Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘When using the ITT method, the issue of how to classify subjects who drop out ofthe study – ie, stop attending for follow-up – must be dealt with before a studystarts. In the worst-case scenario, one could assume that all subjects who withdrawcount as deaths or as ‘no events’, but in reality these subjects will suffer a mixtureof events. In some studies it may be assumed that the event rate in the missingsubjects is the same as in the group of subjects with data available (a methodknown as imputing event rates). This imputation method should only be used if itproves impossible to get outcome data from a large proportion of subjects, and theresults should be presented with and without imputation.How do the results of ITT and PP analyses compare?When performed on data from the same study, either or both of the results fromthe ITT and PP analyses might reach significance. The following are suggestionsfor how these combinations of results can be interpreted:• In most trials reaching significance – ie, where the results from an ITTanalysis support rejecting the null hypothesis (no significant differencebetween treatment groups) – if the results of the PP method are slightlymore significant, this suggests that while some subjects were noncompliant,they were equally distributed between groups.• If the PP results are much more significant than the ITT results,this might suggest a high rate of noncompliance in the study.• If the PP results are not significant but the ITT results are significant,there might be a confounding reason for the difference in outcomesother than it being due to treatment differences. For example, in a trialcomparing medicine and surgery, a significant number of subjects maydie before having surgery.• If the result from an ITT analysis is not significant, while that from aPP analysis is significant, this might be due to a considerable proportionof subjects switching treatments (crossover) in one direction (eg, fromplacebo to new treatment).Are there analysis strategies beyond the ITT and PP methods?Given the limitations of an ITT analysis, occasionally we may feel that an explanatoryPP analysis is more suitable since it attempts to remove the effects of variablecompliance patterns. However, the PP method can lead to a biased comparison ifcompliance itself is associated with the effectiveness of a treatment or the risk ofoutcome events [4]. For example, elderly subjects are more likely to have side-effects259