American Journal of Epidemiology Vol. 154, No. 7 : 685-686
Copyright © 2001 by The Johns Hopkins University School of Hygiene and Public Health
LETTERS TO THE EDITOR |
RE: "CHANGES IN BODY MASS INDEX PRIOR TO BASELINE AMONG PARTICIPANTS WHO ARE ILL OR WHO DIE DURING THE EARLY YEARS OF FOLLOW-UP"
Department of Health and Nutrition Sciences Brooklyn College of the City University of New York Brooklyn, NY 11210
The article by Stevens et al. (1
) sheds fresh light on the confounding of the relation between mortality risk and body mass index by preexisting illness. For example, their analysis suggests that eliminating up to, but no more than, the first 4 years of deaths is effective in removing such confounding. This is a useful guide for future investigators.
Their analysis did not, however, adequately address an important question raised by the meta-analysis of Allison et al. (2): Does eliminating subjects with serious illness and pre-baseline weight loss remove all confounders? Allison et al. (2) showed that, even after eliminating deaths during early follow-up, the relation between mortality risk and body mass index was J- or U-shaped. This suggests that not all confounders were removed, because a J- or U-shaped relation conflicts with clinical trial evidence that weight loss among nonobese subjects reduces risk for major killer diseases, such as heart attacks and strokes (3). The authors' explanation (1) for the finding by Allison et al. (2) is that the majority of the eliminated subjects do not have preexisting illness. This seems illogical, because the authors' own findings (1) show that eliminating early deaths does exclude most subjects with preexisting illness. Moreover, the rationale for eliminating early deaths is the following: 1) Preexisting illness causes a negative relation between mortality risk and body mass index at a low body mass index, and 2) eliminating early deaths removes this negative relation from the mortality risk versus body mass index relation for subjects without preexisting illness and, hence, removes the confounding (4).
An alternative interpretation of the findings by Allison et al. (2) is that there are factors other than preexisting illness that cause a negative relation between mortality risk and body mass index at a low body mass index and that are not removed by eliminating early deaths. Data from the First National Health and Nutrition Examination Survey 1 Epidemiologic Follow-up Study (NHEFS) support such an interpretation. Only 20.6 percent of the subjects in the 1982–1984 NHEFS survey over the age of 55 years who lost weight between the 1971–1975 and 1982–1984 surveys reported that a physician had told them that they had had a heart attack, stroke, or cancer. There was also evidence of a negative relation between mortality risk and body mass index at a low body mass index among all these subjects, which caused an artifactual increase in mortality risk at a low body mass index for the other subjects in survival analyses (5).
These findings suggest that eliminating all subjects 55 years or older with weight loss would do more to remove confounders from the relation between mortality risk and body mass index in the Atherosclerosis Risk in Com-munities' database than the strategy recommended by Stevens et al. (1): eliminating subjects with preexisting illness who die in the first 4 years of follow-up.
REFERENCES
-
Stevens J, Juhaeri, Cai J. Changes in body mass index prior to baseline among participants who are ill or who die during the early years of follow-up. Am J Epidemiol 2001;153:946–53.
[Abstract/Free Full Text] - Allison DB, Faith MS, Heo M, et al. Meta-analysis of the effect of excluding early deaths on the estimated relationship between body mass index and mortality. Obes Res 1999;7:342–54.[ISI][Medline]
- McCarron DA, Oparil S, Chait A, et al. Nutritional management of cardiovascular risk factors. A randomized clinical trial. Arch Intern Med 1997;157:169–77.[Abstract]
- Manson JE, Stampfer MJ, Hennekens CH, et al. Body weight and longevity: a reassessment. JAMA 1987;257:353–8.[Abstract]
- Greenberg JA. Biases in the mortality risk versus baseline body mass index relationship in the NHANES-1 Epidemi-ologic Follow-up Study. Int J Obes Relat Metab Disord 2001;25:1071–8.[ISI][Medline]
THE AUTHORS REPLY
Departments of Nutrition, Epidemiology, and Biostatistics School of Public Health University of North Carolina Chapel Hill, NC 27599
We took Dr. Greenberg's thoughtful letter (1) about our recently published article (2) as evidence that methodological issues surrounding the analysis of the body mass index-mortality association continue to intrigue researchers. Below we address his points.
Dr. Greenberg comments that our analysis did not address the question, "... Does eliminating subjects with serious illness and pre-baseline weight loss remove all confounders?" (1, p. 685). We think it would be a rare circumstance in which an investigator would ever claim that all confounders had been removed. Dr. Greenberg asserts that all confounding has not been removed if the body mass index-mortality association in a study is J- or U-shaped. As evidence for this statement he cites an example of a clinical trial (3) that has shown that "weight loss among nonobese subjects reduces risk for major killer diseases ..." (1, p. 685). We find this evidence not entirely convincing and not necessarily relevant because weight loss is a different issue from weight status. In addition, the study cited (3) actually showed reductions in risk factors for disease, not disease. The shape of the body mass index-mortality association has been a source of controversy for many years.
Dr. Greenberg questioned our comment regarding the finding by Allison et al. (4) that the elimination of deaths that occur early in follow-up resulted in only a small change in the association between body mass index and mortality. We did not say that "... the majority of the eliminated observations do not include preexisting illness" (1, p. 685). We stated, "This [small effect] is not surprising, given that the majority of the observations in an analysis are not impacted by this maneuver" (2, p. 953). In other words, in most of the studies included in the meta-analysis of Allison et al. (4), a relatively small percentage of the subjects was excluded because of early deaths. Those observations that remained after the exclusion were a subset composed of the majority of the original data. We appreciate the opportunity to clarify that point.
We agree with Dr. Greenberg's statement that there are "factors other than preexisting illness that cause a negative relation between mortality risk and body mass index at a low body mass index ..." (1, p. 685). We also agree that the elimination of participants who have lost weight would be an efficient way of controlling for some types of confounding of the body mass index-mortality association. However, we feel compelled to point out that data on weight loss are only rarely available. Valid data on the intentionality of weight loss are even more rare. In addition, the exclusion of participants who had lost weight would not adequately control for factors such as smoking.
Finally, we take issue with the assertion that we recommended "eliminating subjects with preexisting illness who die in the first 4 years of follow-up" (1, p. 685). In fact, we were far more conservative in our recommendations: "... investigators should examine their data before and after exclusion of the first 4 years of deaths to assure themselves that inclusion of those events is not biasing their results. Exclusion of more than 4 years of deaths does not appear to be useful. In addition, ... [we] recommend that deaths occurring in the first year of follow-up be routinely excluded from studies of the BMI [body mass index]-mortality association" (2, p. 953).
We continue to endorse these recommendations.
REFERENCES
- Greenberg J. Re: "Changes in body mass index prior to baseline among participants who are ill or who die during the early years of follow-up." (Letter). Am J Epidemiol 2001;154:685.
- Stevens J, Juhaeri, Cai J. Changes in body mass index prior to baseline among participants who are ill or who die during the early years of follow-up. Am J Epidemiol 2001;153:946–53.
- McCarron DA, Oparil S, Chait A, et al. Nutritional management of cardiovascular risk factors. A randomized clinical trial. Arch Intern Med 1997;157:169–77.
- Allison D, Faith M, Heo M, et al. Meta-analysis of the effect of excluding early deaths on the estimated relationship between body mass index and mortality. Obes Res 1999;7:417–19.[ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||