pCR Score: A Novel Prognostic Method to Estimate the Predictive Probability of pCR in Early-Stage Breast Cancer Patients.

Background: A gap in personalized medicine exists in the absence of a test to assess the probability of pathological complete response (pCR) for early-stage breast cancer patients. We have previously shown that our TumorScope platform, which utilizes pretreatment standard-of-care (SOC) diagnostic data to model in-vivo biologic interactions, can reliably predict a binary outcome of pCR for a given patient with any physician-chosen SOC chemotherapeutic regimen. To drive further utility, we now investigate a pCR score as a continuous outcome (0-100) to establish a prognostic system that evaluates the predictive probability that a patient will achieve pCR with any SOC NAT regimen. Furthermore, we sought to establish thresholds (low, low-mid, high-mid, high) and corroborate the score’s utility in the context of 5-year EFS.

Methods: The total cohort consisted of 1050 patients from seven institutions. The pCR Score was calibrated with a training cohort consisting of 665 patients of all breast cancer subtypes. We used a logistic regression model to calculate the probability of pCR for each individual patient based on NAT regimen, clinical and multiscale simulation predictors. For all breast cancer subtypes, the baseline model included the following clinical variables: age, race, grade, T stage, N stage. For hormone receptor-positive (HR+) and HER2+ subtypes, ER, PR and HER2 status were included. The TumorScope model included the clinical variables and simulation derived features including modeled tumor volume at start and end of therapy (Vs + Ve). We compared the TumorScope model (clinical variables + simulation variables) to the model consisting solely of clinical variables. We then calculated the prognostic ability of the pCR score to corroborate EFS in an independent testing cohort of 385 patients. Patients were stratified according to the likelihood of pCR into high, high-mid, mid-low or low probability and correlated to the 5-year event-free survival (EFS) for all patients and per breast cancer subtype.

Results: TumorScope showed superiority in predicting pCR probability in all breast cancer subtypes calculated as the number of times that TumorScope outperformed the clinical model: HR+/HER2+ (79.6%, n= 162, p= 8×10-13), HR-/HER2+ (72.2%, n= 115, p= 2×10-6), HR+/HER2- (70.5%, n= 233, p=6×10-8), and TNBC (96.4%, n=253, p= 3×10-12). In the overall analysis, patient EFS at 5-year follow-up according to pCR score was as follows: low pCR score (n=76) 61% EFS, low-mid pCR score (n=126) 75% EFS, high-mid pCR score (n=86) 82% EFS, and high pCR score (n=97) EFS 96%. EFS was also calculated for each breast cancer subtype stratified into high or low pCR score; high pCR score correlated with the best 5-year EFS and low score correlated with poor 5-year EFS in alignment with empiric expectations. Of the different subtypes, HR+/HER2+high had the best 5-year EFS (~85-90%) and TBNClow had the worst EFS (55-60%).

Conclusion: The TumorScope pCR (continuous) score system offers an enriched test output to further refine prognostic capability beyond a conventional binary (yes/no) result and permits risk-stratification of patients into predictive categories. Here, we validate the TumorScope pCR score as a reliable metric   using a large, multicenter cohort. The TumorScope pCR score can be correlated with 5-year EFS in an easy-to-understand format, concurrent with currently available prognostic tests in the market.