Understanding and using confidence intervals in clinical research.

Abstract

Confidence intervals contain a wealth of clinically relevant information that is not available in the P value and usual significance testing. Numerous articles discuss the hazards of interpreting study results based solely on the P value, raising both practical and philosophical concerns. The general recommendation is that clinical research should not just test hypotheses, but also describe magnitudes of clinical effect. To this end, the confidence interval is a crucial tool in interpreting clinical studies. In this report, we show how one may use confidence intervals to gain further insight into clinical research. For example, by using confidence intervals, one can identify statistically significant results that are clinically imprecise, or conversely, statistically nonsignificant results that are quite precise. In addition, confidence intervals, like the P value, are influenced by sample size. We show how sample sizes that are sufficiently large to test hypotheses may be too small to generate precise estimates of the magnitude of effect. The application and interpretation of confidence intervals are demonstrated through the use of several examples.