(upper – lower) < 3× (RR-1)
Understanding the confidence in epidemiology results is essential if the uncertainty in liability exposure estimates is to be usefully expressed. Uncertainty is often greater than the central exposure estimate.
Evidence from:
andrew@reliabilityoxford.co.uk
Epidemiological results are often supported by reference to P values. It has become commonplace to refer with great confidence to results with P values less than 0.05. Such confidence may be misplaced.
For example, a P value of 0.04 tells us that if the null hypothesis were true, an association as strong as the one we observe in that particular experiment would occur with a probability of only 4%. That is, it is not very likely that you would get the reported Risk Ratio, based on the data at hand, if in fact, in that particular experiment, the null hypothesis was the true one.
But this does not mean the true result would ever have been found by the experiment in question, only that in the circumstances of the experiment at hand, the null value would have been unlikely.
Ratio of upper to lower 95% Confidence Interval is a better measure of precision, but still will not tell us if the experiment was ever capable of unearthing the true strength of association.
The editors of this journal are often, more persuaded by epidemiological results when the ratio of upper to lower 95% Confidence Interval is three or less. This is an entirely subjective criterion.
In general, it is useful to plot actual data on a graph including the null hypothesis. The reader can then judge for themselves whether the excellent P value, often enthusiastically quoted by the authors, really means anything. Such graphs are rare.