Issues related to modeling the body mass index-mortality association: the shape of the association and the effects of smoking status.

Research on the relationship between body mass index (BMI) and mortality has led to conflicting results; a lack of agreement about how to adjust for confounders, such as smoking status, has added to the problem. Complicating such analyses is the fact that the BMI-mortality association is not a symmetric quadratic relationship; the distribution tends to be skewed to the right, causing the optimal BMI--where mortality is at a minimum--to be overestimated. One way to overcome this problem is by transformation of the BMI distribution to normality. The authors suggest several approaches for doing so, including the use of 1/BMI, or lean body mass index, instead of BMI in modeling. Data sets on 50 cohorts from approximately 30 international studies were used to examine the association (direct, inverse, quadratic or none) between BMI and mortality and to investigate the possible interaction of smoking status. Of the 50 cohorts, 36 showed a quadratic association between BMI and mortality, 10 showed no association and 1 showed a direct association between lean BMI and mortality. Only three cohorts showed a significant interaction between BMI and smoking, which was approximately what one would expect from a 5% significance test, even if no interaction existed. The association between BMI and mortality is not changed when smoking status is ignored in a model or when data on smokers are excluded from analysis. The methodology used in this study could be extended to look for other interactions.