Survival end point reporting in randomized cancer clinical trials: a review of major journals.

PURPOSE

Several publications showed that the standards for reporting randomized clinical trials (RCTs) might not be entirely suitable. Our aim was to evaluate the reporting of survival end points in cancer RCTs.

METHODS

A search in MEDLINE databases identified 274 cancer RCTs published in 2004 in four general medical journals and four clinical oncology journals. Eligible articles were those that reported primary analyses of RCT with survival end points. Methodologists reviewed and scored the articles according to seven key points: prevalence of complete definition of survival end points (time of origin, survival events, censoring events) and relevant information about their analyses (estimation or effect size, precision, number of events, patients at risk). Concordance of key points was evaluated from a random subsample.

RESULTS

After screening, 125 articles were selected; 104 trials were phase III (83%) and 98 publications (78%) were obtained from oncology journals. Among these RCTs, a total of 267 survival end points were recorded, and overall survival (OS) was the most frequent outcome (118 terms, 44%). Survival terms were totally defined for 113 end points (42%) in 65 articles (52%). Accurate information about analysis was retrieved for 73 end points (27%) in 40 articles (32%). The less well-defined information was the number of patients at risk (55%). The reliability was good (kappa = 0.72). Finally, according to the key points, optimal reporting was found in 33 end points (12%) or 10 publications.

CONCLUSION

A majority of articles failed to provide a complete reporting of survival end points, thus adding another source of uncontrolled variability.