Contributors: Roberto Vargas MD, Elif Irem Sarihan MD, Semihcan Doken, Mohamed Abazeed MD PhD
The paper in npj Precision Oncology is here: go.nature.com/2Mt9hhb
Precision oncology is at the leading-edge of personalized medicine efforts. Cancer therapeutics are actively transitioning from generalized treatments toward biological and immuno-based treatments that are tailored to the molecular alterations found in the tumors of individual patients. However, very few patients benefit from personalized treatments in oncology. This is attributed to the inability to predict the response to traditional treatments (chemo-/radio-therapy), the small proportion of patients that are paired with a targeted drug for identified mutations and the not infrequent discordance between assignment to a particular therapy and treatment response. Therefore, precision oncology strategies require substantial optimization—there remains a significant need for models of cancer that can predict treatment response.
What makes this different than other cancer research involving mouse models?
Engraftment of animal or human tumor cells into an animal, usually a mouse, has served as a very useful resource to study cancer biology. Patient-derived xenografts (PDX) in particular have been shown to closely recapitulate the genetic features of their tumors of origin. Although PDX models have been used extensively to study tumor biology, they have yet to lend themselves to informing the treatment of the donor patient directly. Our goal was to test whether a patient’s own model, developed and tested contemporaneous with their clinical care, can guide their treatments. These “avatars” studies must remain a step ahead, testing responses to standard of care treatments, new candidate therapies identified by combining other personalized medicine strategies (e.g. genomics) and/or drugs that are the subject of ongoing clinical trials, before the decision for therapy is made by the clinician.
How can this technology be used to guide treatment?
Let’s formalize the utility of this approach by calculating the probability of responses before and after a theoretical avatar study is completed (figure). Patient populations, depending on the cancer type and chosen therapy, can have low, intermediate or high probabilities of responses to treatments [i.e.pre-test probability or P|X(A)]. This probability can be estimated from historical clinical response rates. After conducting an avatar study, the likelihood of response may vary in the population whose avatar study demonstrates a response (or positive test) compared to the population that does not demonstrate a response (or negative test). If there are no significant differences in the populations after the test is conducted [log(likelihood ratio = 1) = 0, or the diagonal in the figure], then the avatar study does not alter the post-test probability P|X(B). That is, it does not inform clinical decision making. If, however, the avatar study improves the prediction of therapeutic response, then it has significant and quantifiable utility. The extent of utilization will depend on the likelihood ratio, which is likely to be distinct based on individual drugs and cancer types. Viewing avatar-based trials through this lens allows us to understand the nuances of these studies, quantify their effects on prediction and, as a result, identify the clinical contexts that avatar studies may be the most useful.
Figure 1. Graphical depiction of pre- (x-axis) and post-test probabilities (y-axis) as function of the log(likelihood ratio). The test is a theoretical avatar-based study that demonstrates response (positive test) or no response (negative test).
What are the advantages of this approach?
These studies could help oncologists recommend treatments with improved response rates and/or avoid cancer treatments that are unlikely to work. This will result in the prevention of unnecessary treatment-related toxicities and the amelioration of the opportunity cost of receiving a non-efficacious treatment, as there is the obligate trial period of several months to determine if a patient is not responding to a recommended therapy. Finally, for rare cancer types that do not lend themselves readily to randomized control studies, these studies can inform clinical decision making in a setting where there are no well-established guidelines or recommendations.