Confidence intervals in medical research.

ABSTRACT

The utility of confidence intervals in a wide variety of situations in the medical field is re-emphasized, with examples drawn from controlled clinical trials, disease control programmes, vaccine trials and laboratory studies. It is shown that the confidence interval approach is more informative than a mere test of statistical significance, and should therefore be employed as an useful adjuvant. Since proportions are widely quoted in medical literature and as the determination of the exact confidence limits for a binomial proportion is iterative and time-consuming, an assessment is made of 15 published methods which provide approximate confidence limits; the 'Square root transformation' method is recommended since it is accurate and the computation of limits is relatively easy. In the case of a difference between two proportions, the usual method may be employed if sample sizes exceed 75; for smaller sample sizes (even for sizes of 5), the Jeffreys-Perks method is very satisfactory and is therefore recommended.