Interpreting Effect Sizes in Context

This lesson ties the module together by asking how to judge whether an effect actually matters. The first move is to read relative and absolute measures together. A large relative effect (a risk ratio, odds ratio, or hazard ratio far from 1) on a rare outcome can be clinically trivial, while a modest relative effect on a common, serious outcome can be hugely valuable. The absolute measures, the absolute risk reduction, number needed to treat, and mean difference, anchor the relative ones in real-world stakes. Second, compare the effect to the minimal clinically important difference (MCID), the smallest change patients actually notice or that would change management. A statistically significant 1 mmHg drop, or a 1-point shift on a 100-point scale, may fall below the MCID and simply not matter. Third, weigh the effect against harms and costs: a small benefit can still be worthwhile for a cheap, safe drug, while a 'large' benefit may not justify a toxic or expensive one. Fourth, consider the outcome's importance and durability, since an effect on a hard outcome such as death or stroke outweighs the same-sized effect on a surrogate marker, and a lasting benefit matters more than a transient one. Finally, never read an effect size without its confidence interval. A wide interval means the true effect could range from trivial to large, so even an impressive point estimate may be uncertain, while a narrow interval around a meaningful effect is genuinely persuasive. Putting these together, relative plus absolute, measured against the MCID, weighed against harms and costs, for an outcome that matters, and read with its confidence interval, turns a statistical result back into a clinical decision. This judgment, not the p-value, is what should determine whether evidence changes practice.