Author: David F. Williamson
Position: Epidemiologist
Institution: Division of Diabetes Translation, CDC, Atlanta, GA. USA
E-mail: drw1@cdc.gov
Additional Authors: none
Submitted Date: July 13, 2005
Published Date: July 20, 2005
This comment was originally posted as a “Reader Response” on the publication date indicated above. All Reader Responses are now available as comments.

Dear Editor: July 13, 2005

Prof. Stampfer's recent Perspectives commentary [1] on the paper by Sorensen et al.[2] appropriately acknowledges the challenges inherent in using observational epidemiology to determine the impact of weight loss on life expectancy. However, his case that the data of Sorensen et al. do not support their conclusion that intentional weight loss may be hazardous is based, in part, on erroneous statements about the Sorensen et al. study.

Stampfer suggests that 'reverse causation' could account for the findings of Sorensen et al. because he believes they did not do a 'lagged' analysis, in which deaths that occur in the first few years after follow-up are excluded. In the Statistical Analysis section of their paper, however, Sorensen et al. describe using two separate fully adjusted models: one for the first 5 years of follow-up and one for the period thereafter, and they also reported mortality hazard ratios associated with intentional weight loss during each period. Because so few deaths occurred in the first 5 years of follow-up, the estimated mortality hazard ratio (HR) for intentional weight loss during this period (6.26) had such a wide confidence interval (0.33-118) that it was essentially meaningless. However, after excluding the first 5 years of follow-up data, Sorensen et al. still found a clinically and statistically significant association between intentional weight loss and death during the remaining 13 years of follow-up: HR= 1.88 (CI 1.05-3.39).

Stampfer indicates that the authors differentiated only between current smokers and nonsmokers and thus inappropriately combined never smokers with past smokers. In the Methods section of their paper, however, Sorensen et al. reported that they originally used four categories (never smoker, occasional smoker, former regular smoker, and current smoker) to code the smoking status of study participants, before recoding smoking status as a dichotomous yes/no variable. However, as Sorensen et al. described in the Statistical Analysis section of their paper, they analyzed their models using both coding methods to determine whether recoding resulted in residual confounding. Because they found no residual confounding, they chose to report results only from the model with the simpler dichotomous coding of smoking status.

Stampfer also argues that the 'best way?' to remove residual confounding by smoking is to 'simply exclude current and past smokers'. This exclusionary approach for smoking, has been previously examined in a methodologic study that utilized statistical simulation with data from 15 diverse observational studies of body weight and mortality [3]. The study concluded that, 'eliminating smokers from the data sets prior to analysis produces results similar to those expected from the elimination of numerically similar random proportions of the data sets prior to analysis (3, p. 1297. Thus, the practice of excluding smokers in studies of weight loss and mortality is highly questionable.

Sincerely,
/s/ DFW
David F. Williamson, Ph.D.
CAPT U.S. Public Health Service
Centers for Disease Control and Prevention
Division of Diabetes Translation (K-10)
4770 Buford Hwy, N.E.
Atlanta, GA 30341-3727
(P) 770-488-1054
(F) 770-488-1148

References

1. Stampfer M (2005) Weight loss and mortality: What does the evidence show? PLoS Med 2(6);e181:476-477.
2. Sorensen TIA, Rissanen A, Korkeila M, Kaprio J (2005). Intention to lose weight, weight changes, and 18-y mortality in overweight individuals without comorbidities. PLoS Med 2(6);e171:1-11.
3. The BMI in Diverse Populations Collaborative Group (1999). Effect of Smoking on the Body Mass Index-Mortality Relation: Empirical Evidence from 15 Studies. Am J Epidemiol;150:1297-1308.