Modern human beings are fond of adding weight to their ugly bodies to such an extent that many have to turn aside to get into their bedrooms. Other than genes, which approximately accounts for 30~ 60% of obesity and which we can do nothing about, poor diet and sedentary lifestyle are likely the culprits. To prevent obesity and associated consequences, we should address our modifiable lifestyles.

Recently, the Center for Disease Control and Prevention (CDC) bravely addressed this question: “how many deaths are attributed to our lifestyles?” The paper published in JAMA in March, 2004 drew a great deal of interests from many groups including the US Department of Health and the Congress. Unfortunately, the attributed death number due to poor diet and physical inactivity was wrong. This mistake provoked widespread discussions in both mass media and academia.

People often publish wrong results due to typos, miscalculations, or even wrong models. Randomly opening a scientific journal, you will see a section called “errata” at the end of the journal in almost all issues. CDC also made mistakes in its publications before. Why did things get so “violent” this time?

The main reason is that this paper was supposed to be very important in public health. It was written by a government agency, and the authors were high rank directors in the CDC. Future health policies and funding opportunities would be significantly affected by papers like this one. For example, just right after the appearance of this paper, the Robert Wood Johnson Foundation issued memos about cutting back tobacco funding while increasing obesity related intervention funding. The Secretary of Health and Human Services Tommy Thompson also cited this paper in his public speech. The importance of this paper is enormous.

So what was the mistake the authors had made in that paper? Well, the mistake is not easily detectable if you don’t read it carefully. I will try my best to explain their methods at the end of this essay. But if you are not interested in the methodological issues, here is a brief description of key problems in that paper.

In that paper, the authors used different formula to compute attributed deaths for tobacco and for poor diet/physical inactivity. Furthermore, they didn’t account for sex, age, smoking, and other factors when computing the attributed deaths for poor diet/physical activity. Some assumptions were unfounded, but they used them liberally.

These and other neglects were instantly caught by some keen readers. In fact, the accompanied editorial also noticed the methodological inconsistency in that paper but thought it was OK. Among many letters to editor, two letters correctly pointed out the methodological flaws which I also detailed at the end of this essay. The healthy scientific skeptics, however, lead to something bigger than science.

People working on tobacco research first opened fire on this paper. They not only questioned the scientific merits of this paper but also suspected conspiracies behind the paper. They argued that it was unbelievable that this paper was passed through the internal review in CDC. Furthermore, within the CDC, there were some debates on related methodological issues. Two papers from the author’s division suggested that traditional methods (the same method used in the JAMA paper) might overestimate attributable deaths.

Quickly, mass media heated the debate. By the end of November last year, all major newspapers reported this incidence. For example, In a Wall Street Journal interview with Dr. Julie Gerberding, the director of Chronic National Center for Chronic Disease Prevention and Health Promotion, who was also one of the authors in the JAMA paper, Dr. Gerberding acknowledged the problem and stated that they had initiated internal investigation right after they received concerns from readers. She assured the reporter that there was no misconduct in the paper and claimed that scientific integrity was important in research. However, her comment– “This paper in and of itself is a very minor contributor to our knowledge of obesity”, is a sure sign of discrediting that important paper.

This week, the authors finally gave a letter of correction in JAMA stating that due to errors in computation, “(formula) modifications were not copied properly from the first to the next 5 spreadsheet tables.” The final attributed deaths for poor diet and physical inactivity were 365,000, about 35,000 less deaths than that in the original report.

The corrected number is not important now because nobody will cite that paper any more. Damages were done. There will sure be more internal debate within CDC and academia. However, the scientific integrity and political involvement are again in question. Let’s wait and see.

_______________________________________________
Appendix: methodological issues in CDC JAMA paper.
_______________________________________________

To compute the deaths attributed to risk factors such as tobacco and poor diet, you first need attributable fraction (or attributable risk ratio) for any specific risk factor. In that paper, the attributable ratios were obtained from several studies, and then the ratios were weighted by the percent of people engaging that risky behavior to get an overall attributable fraction. Because many risk behaviors have several levels, e.g., never smoked, former smoker, or occasional smokers, or heavier smokers, the weighting strategy is reasonable. The attributable fraction was then applied to the vital statistics–which is conveniently available in CDC, to obtain the attributed deaths. However, the key problem here is where you get the original attributable ratios. Another problem is whether the attributable ratio is adjusted (or controlled) for other factors such as sex and age.

The dispute was centered on the different methods used in computing tobacco attributed deaths and diet and physical activity attributed deaths. In computing tobacco related deaths, the authors used well established attributable ratios from other studies, and the tobacco use prevalences from Behavioral Risk Factor Surveillance Study (BRFSS). The computation was stratified by sex and age.

In computing deaths related to poor diet and physical inactivity, the author took an indirect approach. They used attributed ratios for obesity (and also data) from a study done by Allison DB et al., in which raw data from six studies were pooled. They argued that poor diet and physical inactivity accounts most of the hazardous effects of obesity, which is probably true. However, the attributable ratios were computed differently from those for tobacco. The obesity attributable ratios were marginal ratios adjusted for sex and age, and the authors used these ratios directly on the whole population instead of stratifying by sex and age, as what they did in the tobacco computation. People argued that obesity is less harzardous with the increase of age. The way the authors did was incorrect.

Furthermore, the authors thought (correctly) that poor diet and physical inactivity had other detrimental effects on health besides obesity. For example, poor diet is related to high cholesterol, and physical inactivity may be related to worse cardiovascular fitness. Therefore, the authors added 15,000 deaths in additional to those from obesity alone. However, they didn’t give any references on how they got this magic number.

My impression is that although the authors may overestimate the obesity related deaths, they may actually underestimate deaths from poor diet and physical inactivity because poor diet and physical inactivity can directly and significantly increase the risk of cardiavascular diseases and cancer, which are top 3 causes of death. In a previous report done by other investigators which answered similar questions, they directly examined the poor diet and physical inactivity. Their method was valid.