Published in Cancer Detection and Prevention 1996; 20(5).

Biostatistical issues in the toxicological evaluation of chemopreventive agents

CD Sherman PhD

A major biostatistical issue facing any toxicologist with the task of evaluating a compound for its ability to prevent carcinogenesis is statistical power. Statistical power may be defined as the probability of detecting a chemopreventive agent. In general, statistical power is maximized when the differences between the control group and the chemopreventive treated groups are large. In most experimental and clinical settings, the difference in the proportion of tumors between treatment and control is not particularly large and can never get very large since the lower bound on the proportion responding is zero. To maintain statistical power, the statistician may easily suggest an increase in sample sizes for each treatment group; however budgetary, experimental, and time constraints tend to drastically attenuate these recommendations. Experimental design features, statistical analysis approaches, and biomathematical models will be described that can accommodate these constraints and provide the information needed to meet the objectives for evaluating the effects of chemopreventive agents. Methods for simultaneous analysis of larger and more diverse forms of information than a simple cancer bioassay will be described. The focus will be on mechanistic modeling and its role as a tool for data analysis and interpretation.

Paper presented at the International Symposium on the Impact of Cancer Biotechnology on Diagnostic and Prognostic Indicators; Nice, France; October 26 - 28, 1996; in the section on Chemotherapy.

http://www.cancerprev.org/Journal/Issues/20/5/505/1544