How much of the data published in observational studies of the association between diet and prostate or bladder cancer is usable for meta-analysis?