Mathematics and Statistics
Assessing the Prognostic and Predictive Ability of an Avatar Mouse Model in Cancer Treatment
Siqi Fan, Vassar College ’17, Melanie Lai Wai, Vassar College ’16 and Prof. Ming-Wen An and Dr. Sumithra MandrekarModern cancer treatment research is moving towards personalized therapy. Biomarkers can be important for treatment selection. A biomarker is predictive for a treatment T if the treatment effect (T vs. control) differs between biomarker subgroups, and prognostic if the outcomes of one subgroup are better than those of another, regardless of treatment. Avatar mouse models are a novel technology where mice are engrafted with a patient’s tumor, given different treatments, and then followed-up for outcome. These models can serve as markers that inform on patient outcomes to different approved treatments. However, the usual definitions of prognostic/predictive do not readily apply, as there is no control arm in our patient data. We therefore aim to develop a modified definition. We used simulation studies to understand prognostic/predictive first in the usual sense, and then in our setting. Specifically, we simulated data for mouse and patient outcomes under various clinically relevant scenarios. Based on the results, we proposed the following modified definition: Given a set of N treatments, consider N binary biomarkers, each associated with the mouse response to one of the N treatments. A biomarker associated with treatment X is predictive for X if the relative effect (X vs. remaining treatments) differs between biomarker- defined subgroups; and prognostic if the outcomes in one subgroup are better than those in the other, regardless of treatment. Under this definition, the prognostic/predictive behaviors of biomarkers are relative to the others under consideration, which poses a limitation. For future work, we will refine this definition and develop clinical trial designs for validating the prognostic/ predictive behavior of biomarkers arising from Avatar models.