Bayesian Analysis of the Effect of Intentional Weight Loss on Mortality Rate.

The effect of weight loss on mortality rate is widely studied and of importance in the field of obesity. Separating the effects of intentional weight loss (IWL) from unintentional weight loss (UWL) has been a challenge. Most studies addressing this issue have used weight loss among people intending to lose weight as a surrogate of IWL. Coffey et al. (2005) [1] showed that these were not equivalent and developed a preliminary model to separate the effects of IWL from those of UWL. In this study we construct and implement Bayesian latent-variable linear models that allow the separation of the effects of IWL and UWL. The key idea of our method is to augment the unobserved UWL by using the information of observed weight loss among individuals not intending to lose weight. This data augmentation approach offers a way to estimate the effects of IWL and UWL as well as any parameters of interest. We applied our method to a real data set of rodent caloric restriction studies: our results suggest that IWL has a beneficial effect on mouse lifespan in contrast to UWL. Extensions to human data involving censored outcomes are discussed.