Site/Central discordance: late phase oncology trial with differences between site and central reads found
Adjudicator agreed with site reader over central reader 5:1.
RadMD performed central over-read and identified that all central reviewers were making similar type errors. Central readers from single institution, analysis identified possible informed bias in central reads - all reviewers missed same type of lesions on certain modalities.
All reviewers from the same institution, possibility of informed bias; independent review appeared less robust than site read; primary endpoints (central review results) at risk.
Replace central reader pool with radiologists from different institutions to reduce any potential bias.