Understanding confidence intervals

In this episode, hear from Dr Kate Scrivener, Dr Aidan Cashin and Clinical Associate Professor Mark Elkins about understanding confidence intervals.

The point of studies that compare the effects of treatments is to give readers an idea about what would happen if a patient received one treatment versus another. The study does this by producing an ‘effect estimate’. For continuous measures this is the between-group difference; the mean outcome score for the intervention group minus the mean outcome score for the control group. Note that we are not talking about p-values here, for a range of reasons p-values are not useful for informing treatment decisions.

It is important to recognise though that the effect in the study comes from a study sample. One implication of this is that the best the researchers can do is provide an estimate of the effect in the whole population. All estimates are imprecise, and it matters how imprecise they might be. The most important and useful tool researchers have to describe the precision of an effect estimate is the confidence interval.

Confidence intervals are often misinterpreted. They do not represent the range of effects that 95% of patients will experience, or the largest and smallest effects an individual patient can expect.

The technical explanation of a confidence interval is quite complicated but there is a way to interpret them that is close enough for clinical purposes. The confidence interval is the range of values that the population effect most likely falls into. So, if a trial has a mean between-group difference of 2 points, with a confidence interval from 1 to 3, then the best estimate of the treatment effect is 2 points, but it could be anywhere from 1 point to 3 points.

For a clinician, the range of plausible effects (values within the confidence interval) can form part of the discussion with a patient about treatment options in coming to a shared decision.