Do We Need “Big” Or Rather Consecutive Studies In Epidemiology?


Authored by  G Kallischnigg, E Kaelin, HJ Urban, R Weitkunat

Presented at Society for Epidemiologic Research Meeting (SER) 2007     

Abstract

Many investigators have suggested that very large studies are needed in epidemiology. Large studies, which are often organized as multi-center or even multi-national studies, are indeed essential if effects are small, as they often are in molecular and genetic epidemiological research. As with diagnostic tests, however, the proportion of true positive findings obtained in such studies depends to a large degree on the prior probability of the effect as well as on the sensitivity and specificity of the measurement. In statistical terms, the latter two correspond to power and type I error, respectively. In analogy to the calculation of the positive predictive value (PPV), a parameter often used to assess the results of a given study, the true positive report probability (TPR) can be calculated. The TPR is the probability of the alternative hypothesis being true, given that a significant finding has been obtained. It can be shown that the TPR does not increase substantially by increasing the sample size beyond 1000 individuals, if the prior probability of the true alternative hypothesis is low. However, as in diagnostic tests, the TPR will increase dramatically if the finding is replicated. From this it follows that, independent of sample size considerations, it is necessary to conceive research strategies which pre-plan for replications in order to avoid false positive results.