Comparison between protocols and publications for prognostic and predictive cancer biomarker studies.

Protocols are generally not accessible or not used for cancer biomarker studies. Publications were often explicitly discordant with protocols, particularly regarding biomarkers and endpoints. Our findings point to common unaddressed risk of bias in publications of major journals reporting the relationship between cancer biomarkers and clinical endpoints.