A statistical simulation study finds discordance between WHO criteria and RECIST guideline

Objective: Tumor shrinkage has been adopted as an end point for evaluating the effectiveness of new anticancer agents. The WHO (World Health Organization) criterion suggested measuring the tumor shrinkage by the change in the product of maximal diameter (MD) and the corresponding largest perpendicular diameter (LPD). The RECIST (Response Evaluation Criteria In Solid Tumor) guideline proposed using the change in MD only, based on the observation that this measure is more linearly related to tumor cell kill than the cross product (MD*LPD). Both criteria classify patients into four categories of response: complete response (CR: total disappearance), partial response (PR), stable disease (SD), and progressive disease (PD) but the criteria used in the definition of PD vary. It was anticipated that patients' actual response categorization would not be considerably affected by utilizing the RECIST criteria instead of WHO. Empirical evidence supporting this fact was provided by retrospective analysis of several large datasets. Study Design and Setting: A statistical simulation is performed to generate tumor measurements and patient response data under meaningful probability distributions with parameters based on data from 130 patients on clinical trials at a cancer center. Concordance measures between the two response criteria (Kappa coefficient and percentage disagreement per response category) are assessed systematically over various combinations of the percentage of elliptical tumors at baseline and the percentage of tumors changing shape from baseline to follow-up. Results: The overall percentage of disagreement between the two methods of response assessment is found to be in the range of 14-20%. The patients categorized by WHO in the PR, SD, and PD groups fall into a different category when assessed by RECIST between 8-16%, 3-12%, and 32-35% of the times, respectively. The kappa coefficient ranges between 0.68-0.77. The proportion of elliptical tumors at baseline does not greatly impact the concordance, but the magnitude of the change in the aspect ratio has a large impact. Conclusion: Response assessment as measured by RECIST, with both a change in the underlying metric and change in definition of progression, often results in different categorization of response compared to WHO. The difference in response categorization may be problematic when new experimental therapies are compared to conventional agents whose response rates have been established in historical trials. The apparent lower rate of disease progression with RECIST may mean that more patients remain on therapy. Higher percentages of patients with SD need to be interpreted cautiously by distinguishing those due to the change in the response criterion as opposed to those induced by drugs using pathways such as angiogenesis where disease stabilization is expected rather than shrinkage of tumor. (C) 2004 Elsevier Inc. All rights reserved.