Repeated measurements and multiple comparisons in cardiovascular research.

In cardiovascular research, experiments are commonly performed in which repeated measurements are made in the same individual at predetermined intervals of time or at ascending levels of stimulus or dose of drug. The goal is usually to test the effects of treatments or disease state on the time course of the response, or on the stimulus-response relationship. Since the passage of time or the order of stimuli or doses is fixed, statistical analysis of the results of such experiments is associated with an excessive risk of false positive interferences (type I error) unless special precautions are taken. The nature of the statistical problems associated with repeated measures experimental designs, and several solutions to them, have been discussed. An approach much favoured by cardiovascular investigators is to make multiple pairwise contrasts between treatments at each time or dose, or between times or doses within each treatment. This greatly inflates the risk of type I error unless special precautions are taken, and the information provided by making multiple contrasts is of limited value. I believe that repeated measures analysis of variance, with a correction for multisample asphericity, usually provides the most informative and least biased test of the biological hypotheses proposed by cardiovascular investigators. Other analytical techniques, such as comparing areas under curves and regression analysis, have also been discussed. Summary recommendations are given in the table.