Frederick Howard, MD, a medical oncology fellow in the Department of Medicine at the University of Chicago, discussed the clinical benefits of the SimBioSys TumorScope model for the treatment of early breast cancer in an interview with Targeted Oncology.
Even with aggressive therapy, the majority of women with early breast cancer will not achieve a pathological complete response (pCR) with neoadjuvant chemotherapy. Investigational agents such as immunotherapy may increase pCR rates, but are also associated with irreversible immune-related toxicities, research shows.
SimBioSys TumorScope is a biophysical model designed to predict how patients will respond to treatment. The device could help to determine if patients would benefit from neoadjuvant chemotherapy intensification or de-escalation and allow for the use of precision medicine.
A retrospective analysis looked at 144 tumors from 141 patients who had an average age of 52 years. Of those patients, 65% had stage II disease and 19% had stage III disease. The device was used to predict pCR in patients.
Frederick Howard, MD, a medical oncology fellow in the Department of Medicine at the University of Chicago, discussed the clinical benefits of the novel predictive model for the treatment of early breast cancer in an interview with Targeted OncologyTM.
TARGETED ONCOLOGY™: Can you give me an overview of independent validation of SimBioSys TumorScope to predict response to neoadjuvant chemotherapy in early breast cancer?
HOWARD: A pre-treatment MRI is used as the basis of the model. And the tumor is segmented into the actual tumor body, the vasculature, the fatty tissue, and then the treatment regimen is inputted and that is used to simulate how the tumor is going to respond. That takes into account a lot of complex factors such as the areas of tumor that are going to have high exposure to drugs related to the proximity to the vasculature. The individual tumor sensitivity based on the receptor subtype is also accounted for in the model.
The model was developed using several publicly available imaging data sets. And so, this was an approach to validate that model in an independent population. So, what we did is we queried our institutional database for all patients who had received neoadjuvant chemotherapy and had a pretreatment MRI along with pathologic response data available in the last 10 years. We found 144 patients who received standard of care regimens that could be modeled with the tumor scope program. We essentially performed this validation in a blinded fashion. We provided the pretreatment imaging as well as some patient demographic characteristics and the planned treatment regimen. Then, a simulation was made for each patient's response to chemotherapy over time. There was a pre specified cut off. The model essentially is a 3D model of the tumor which shrinks over time. There's a pre specified cutoff of what percentage tumor shrinkage or volume reduction at tumor core corresponded to pathologic complete response in some of the prior modeling of patients. The model predictions were then compared to the actual pathologic complete response data. Then, we saw very encouraging predictive accuracies with around 90% accuracy in predicting pathologic complete response overall as well as consistent performance across tumor subgroups and regimens.
What was the study design and what were you hoping to achieve with this analysis?
This was a retrospective blinded evaluation of tumor scope, and the primary end point was accuracy and prediction of pathologic complete response. Secondary end point measures included the prediction of the correlation of prediction of pathologic complete response with survival outcomes, including disease free interval event free survival and overall survival, as well as the accuracy of inter regimen, MRI volumes compared to model predictions of inter regimen volumes. In other words, not only can the model not only predict pathologic complete response, but how accurate is the model simulation of tumor size over time?
What were the results? Did anything surprise you?
The company has been developing this tool previously. This was our first exposure working with this promising new technology. To have a predictive accuracy of around 90%, to predict pathologic complete response from data that is available pretreatment is outstanding and exceeds anything that I think is really available in this field. It just makes it a an incredibly promising tool that could be used to help guide and direct therapy in the future, because of how accurate the predictions are. The other thing that was surprising is that not only were the predictions accurate, but in the cases that were misclassified, in other words, where the model predicted that the patient would have a pathologic complete response, but they didn't, that 8 of the 10 patients had cellularity of the tumor bed of less than 5%. I think 3 or 4 of the patients had a lymph node only disease and no residual cellularity within the tumor bed. And none of the patients who the model predicted to have a pathologic complete response but who didn't, ended up having disease recurrence, just illustrating that the model accurately stratifies patients who will have a good outcome or a poor outcome with neoadjuvant chemotherapy. I think that this is desperately needed in the field of breast cancer with the advent of immunotherapy and other more aggressive neoadjuvant treatment regimens, we expect that immunotherapy may be approved in the near future. But they're still with all the all the neoadjuvant immunotherapy trials, there has been no consistent biomarker to identify which patients are going to benefit from the addition of immunotherapy. But when you have models that can accurately predict response to standard chemotherapy, you can perhaps identify patients for whom the standard treatment is going to result in a pathologic complete response and good long-term outcomes. The patients who really do require that intensification of therapy, and which does come with its own potential risks, including the risks of long-term autoimmune toxicities from immunotherapy.
How can this model be incorporated into a community clinical practice?
So, there is further validation of this model ongoing and planned in several different cohorts. I think that prior to the model being incorporated into clinical practice, it's going to require FDA approval. I know SymBioSis has been seeking that, but I think further validation is planned prior to the ability to achieve FDA approval. I think that once approval is obtained if these results are consistent and further validation studies, then the applications are diverse in really a number of interesting applications that could be explored. So, for example, with a lot of our breast cancer neoadjuvant treatments for the standard of care treatments, we are thinking about the inclusion of anthracyclines. For example, for triple negative patients that have tumors that are above a centimeter or for hormone receptor positive patients who have axillary nodal disease. But these aren't based on tumor biology. These are based on kind of the risk of the cancer’s recurrence. If these models can accurately predict patients that will respond well to non-anthracycline based therapies, for example, then you can save women the risks of cardiotoxicity that come with anthracyclines, if we knew that they're going to respond well to an anthracycline free regimen. We also think that this is going to be a promising tool that can be used, as I mentioned, to help guide which patients need intensification of therapy with the addition of immunotherapy or who will be the best candidates for enrollment in neoadjuvant clinical trials.
What are the planned next steps for this research?
There are a number of avenues that we're exploring. Aside from further valid of the model, the model can also be used for in silico experiments. So, one avenue that we're exploring is the use is is in silico modeling for patients who received an immunotherapy regimen. One thing that we're seeing in a small subset of patients, the model currently doesn't accurately model response to immunotherapy. But it models response to standard of care therapy. So, what you can do is if a patient received an immunotherapy plus standard of care regimen? If they have a response, you don't know whether that response was due to the immunotherapy or because they would have had a good response to standard of care. So, what you can do is then take a patient and then model what their response would have been, theoretically with a highly accurate model to standard of care therapy and then see, was this patient someone who had received a great benefit from immunotherapy? Or was this patient who was going to respond anyway to their standard of care treatment? That allows you to do analyses on these patients for further biomarker discovery for immunotherapy response.