Background: Clinical trials can be logistically burdensome, financially expensive, and potentially detrimental to patient outcomes. The NeoSphere study, which investigated the efficacy of docetaxel (T), pertuzumab (P), and trastuzumab (H) in combination with one another, was performed recently over an eight year stretch with n=254 operable breast cancer patients distributed across four study arms. Given the inherent hurdles of clinical trials, we set out to replicate the NeoSphere Trial in silico. We posit that never-before-possible in silico clinical trials can accurately forecast regimen efficacy/pathological complete response (pCR) rates and expand insights well in advance of human clinical trials. Analysis of tumor microenvironment and chemical interactions therein can enable drug developers to garnish more comprehensive insights than ex vivo or in vitro techniques. We aim to comprehensively understand cancer by leveraging our commercially available biophysical modeling platform and using all major hallmarks of cancer variation as input.
Methods: Selecting from the SimBioSys Virtual TumorBank of 3000+ past patient breast tumors and accompanying standard of care (SoC) clinical data, we matched NeoSphere sample composition based on age, tumor/node stage, cancer subtype, and HER2 receptor status. This generated a sample of n=144 patients, which were included across all four study arms (TH, THP, HP, and TP). Drug models used in our simulations were derived from publicly available data. Our main outcomes of interest (in operable breast cancer) included pCR rate by treatment arm, and pCR by treatment arm stratified by hormone receptor (HR) status (+/-). pCR was assessed per patient based on final simulation volume threshold of 0.1cc. Parameters from the NeoSphere study were mirrored in our SimBioSys Virtual Trial where possible, including treatment time, and
time to surgery. We did not simulate post-surgery treatment. pCR rates were compared between virtual and clinical trials using two-tailed Fisher’s Exact tests.
Results: In non-stratified analyses, we observed pCR rates of 20.1% (TH), 42.4% (THP), 2.8% (HP), and 20.8% (TP) per arm, compared to 23.4% (TH), 47.7% (THP), 16.9% (HP), and 26.7% (TP) in the NeoSphere trial. Our non-stratified results were not significantly different from NeoSphere in the TH (p=0.72, OR=0.86), THP (p=0.69, OR=0.89) or TP (p=0.48, OR=0.78) arms. In our HP arm, we observed lower pCR rates (p=0.002, OR=0.17) than NeoSphere. Importantly, we predicted disease progression in 8.3% of HP patients, closely mirroring NeoSphere HP disease progression (7.4%). In our HR+ analyses, none of our results from the TH, THP, HP, or TP arms were significantly different than NeoSphere (prange = 0.34 to 0.69, ORrange = 0.61 to 1.36). In our HRanalyses, our results were not significantly different from NeoSphere in the TH (p=1.00, OR=1.03), THP (p=0.87, OR=0.92) or TP (p=0.55, OR=1.33) trial arms. However, as in our non-stratified analysis, in our HR-negative sample we observed a significantly lower proportion of pCR in patients treated with HP than in NeoSphere (p=0.0003, OR=0.00).
Conclusion: Using bleeding-edge biophysical simulations, our never-before-possible in silico clinical trial retrospectively predicted regimen efficacy, pCR rate, and disease progression of the NeoSphere clinical trial. Currently, we are investigating the impact of drug delivery, drug sensitivity, metabolism, and spatial heterogeneity on tumor responses within our trial. In the near future, our virtual clinical trials will curtail the negative aspects of real-world clinical trials, while enabling drug developers to garnish more comprehensive insights than ex vivo or in vitro techniques.
