Confidence interval in the analysis of clinical signicance

Abstract

In the clinical study researchers want to answer the most important question whether a new therapy is better than the old one. Many researchers continue to frame the question in terms of null hypothesis and answer the question in terms of P value. The null hypothesis itself is typically not plausible, and even in a study where the null hypothesis was plausible, our concern is typically not only limited to the issue of whether or not the treatment has any effect but we also want to know how much the impact. In the study which the researcher pressed the P value into service as an indicator of effect size, it, lends itself to misinterpretation because it combines information about the magnitude of an effect with information about the precision with which that effect is estimated. By contrast, confidence intervals (Cis) focus one's attention on an estimate of a more meaningful parameter (e.g. the rate difference) and, as a separate matter, on the precision of the estimate. The CI is a range of values that is likely to cover the true but unknown value (the extremely low up to the extremely high value of e.g. rate difference, mean difference, and Odds Ratio) if we measure the value many times on samples using the same method. Cis in a clinical trial where the result is statistically significant we might find that it is of no clinically Importance (in a very large sample). On the other hand, the effect of a treatment might be statistically not significant but In fact it is of clinical importance (in a small sample). Cl Is affected by sample size: the larger the sample size the narrower the interval of Cl. The interval Is also affected by the standard error and hence by standard deviation and the confidence level we claim. The higher the confidence level (90%, 95% or 99%, arbitrarily 95% Cl is commonly used) the wider the confidence interval. In case that the 95% Cl does not include zero value (in mean or proportion difference) or one (in Odds ratio or relative risk) it also reflects statistical significance (p<0.05) with a = 5%. A sample size that is enough in terms of power might not be enough in terms of precision due to the confidence Interval level we choose.

Key Words : P value - confidence interval - clinical importance - sample size - statistical significant