A “Diagnosis Method" for the Analysis of Epidemiological Studies. A Reevaluation of Pan, et al. Arch Intern Med 2012, 172:555.

Abstract

The medical literature has been subject to extensive criticism for a lack of accuracy and reproducibility. A major cause, in our view, is misleading or inappropriate presentation and interpretation of statistical results. Emphasis on percentages or relative values (odds ratios, relative risk, and hazard ratio) always reduces and often obscures information. We suggest a more direct method using the 2 x 2 matrix commonly employed in diagnosis testing. We demonstrate the “diagnosis method,” by evaluating a previously published paper (Pan A, et al.: Arch Intern Med 2012, 172:555-563). on the risk of red meat for disease, focussing on cancer risk. The direct analysis shows very small effects of red meat. Given the generally large errors in the independent variable measured by food frequency questionnaire, even this low predictability is likely to be unreliable. In combination with other studies in the literature, the analysis suggests that red meat is unlikely to be a risk for cancer. The major point is that statistical significance does not necessarily point to biological or clinical importance.