Re: False-Positive Results in Cancer Epidemiology: A Plea for Epistemological Modesty

Extract

Like the false-positive results discussed by Boffetta et al. ( 1 ), false-negative results are also an inherent feature of epidemiology. To be considered as positive, epidemiological studies need to be “statistically significant,” that is, to overcome the conventional limit of the false-positive rate set at 1/20 or P less than .05. The observational nature of epidemiology, which limits the number of patients, and hence the power to attain the conventional P value, as well as other constraints (eg, financial) are the main reasons for results that are apparently negative, that is, do not show a statistically significant association. The concept of the P value has been criticized ( 2 ), but it remains in wide use, so that associations that are plausible on various grounds, for example, previous positive results or established mutagenicity, are dismissed because of a non–statistically significant P value. Other reasons for “negative” findings are exposure misclassification ( 3 ), comparison bias and dilution effect ( 4 ), and business bias ( 5 ).