Individual patient data analysis to assess modifications to the RECIST criteria.

BACKGROUND After the initial RECIST 1.0 were published in 2000, the criteria were widely implemented in the scientific oncology community. Since then, the RECIST working group has identified several issues to examine further. Two key issues that required careful, data-based assessment were the maximum number of lesions that should be assessed at each evaluation and the added value of requiring confirmation of response. METHODS To address these questions, data were obtained from 16 clinical trials in metastatic cancer, with patients enrolled between 1993 and 2005. A total of 6512 patients were included in the primary analysis dataset, accounting for over 18,000 potential target lesions. Nine percent of the included patients (n=585) had six or more reported target lesions. The response and progression outcomes in the database were calculated using an adjusted RECIST methodology with a maximum of 5 (or 3) target lesions with/without confirmation and this was compared to the original RECIST version 1.0 which required up to 10 target lesions plus confirmation of response. RESULTS Assessment of 5 lesions per patient led to a difference in best overall response assignment for an estimated 209 (3.2%) patients as compared to RECIST version 1.0. However, these changes did not affect the overall response rate. Progression-free survival was only minimally affected by measuring fewer lesions. In contrast, removing the requirement for response confirmation led to a significant increase in the numbers of patients classified as responders, resulting in a relative increase of approximately 19% in response rate. An algorithm using a maximum of three target lesions shows high concordance with the 10 lesions requirement in terms of response and TTP assignment. Concern that appropriate assessment of disease within an organ requires two lesions to be followed per organ suggests the approach of following two target lesions per organ, up to a maximum of five target lesions overall. Both strategies seem reasonable based on the data warehouse. The requirement of response confirmation in trials where this is a primary end-point is recommended to be maintained as its removal would substantially increase reported response rates.