Neoadjuvant Therapy (NAT) for breast cancer results in higher rates of breast conservation surgery. Importantly, there is data supporting the use of multiple NAT regimens, with little guidance as to which one would be most effective for a given patient. Thus, tools that can accurately predict patient outcome before cancer treatment can provide guidance during treatment planning and increase the chances of therapy success.
Here we present the first biophysical modelling platform, TumorScope (TS), capable of constructing a 3D model of the tumor microenvironment by integrating models for tumor morphology, metabolism, drug delivery and behavior, and vascularity. Thereby allowing visualization of response to multiple NATs for a given tumor. TS integrates patient data to create an in-silico tumor model capturing the tumor response trajectory, volume, dimensions, pathological complete response (pCR), residual cancer burden and tumor morphology, and risk of recurrence (ROR). To illustrate the clinical applicability and to validate TS we performed a single center validation study.
Previously acquired pre-treatment standard of care (SOC) diagnostic data (demographics, drug regimen, imaging (DCE MRI), and pathology) was used as TS inputs. Using this baseline data only, TS predicted weekly volumetric response through the treatment time and used simulated residual volume to predict pCR and assessed the ROR. The validation was performed using ground truth from post treatment assessment of pCR, radiographic volumes extracted from MRIs and ROR information.
A chart review was performed to identify consecutive patients who had received NAT and had a corresponding pre-treatment MRI at University of Cincinnati. A validation set was generated from this list and data processed through TS (n=81, after excluding 9 patients with poor quality DCE-MRIs). TS predicted pCR was predefined as a residual tumor volume < 0.01 cm3 or a 99.9% or greater reduction in tumor volume. Performance metrics of TS were calculated.
TS response (pCR vs. residual) prediction accuracy had an area under the curve (AUC) =0.912 with sensitivity and specificity of 88.9% and 92.5%, respectively. The performance was robust across all subtypes.
Using SOC diagnostic data TS predicted the reduction in tumor volume with a median absolute volumetric error of 3.4% vs radiographic volume from pre-surgery DCE MRIs (n=59). TS prognostic ROR (High vs Low) had a hazard ratio [HR] 12.705, p=0.01 which was comparable as a prognostic indicator to actual pCR which had a [HR] 5.81, p=0.08.
In summary, TS accurately predicts patient-specific response to NAT in early breast cancer using only baseline SOC pre-treatment data. This approach to a comprehensive view of the tumor represents a major step-forward in personalized clinical decision making. The results from this study will be used in a multi-center meta-analysis of TS performance characteristics.
View the full publication here.