The analysis of failure time data in crossover studies.

Clinical trials of new drugs in the treatment of angina pectoris frequently make use of exercise tests to evaluate efficacy. The crossover design is often employed. The methods commonly used to analyse the various exercise times, for example, 'time to pain', are insensitive and potentially biased by the manner in which they deal with the censored nature of the data. Survival analysis can be adapted for use in crossover trials, both in a relatively simple way, and also through the full power of the Cox model. This is considerably more sensitive and not subject to the same bias. This methodology leads to the use of median survival times to illustrate treatment effects and this provides a practical interpretation of clinical relevance. The estimation of median survival times in crossover trials poses some special problems. The methodology is illustrated throughout by means of a specific two-period example in which atenolol was compared with the combination of atenolol and nifedipine. The three-period design is also briefly discussed.