Influence of measurement error on assessment of response to anticancer chemotherapy: proposal for new criteria of tumor response.

The decision to use a given type of chemotherapy to treat cancer patients is often based on the prior demonstration that a proportion of similar patients has "responded" in a clinical trial. Most responses are recorded as a partial shrinkage of tumor, defined usually as a greater than 50% shrinkage of the sum of cross-sectional areas of index lesions for at least one month. The errors in categorization of response have been estimated by comparing measurements of several physicians on real or simulated malignant lesions. False categorization of partial response based on a comparison of two measurements of the same lesion was 1.3% and 12.6% for large and small simulated nodules, respectively, 13.1% for malignant neck nodes, and 0.8% for metastatic lung nodules. Partial response for hepatic lesions has been defined by a 50% or 30% decrease in liver span below the costal margin; these definitions led to a false categorization of partial response of 8.5% and 18.4%, respectively. Larger errors are evident when using the current definition of disease progression that requires only a 25% increase in area. False categorization of response is increased by comparing any of serial measurements with the initial lesions, as is usually done clinically. Many published trials have used criteria for response that are subject to large errors; an uncritical interpretation of their results may lead to inappropriate treatment of patients. Based on the results, new criteria for evaluating tumor response are proposed.