Confidence interval

A confidence interval (CI) gives, around an effect estimate, the range of plausible values within which the true effect may lie. According to the Cochrane Handbook, the confidence interval conveys the precision (uncertainty) of the summary estimate, whereas the p value conveys the strength of evidence against the null hypothesis. ^[1] The 95% confidence interval is given most often.

How is it read?

Two things are read together: the width of the interval and whether the interval includes the “no effect” value. For ratio measures (risk ratio, odds ratio) the “no effect” value is 1, and for difference measures (mean difference, risk difference) it is 0. A narrow interval means a precise estimate, and a wide interval an uncertain one. Rather than calling results “statistically significant” or “non-significant” and relying too heavily on p value thresholds, the Cochrane Handbook recommends that the confidence interval be reported together with the exact p value; it describes the 0.05 threshold as “arbitrary.” ^[2]

An example

In Cochrane’s sponsorship review, for a favorable efficacy result in industry-sponsored studies the RR was 1.27 (95% CI 1.17–1.37; 25 papers); the interval lies entirely above 1. In the same review, for harm outcomes the interval of RR 1.37 (95% CI 0.64–2.93; four papers) includes 1 and is very wide; the estimate is not precise. ^[3]

Point to note

A confidence interval reflects only uncertainty due to chance; it does not show design flaws of the study (bias) or whether the result can be applied to other groups of people. These are a separate topic of levels of evidence. For the order of reading and examples, see the page How to read a study?, and for closely related terms, the entry Absolute and relative risk.