The calculation of hazard ratios from data published in reports of randomised controlled trials; are they valid?

Article type
Authors
Glenny A1, Furness S1, Worthington H1
1Cochrane Oral Health Group, UK
Abstract
Background: A meta-analysis of individual patient data (IPD) offers benefits in terms of the quality of data. However, despite being the gold standard of systematic reviews, they are not always feasible and most systematic reviews rely on published data.

Objectives: To compare the hazard ratios (HR) calculated from IPD with those calculated from data published in reports of randomised controlled trials.

Methods: A Cochrane systematic review has been undertaken to assess the effectiveness of radiotherapy regimens for oral cavity and oropharyngeal cancers. The primary outcome was total mortality expressed as a HR. If HR were not quoted in the included trials, the log HR and the standard error (SE) were calculated from the available published summary statistics or Kaplan-Meier curves, or these data were requested from authors. A separate Cochrane review of IPD for altered versus conventional fractionation for head and neck cancer has also been published. There is overlap in the trials included in the two reviews. A comparison of the HR calculated from the published data/graphs and from the IPD data has been made.

Results: Thirty trials were included in the review of radiotherapy for oral cavity/oropharyngeal cancers and 15 trials in the review of altered fractionation for head and neck cancers. There was an overlap of nine trials (11 comparisons) within the two reviews. The HRs calculated by the two methods varied by less than 0.1 in 8/11 comparisons made; the largest variation in HR was 0.28. The direction of effect did not differ and in only one comparison would the interpretation of the findings have been altered.

Conclusions: A meta-analysis of IPD is the ideal. However, where this is not feasible, it is possible to calculate reasonably valid HR from published data.