Review Illustrates Statistical Principles in Clinical Trials
The first in a series of State of the Art Reviews on statistical principles, published Nov. 30 in the Journal of the American College of Cardiology (JACC), illustrates how best to display results, use confidence intervals and assess p-values in reports on randomized clinical trials. Throughout the series, Stuart J. Pocock, PhD, et al., use real trials as topical examples and provide solutions to common misconceptions.
According to the authors, the first table displayed in any clinical trial should report the patients’ baseline characteristics, including key demographic variables and related medical history, separated by treatment group. In order to reduce the size of the first table, a column showing the combined results for all groups may be unnecessary. In addition, reporting means or percentages to one decimal place is sufficient.
Following this, “the key table for any clinical trial displays the main outcomes by treatment group,” state Pocock, et al. “For trials concentrating on clinical events during follow-up, the numbers (in percentages) by group experiencing each type of event should be shown.” Most major trials also use a Kaplan-Meier plot of time-to-events to show the primary outcome(s), they add.
The authors explain three main types of outcome data – binary responses; time-to-event outcomes; and quantitative outcomes – and the standard methods for estimating treatment effects and their confidence intervals. They further note that in order to express statistical uncertainty, researchers should use a 95 percent confidence interval around any estimate in a clinical trial.
The review also addresses p-values, which, according to Pocock, et al., are often misused in classifying the results of clinical trials as positive or negative. “This oversimplification is an abuse of the p-value, which can be a valuable statistical tool when interpreted appropriately,” the authors write. “Alongside an estimate of treatment difference and its 95 percent confidence interval, the corresponding p-value is the most succinct, direct route to expressing the extent to which it looks plausibly like a real treatment effect, or rather could readily have arisen by chance.”
Listen to the audio commentary by Valentin Fuster, MD, PhD, MACC, editor-in-chief of JACC here.
< Back to Listings