Cumulative local recurrence rate is a misleading and non-representative outcome measure for early breast cancer trials

Abstract

In many breast cancer radiotherapy trials, the results are presented in the form of cumulative incidence rates of local recurrence or Kaplan-Meier plots, in which deaths are censored. Censoring - using patients' length of follow up until the point when they had last been seen alive - is included in the statistical model, under the correct assumption that they will continue to have a risk of developing a local recurrence. Censoring should be non-informative and balanced. However, if shorter follow up is unbalanced between treatments, or if shorter follow up is due to death (from whatever cause), these assumptions and therefore the model is no longer valid. It is therefore ambiguous to statistically ignore deaths when reporting local recurrence, by censoring them. We illustrate, with examples from randomised trials, why and how such graphs cannot give patients and clinicians a clear indication of the effects of treatments or prognosis. For instance, in one of these examples, 60% of patients were alive at 10 years, so those alive without a local recurrence should inevitably be lower than 60%, rather than the 90% estimated using the above method. The simple way to avoid this error is to turn the analysis on its head, by reporting chances of success rather than failure, by reporting the probability of being free of local recurrence (i.e. both death and local recurrence are events). This estimate truly represents what really happens to patients in terms of local control and the relative effectiveness of treatment(s) comprehensively. It also conforms with the recommendations of ICH-GCP, European (DATECAN) and American (STEEP) guidelines.