Michael Hayden Rosenthal MD, PhD, Presenter: Nothing to Disclose

Nikhil H. Ramaiya MD, Abstract Co-Author: Nothing to Disclose

The Response Evaluation Criteria in Solid Tumors (RECIST) are the most widely accepted standards for assessing tumor response in clinical trials, but the classification thresholds in RECIST are based on historical data that predated the adoption of cross-sectional imaging. The impact of these thresholds on the statistical power of clinical trials that use RECIST is unknown. This work tests that relationship in a simulated environment.

A clinical trial simulator was constructed using the Matlab software environment. Tumor progression and subject mortality were modeled using logistic functions that incorporated common demographic and tumor biologic risk factors. Tumor growth was simulated using an exponential model, and tumor size was a continuous factor in the logistic model of mortality. Tumor measurements were corrupted by white noise proportional to the square root of their magnitudes to simulate measurement error. Progression, mortality, and tumor growth were simulated for sixty time points. Progression of disease was defined at each time point for each subject for a range of thresholds from 5-400%. Only the first progression event was counted. The relationship between the statistical strength of the log-rank test of survival and a range of possible response thresholds was evaluated via Monte Carlo simulation.  

The simulated clinical trial had a total of 10,000 subjects randomized to two treatment arms. The odds ratio for progression between the treatment arms was assumed to be 0.3. The Z-score of the log-rank test of survival varied significantly over the range of possible tumor response thresholds (range 4.3 - 16.9). This variation was dependent upon both the assumed tumor growth rates and magnitude of the measurement noise.

In this computational model, the statistical power of a simulated clinical trial varied significantly with changes in the tumor response threshold. This suggests that more optimal threshold selection could improve the power of clinical trials without increasing the number of subjects. Further investigation is needed using actual clinical trial data.

This work could lead to improvements in the assessment of tumor response, which could improve cancer care and the power of clinical trials of new cancer treatments.

Rosenthal, M, Ramaiya, N, Analysis of the Relationship between Tumor Response Criteria and the Statistical Power of Clinical Trials in a Computational Model of Cancer Imaging.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12043910.html