Meta-analysis of animal experiments on dietary fat intake and mammary tumours.

I describe a meta-analysis of animal experiments designed to investigate the effects of dietary fat intake upon mammary tumour development. These experiments involve one or more of three types of dietary manipulation: (i) varying the proportion of fat in the diet; (ii) varying the source of fat; and (iii) varying the amount of energy consumed. I use linear logistic regression models to investigate questions about the relative effects of fat and total energy intake upon mammary tumour development and about the effects of different sources of fat. The results of these analyses are presented and discussed. The regression models employed are characterized by the inclusion of an experiment 'effect' to ensure that estimated nutrient effects are based only upon within-experiment comparisons. Homogeneity of nutrient effects across experiments is assumed. The method differs from the usual meta-analysis of clinical trials in that the nutrient effects may not be estimable from each individual experiment, but only from combinations of experiments. This poses problems for the development of a model that allows heterogeneity of nutrient effects.