Rethinking outcomes and statistical analyses for paediatric cancer trials

Paediatric cancers are rare diseases, and trials in this field have the major problem that sample sizes are limited, often severely. It is as important in rare diseases as in common ones that treatments are based on high quality evidence from randomised controlled trials, so there is a challenge for trial design. Several innovations have been introduced to improve the efficiency of paediatric cancer trials, but one area that has not been addressed is the outcomes and statistical analyses used. Trials invariably use outcomes that are similar to adult cancer trials: tumour responses, usually analysed as a binary outcome (response/non-response), and time to event outcomes, usually overall survival time and various forms of event-free survival time. However, when the sample size is restricted, it is crucial for the analysis to be as efficient and informative as possible. The problem is that the conventional methods do not use information efficiently, and this directly impedes our ability to identify the best treatments. Improving the efficiency of outcome measurement and analysis would improve the quality of evidence provided by trials in paediatric cancer, enabling better evaluation of treatments and improving health. In this project we will investigate whether a novel approach to outcome measurement and analysis, the use of ordinal longitudinal outcomes, would be advantageous for paediatric cancer trials. This approach records patients’ outcomes on an ordinal scale that includes all relevant events, and is recorded repeatedly throughout follow-up, giving a trajectory through time of the patient’s clinical course. This type of outcome includes much more information than conventional methods, and thus should enable more efficient discrimination between therapies. Analytical methods, developments of ordinal logistic regression, allow multiple clinically useful summaries, facilitating greater learning from trials. This project will consist of five phases: Systematic review of outcomes and statistical methods used in paediatric cancer trials; Selection of four case studies based on trial data sets held by Cancer Research UK Clinical Trials Unit, and development of an ordinal longitudinal outcome for each of these; Reanalysis of each trial using ordinal longitudinal models, and comparison of the results of the reanalysis with the original trial’s outputs, asking the questions: would the conclusions of the trial have been different? could the trial have reached its conclusions more quickly?; is the quality of the information better (e.g. less uncertainty) or are additional insights available from the ordinal longitudinal analysis?; Consideration of issues in trial planning, and reporting and communication of results from ordinal longitudinal models; Dissemination of the results widely among the scientific community, and other stakeholders including regulators and policy makers. The results of this project will be directly relevant to trial design and analysis in other areas of cancer, and in rare diseases generally, and will help to establish the methods as an alternative way to conduct cancer trials.