A case study on the reporting of effect size estimates in breast cancer trials

Article type
Wilson M1, Ghersi D2, Askie L3
1Cochrane Breast Cancer Group, NHMRC Clinical Trials Centre, University of Sydney, Australia
2Research Translation Group, National Health and Medical Research Council, Canberra, Australia
3Systematic Reviews Health Technology Assessment, NHMRC Clinical Trials Centre, University of Sydney, Australia
Background: Guidance on trial reporting from the CONSORT statement outlines that primary and secondary outcomes should be presented with estimated effect sizes and range (item 17; [1]). In oncology trial reports, time-to-event outcomes are commonly reported in the form of survival curves. When pooling time-to-event data for systematic reviews, the Hazard Ratio (HR) from survival curves is the most appropriate way to express the results [2].

Objectives: To conduct a preliminary assessment on whether trial reports comply with CONSORT by providing the appropriate effect estimate and range for each time-to-event outcome.

Methods: A Cochrane Review update on Taxane-containing regimens for metastatic breast cancer was used as a test case. For each included study that had a full-text publication, details on the time-to-event outcome reported and the presence or absence of (i) a survival curve (ii) a P-value and (iii) HR, median survival time or risk ratio and associated Confidence Interval (CI)were extracted.

Results: Twenty-four trial publications were included and publication dates ranged from 1995 to 2011. Nearly all trial reports (23 out of 24) provided survival curves for overall survival and time-to-progression or time-to-failure and P-values, and reported one effect estimate (i.e. either median survival times, risk ratios or HRs). The ranges for median survival times or risk ratios were absent in 6 out of 24 trial reports. Also, 6 out of 24 trials reported HRs and CIs, and this was observed in trials published in the early 2000s and 2011.

Conclusions: This case study showed that effect size estimates for time-to-event outcomes were consistently reported, suggesting compliance with CONSORT. However, only a small number of trials (6/24) reported the HRs and CIs from the survival curves. The lack of reporting relevant effect sizes, such as HRs, for time-to-event outcomes may have implications when undertaking meta-analyses.