Confidence Interval
Definition: A confidence interval is a range of plausible values for the measure of association. (Methods in Observational Epidemiology, Kelsey et al, 1996)
Confidence intervals allow the reader to see the range of values consistent with the reported results and compare that range with other tests.
Example: Consider an existing diagnostic test, tested in hundreds of patients. The test is 80% sensitive and 70% specific. A new diagnostic test is reported to be positive in 5 out of 5 patients with the disease (100% sensitive), and negative in 9 out of 10 without disease (specificity of 90%). Although impressive, these results are based on just a few subjects. In fact, the 95% confidence interval for the new test's sensitivity runs from about 57% to 100%; for the specificity, it runs from 60% to 98%. Both of these intervals overlap the values for the old test. Thus it is not clear if the new test is an improvement. (Designing Clinical Research, Hulley and Cummings, 1988)
