Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.

Accept all cookies Reject all non-essential cookies Find out more

Synthesis methods

No. 14

Task: Describe the meta-analysis methods used to synthesise IPD. Specify any statistical methods and models used. Issues should include (but are not restricted to):

  • Use of a one-stage or two-stage approach.
  • How effect estimates were generated separately within each study and combined across studies (where applicable).
  • Specification of one-stage models (where applicable) including how clustering of patients within studies was accounted for.
  • Use of fixed or random effects models and any other model assumptions, such as proportional hazards.
  • How (summary) survival curves were generated (where applicable).
  • Methods for quantifying statistical heterogeneity (such as I2 and 2).
  • How studies providing IPD and not providing IPD were analysed together (where applicable).
  • How missing data within the IPD were dealt with (where applicable).

Text: The methods used in the standard analyses for the EBCTCG meta-analyses are those described in the mid 1970s,* and rely on the use of time to event data to calculate the log-rank statistic for each contributing trial, analyzed separately. The analyses are stratified by trial, age, ER status and nodal status . These statistics for each trial are then combined to estimate the relative effects of the interventions and the associated confidence intervals for the meta-analyses. . If a log-rank statistic (o − e) has variance v, then, defining z = (o − e) / √v and b = (o − e) / v, the event rate ratio (RR, newer treatment vs control) is estimated as exp(b) with standard error SE = (RR − 1) / z. Either RR and its SE are cited, or confidence limits for RR are derived from those for b (by normal approximations). 2p indicates two-sided significance. An estimate of the absolute effects of the intervention and survival curves are calculated and plotted using methods described in detail in:Early Breast Cancer Trialists' Collaborative Group Treatment of Early Breast Cancer. Volume 1. Worldwide Evidence 1985-1990. Oxford University Press: Oxford, 1990. Available from https://www.ctsu.ox.ac.uk/research/ebctcg/furtherinformation/original-methods-for-ebctcg-meta-analyses

Since the mid 1990s,† the effects of treatments on breast cancer mortality have been calculated using log-rank subtraction, in which the log-rank statistics for mortality without recurrence [ie, censored at recurrence] aresubtracted from those for overall mortality to estimate breast cancer mortality.

* Peto R, Pike MC, Armitage P, et al. Design and analysis of randomized clinical trials requiring prolonged observation of each patient. II. Analysis and examples. British Journal of Cancer 1977; 35(1): 1-39.

† Early Breast Cancer Trialists' Collaborative Group. Effects of radiotherapy and surgery in early breast cancer: an overview of the randomised trials. New England Journal of Medicine 1995; 333: 1444–1455.

In this section