI categories from the sample with exclusions instead of the relative risks for BMI categories calculated from the full sample. This allows the use of relative risks for smoking even when smokers are excluded from the sample used to estimate the relative risks for BMI categories. This set of relative risks was then applied to the joint distribution of the covariates in the full NHANES 1999–2000 sample.

Using BMI relative risks based on exclusions provides a sensitivity analysis with which to isolate the potential impact of the exclusions on the attributable fractions of deaths, independently of any changes in the relative risks for the other covariates. (We do not recommend this procedure for estimating attributable fractions or excess deaths in the entire population, however, because it does not take into account empirical data from the excluded portions of the population.)

As the various exclusions are made, the balance among the three surveys providing follow-up data can also change (i.e., a given exclusion may result in deleting a higher proportion of deaths in one survey than in another). In addition, the available weight loss variables differed across surveys. Thus, further examination of these issues was carried out within each individual survey.

To maintain adequate sample sizes for the within-survey analyses, we used two age groups (<70 years and ≥70 years) and the following BMI categories: <22, 22–<25 (reference category), 25–<30, and ≥30. We carried out analyses similar to those above, with the additional exclusions in NHANES II and NHANES III of participants who had lost weight. In NHANES II, participants who responded affirmatively to a question on whether they had lost weight in the past 6 months without intending to were excluded. In NHANES III, participants whose self-reported weight 10 years previous to baseline was more than 2.5 kg higher than their self-reported weight at baseline were excluded.

Data were analyzed using the SAS for Windows (release 9.1; SAS Institute Inc., Cary, North Carolina) and SUDAAN (release 9.0; RTI International, Research Triangle Park, North Carolina) software programs. All analyses included sample weights that accounted for the unequal probabilities of selection due to oversampling and nonresponse. All variance calculations incorporated the sample weights and accounted for the complex sample design.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
The number of deaths and the unweighted sample size for each exclusion criterion are shown in table 1. Note that these sample sizes refer to the number of participants with data for each age group and that participants originally examined in one age group who survived to enter another age group would have been counted in both age groups. However, deaths were counted only once, in the age group in which the death occurred.


View this table:
[in this window]
[in a new window]

  		TABLE 1. Numbers of deaths and relative risks for mortality according to age group, body mass index category, and sequentially applied exclusion criteria in NHANES* I (1971–1975), NHANES II (1976–1980), and NHANES III (1988–1994) combined

 
Cumulatively, these exclusions resulted in the deletion of large numbers of participants and a high proportion of deaths. After all exclusions, only 20 percent (198/976) of the deaths in the age group 25–<60 years, 14 percent (200/1,406) of the deaths in the age group 60–<70 years, and 13 percent (782/6,036) of the deaths in the age group ≥70 years remained. With this sequence of exclusions, the largest proportionate reduction in sample size was seen from the deletion of ever smokers. The effects of exclusions on unweighted sample sizes, numbers of deaths, and person-years are shown in table 2 by BMI category.


View this table:
[in this window]
[in a new window]

  		TABLE 2. Numbers of deaths, unweighted sample sizes, and person-years of follow-up according to body mass index category and sequentially applied exclusion criteria in NHANES* I (1971–1975), NHANES II (1976–1980), and NHANES III (1988–1994) combined

 
The relative risks are also shown in table 1 for the full sample and after each exclusion. In the full sample, relative risks were generally near 1.0 for all BMI categories, with the highest relative risks being found either in the underweight category (in two age groups) or in the most obese category (in one age group). Exclusion of participants with cancer or cardiovascular disease at baseline had little effect on the relative risks. For the overweight category (BMI 25–<30), relative risks were always less than 1 and remained so after each exclusion. For the underweight category (BMI <18.5), grade 1 obesity (BMI 30–<35), and higher levels of obesity (BMI ≥35), relative risks did not change uniformly across age groups. In most cases, differences between relative risks for the full sample and relative risks calculated after exclusions were small.

Results from sensitivity analyses summarizing the effect of exclusions on the estimated attributable fractions of deaths are shown in table 3. These are synthetic estimates, arrived at by using the postexclusion relative risks for BMI categories in place of the BMI relative risks from the full sample, holding all of the relative risks for other covariates constant, and applying these results to the NHANES 1999–2002 data. When the postexclusion BMI relative risks were used, the estimated attributable fractions of deaths associated with underweight and with higher levels of obesity increased and the estimated attributable fractions of deaths associated with overweight and with grade 1 obesity decreased.


View this table:
[in this window]
[in a new window]

  		TABLE 3. Sensitivity analysis of the effect of sequential exclusions on estimates of the attributable fraction (%) of deaths in 2000, according to body mass index category, in the National Health and Nutrition Examination Surveys

 
Survey-specific relative risks obtained using different BMI categories are shown in table 4. The effect of exclusions was variable within the six survey-age group categories and did not show systematic effects.


View this table:
[in this window]
[in a new window]

  		TABLE 4. Relative risks for mortality according to survey, age group, body mass index category, and sequentially applied exclusion criteria in the National Health and Nutrition Examination Surveys

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
It has been suggested that to estimate relative risks correctly in studies of weight and mortality, participants who have ever smoked or who may have lost weight prior to baseline due to preexisting illness should be excluded from analysis, because smoking and illness-induced weight loss may both modify the weight-mortality relation. However, in these NHANES data, the use of a set of simultaneous exclusions related to smoking and baseline illness in most cases had little effect on estimated relative risks or attributable fractions for BMI categories, and the effects of exclusions often pointed in the direction opposite that predicted.

These exclusions resulted in the elimination of a large proportion of the sample and consequently a large proportion of deaths in the sample. After all exclusions, only 17 percent of the deaths occurring among younger people and only 13 percent of the deaths occurring among older people remained. This is typical of the effects of such exclusions. For example, in the Nurses' Health Study cohort, Manson et al. (4) began with 4,726 deaths among women with no history of cancer or cardiovascular disease, but after exclusions for smoking, early deaths, and weight stability, their final analyses included only 531 deaths (11 percent of the total). Similarly, in their final analyses, Lee et al. (5) included only 510 (12 percent) of the 4,370 deaths occurring in the Harvard Alumni Health Study among men who were free of heart disease, cancer, and stroke at baseline. In Cancer Prevention Study II, a large study that contained over 1 million participants, 24 percent of the deaths among men and 26 percent of deaths among women occurred among nonsmokers without a history of disease at baseline (3).

The exclusion of such a large proportion of the sample leads to some loss of power to detect statistically significant associations. Typically, the relative risks for mortality associated with weight categories are low, being generally below 2 and often well below 1.5. These relatively weak associations make it difficult to detect statistically significant effects when the sample size and numbers of deaths are reduced. Thus, the individual relative risks are often not statistically significant, and only an overall test for trend shows significance. The reduced sample size and lower power would make it more difficult to detect a nonlinear relation if one existed. Some research has suggested that the effect seen from deleting smokers, for example, is similar to that seen from simply randomly deleting the same number of subjects (27). The estimated attributable fractions will have increased variability in studies with many exclusions, but attributable fraction estimates depend on the point estimates of relative risks, not on the variances of the relative risk estimates. If the point estimate of relative risk is lower, the estimated attributable fraction of deaths will be lower.

In Cancer Prevention Study II, even after exclusions the plot of mortality relative risk against BMI was U-shaped, and through much of the range of BMI there was little difference between the various subgroups of smokers and nonsmokers with or without a history of illness at baseline (3). A large cohort study in China with measured heights and weights similarly showed little effect of exclusions based on smoking status, alcohol consumption, and baseline illness (28).

Exclusions for early mortality, for history of specific diseases, and for weight loss before or shortly after baseline have been proposed as methods of reducing potential bias due to illness-induced weight loss prior to baseline (12, 13, 16). These exclusions, however, provide only indirect evidence regarding the issue of whether illness-induced weight loss has occurred before baseline and whether such weight loss has biased the results. Depending on the characteristics of the subgroup, confounding by other variables might even be increased after such exclusions. In studies with self-reported weights and heights, differences in reporting error patterns between the full sample and the subgroup could also potentially affect the results.

Estimates of relative risk can be translated into corresponding estimates of attributable fractions of deaths and excess deaths in a particular population, if the relative risks apply to that population. The various exclusions can be regarded as sensitivity analyses with respect to relative risk, but it is not clear to what population these various relative risks would apply. These sensitivity analyses offer no indication that the previous results for excess deaths were substantially biased by these factors. The relative risks for BMI categories did not show large or systematic changes after exclusion of ever smokers, persons with a history of cancer or cardiovascular disease, and persons who died early in the follow-up period or had their heights and weights measured at older ages.

The variances and confidence intervals for estimates of excess deaths are calculated by taking into account the variability in the estimated relative risks and in the estimated joint distribution of BMI and other covariates in the model. These estimates are calculated using data across all age groups. Thus, even if the relative risks within smaller age groups are not statistically significant, the overall estimate of excess deaths can still be statistically significantly different from zero. It could be inaccurate, therefore, to conclude from the confidence intervals for the relative risks that an overall estimate of excess deaths would not be statistically significant.

The relative risks for overweight and obesity in table 3 are somewhat larger than those in table 1, because we used a BMI of 22–<25 as the reference category in table 3 in order to have sufficient numbers of deaths in each BMI category to examine the effects of exclusions. This reference category has lower mortality rates than the widely used normal weight reference category, BMI 18.5–<25. Additionally, the estimates shown in table 3 incorporate weight loss history and indicate that within a given survey there were no strong or systematic effects of exclusions.

Members of a cohort initially categorized by BMI level, smoking, and health status may change their characteristics over time, leading to complex effects on mortality rates. In an observational study, it is unlikely that any simple analysis, such as regression adjustment for baseline covariates or exclusions based on initial smoking or health status, can accurately and completely characterize the association between baseline levels of BMI and mortality. In an experimental study, it is difficult to define the causal effects of weight on mortality separately from the effects of the specific interventions employed in the study. Nonetheless, measures such as the attributable fraction of deaths or the numbers of excess deaths associated with BMI levels provide useful summaries of the association between BMI and mortality in the population.

These sensitivity analyses suggested that residual confounding by smoking or preexisting illness had little effect on previous estimates of attributable fractions from nationally representative data with measured heights and weights.


   ACKNOWLEDGMENTS
 
Conflict of interest: none declared.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. NHLBI Expert Panel on the Identification, Evaluation and Treatment of Overweight and Obesity in Adults. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults. Obes Res 1998;6(suppl 2):51S–209S. (Also available at www.nhlbi.nih.gov/guidelines/obesity/ob_home.htm). (Accessed June 7, 2006).
  2. Troiano RP, Frongillo EA Jr, Sobal J, et al. The relationship between body weight and mortality: a quantitative analysis of combined information from existing studies. Int J Obes Relat Metab Disord (1996) 20:63–75.[Web of Science][Medline]
  3. Calle EE, Thun MJ, Petrelli JM, et al. Body-mass index and mortality in a prospective cohort of U.S. adults. N Engl J Med (1999) 341:1097–105.[Abstract/Free Full Text]
  4. Manson JE, Willett WC, Stampfer MJ, et al. Body weight and mortality among women. N Engl J Med (1995) 333:677–85.[Abstract/Free Full Text]
  5. Lee IM, Manson JE, Hennekens CH, et al. Body weight and mortality. A 27-year follow-up of middle-aged men. JAMA (1993) 270:2823–8.[Abstract/Free Full Text]
  6. McGee DL; Diverse Populations Collaboration. Body mass index and mortality: a meta-analysis based on person-level data from twenty-six observational studies. Ann Epidemiol (2005) 15:87–97.[CrossRef][Web of Science][Medline]
  7. Rognerud MA, Zahl PH. Social inequalities in mortality: changes in the relative importance of income, education and household size over a 27-year period. Eur J Public Health (2006) 16:62–8.[Abstract/Free Full Text]
  8. Pekkanen J. Commentary: use of antibiotics and risk of asthma. Int J Epidemiol (2004) 33:564–5.[Free Full Text]
  9. Bech BH, Nohr EA, Vaeth M, et al. Coffee and fetal death: a cohort study with prospective data. Am J Epidemiol (2005) 162:983–90.[Abstract/Free Full Text]
  10. Allam MF, Campbell MJ, Del Castillo AS, et al. Parkinson's disease protects against smoking? Behav Neurol (2004) 15:65–71.[Web of Science][Medline]
  11. Macleod J, Carroll D. Commentary: what should we make of associations between vital exhaustion and heart disease? Int J Epidemiol (2003) 32:997–9.[Free Full Text]
  12. Manson JE, Stampfer MJ, Hennekens CH, et al. Body weight and longevity. A reassessment. JAMA (1987) 257:353–8.[Abstract/Free Full Text]
  13. Willett WC, Dietz WH, Colditz GA. Guidelines for healthy weight. N Engl J Med (1999) 341:427–34.[Free Full Text]
  14. Stampfer M. Weight loss and mortality: what does the evidence show? PLoS Med (2005) 2:e181. (Electronic article).[CrossRef][Medline]
  15. Flegal KM, Graubard BI, Williamson DF, et al. Excess deaths associated with underweight, overweight, and obesity. JAMA (2005) 293:1861–7.[Abstract/Free Full Text]
  16. Willett WC, Hu FB, Colditz GA, et al. Underweight, overweight, obesity, and excess deaths. (Letter). JAMA (2005) 294:551.[Free Full Text]
  17. Hu FB. Overweight and increased cardiovascular mortality: no French paradox. Hypertension (2005) 46:645–6.[Free Full Text]
  18. Flegal KM. Supplemental analyses for estimates of excess deaths associated with underweight, overweight, and obesity in the U.S. population. (2007) Hyattsville, MD: National Center for Health Statistics. (http://www.cdc.gov/nchs/products/pubs/pubd/hestats/excess_deaths/excess_deaths.htm). (Accessed June 1, 2006).
  19. Miller HW. Plan and operation of the Health and Nutrition Examination Survey. United States— 1971 –1973. Vital Health Stat 1 (1973) (10a):1–46.
  20. Engel A, Murphy RS, Maurer K, et al. Plan and operation of the HANES I augmentation survey of adults 25–74 years United States, 1974 –1975. Vital Health Stat 1 (1978) 1–110.
  21. McDowell A, Engel A, Massey JT, et al. Plan and operation of the Second National Health and Nutrition Examination Survey, 1976 –1980. Vital Health Stat 1 (1981) (15):1–144.
  22. Plan and operation of the Third National Health and Nutrition Examination Survey, –94. Series 1: programs and collection procedures. Vital Health Stat 1 (1988) 1994:1–407.
  23. Cohen BB, Barbano HE, Cox CS, et al. Plan and operation of the NHANES I Epidemiologic Followup Study: 1982 –84. Vital Health Stat 1 (1987) 1–142.
  24. Cox CS, Mussolino ME, Rothwell ST, et al. Plan and operation of the NHANES I Epidemiologic Followup Study, 1992. Vital Health Stat 1 (1997) 1–231.
  25. Loria CM, Sempos CT, Vuong C. Plan and operation of the NHANES II Mortality Study, 1992. Vital Health Stat 1 (1999) 1–16.
  26. Korn EL, Graubard BI, Midthune D. Time-to-event analysis of longitudinal follow-up of a survey: choice of the time-scale. Am J Epidemiol (1997) 145:72–80.[Abstract/Free Full Text]
  27. Effect of smoking on the body mass index-mortality relation: empirical evidence from 15 studies. BMI in Diverse Populations Collaborative Group. Am J Epidemiol (1999) 150:1297–308.[Abstract/Free Full Text]
  28. Gu D, He J, Duan X, et al. Body weight and mortality among men and women in China. JAMA (2006) 295:776–83.[Abstract/Free Full Text]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?



This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
166/8/975    most recent
kwm152v2
kwm152v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (5)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Flegal, K. M.
Right arrow Articles by Gail, M. H.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Flegal, K. M.
Right arrow Articles by Gail, M. H.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?