Response to “Incident CHD and Excess Body Weight in the US Population”

To The ediTor: Flint et al. (1) attempted in their article titled "Excess Weight and the Risk of Incident Coronary Heart Disease Among Men and Women" to extrapolate from several nonrepresentative study samples to estimate the fraction of incident coronary heart disease (CHD) in the US population attributable to excess body weight.Without information on the number and distribution of incident CHD cases occurring in the US population, such extrapolation is a difficult challenge.The distribution of cases in Flint et al.'s highly selected, age restricted, and professionally homogeneous study samples of female nurses and male health professionals may well differ from the distribution of cases in the US population, which is more diverse in terms of age, race, geographic region, income, education, access to medical care, and multiple other factors that affect incident CHD.Flint et al. state that they used the weighted sum method of calculating attributable fractions, citing Benichou (2).However, rather than the weighted sum method as described by Benichou, Flint et al. appear to be applying some different method of calculating attributable fractions, which is incompletely described and whose derivation, properties, and validity are unclear.In the weighted sum method, attributable fractions are calculated within each of the cells formed by the joint distribution of the covariates.Then, the overall attributable fraction is calculated as the weighted sum of the attributable fractions within each cell.In calculating the weighted sum, the weight for the attributable fraction from a given cell is the proportion of the total cases in the sample that fall within that cell, not the proportion of the total sample in the cell.