Putting clinical studies into better perspective by defining accuracy when effect size is small

In recent years, computational techniques have improved to where tens and even thousands of observations can be rapidly analyzed on computers. Large numbers of observations may improve statistical certainty but may increase clinical/epidemiological uncertainty when the effect size is so called "modest" or even "moderate," because, although p-values may reach significance, clinical discrimination may be poor. Moreover, when the effect size is small, studies show that the conclusions may be faulty depending on unknown confounders, sample selection, the statistical models used for calculation, and other factors. Although clinical/epidemiological results may be viewed as efficacy or effectiveness, one may ask whether or not such results can be translated into measures of accuracy. The objective here is to show how indices such as risk ratio or odds ratio can be translated into more absolute parameters that better define clinical usefulness.

In this review, techniques are discussed that may help providers place statistical results into a perspective that allows for better decision making. Examples are given and calculations are described whereby commonly provided more relative statistics, such as risk, hazard and odds ratios can easily be converted/examined in more absolute terms by converting to or calculating diagnostic statistics or number needed to treat/be exposed (NNT/NNE) that provides a better practical perspective.

As described, statistically significant studies with small effect sizes are always suspect but may be converted or viewed in parameters that allow observers to better evaluate the results in term of accuracy. This may provide a more practical perspective.