Machine Learning Frameworks to Predict Neoadjuvant Chemotherapy Response in Breast Cancer Using Clinical and Pathological Features

PURPOSE Neoadjuvant chemotherapy (NAC) is used to treat locally advanced breast cancer (LABC) and high-risk early breast cancer (BC). Pathological complete response (pCR) has prognostic value depending on BC subtype. Rates of pCR, however, can be variable. Predictive modeling is desirable to help identify patients early who may have suboptimal NAC response. Here, we test and compare the predictive performances of machine learning (ML) prediction models to a standard statistical model, using clinical and pathological data. METHODS Clinical and pathological variables were collected in 431 patients, including tumor size, patient demographics, histological characteristics, molecular status, and staging information. A standard multivariable logistic regression (MLR) was developed and compared with five ML models: k-nearest neighbor classifier, random forest (RF) classifier, naive Bayes algorithm, support vector machine, and multilayer perceptron model. Model performances were measured using a receiver operating characteristic (ROC) analysis and statistically compared. RESULTS MLR predictors of NAC response included: estrogen receptor (ER) status, human epidermal growth factor-2 (HER2) status, tumor size, and Nottingham grade. The strongest MLR predictors of pCR included HER2+ versus HER2− BC (odds ratio [OR], 0.13; 95% CI, 0.07 to 0.23; P < .001) and Nottingham grade G3 versus G1-2 (G1-2: OR, 0.36; 95% CI, 0.20 to 0.65; P < .001). The area under the curve (AUC) for the MLR was AUC = 0.64. Among the various ML models, an RF classifier performed best, with an AUC = 0.88, sensitivity of 70.7%, and specificity of 84.6%, and included the following variables: menopausal status, ER status, HER2 status, Nottingham grade, tumor size, nodal status, and presence of inflammatory BC. CONCLUSION Modeling performances varied between standard versus ML classification methods. RF ML classifiers demonstrated the best predictive performance among all models.

Variation in breast cancer grading in 1,636 resections assessed using control charts and in silico kappa

Objective Assess interpretative variation in Nottingham grading using control charts (CCs) and in silico kappa (ISK). Methods In house invasive breast cancer cases (2011–2019) at two institutions with a synoptic report were extracted. Pathologist interpretative rates (PIRs) were calculated and normed for Nottingham grade (G) and its components (tubular score (TS), nuclear score (NS), mitotic score (MS)) for pathologists interpreting >35 cases. ISKs were calculated using the ordered mutually exclusive category assumption (OMECA) and maximal categorical overlap assumption (MCOA). Results The study period included 1,994 resections. Ten pathologists each assessed 38–441 cases and together saw 1,636; these were further analyzed. The PIR medians (normed ranges) were: G1:24%(18–27%), G2:53%(43–56%) and G3:26%(19–33%). The MCOA ISK and the number of statistical outliers (p< 0.05/p< 0.001) to the group median interpretive rate (GMIR) for the ten pathologists was G1: 0.82(2/0 of 10), G2: 0.76(1/1), G3: 0.71(3/1), TS1: 0.79(1/0), TS2: 0.63(5/1), TS3: 0.66(5/1), NS1: 0.37(5/4), NS2: 0.60(4/3), NS3: 0.59(4/4), MS1: 0.78(3/1), MS2: 0.78(3/1), MS3: 0.77(2/0). The OMECA ISK was 0.62, 0.49, 0.69 and 0.71 for TS, NS, MS and G. Conclusions The nuclear score has the most outliers. NS1 appears to be inconsistently used. ISK mirrors trends in conventional kappa studies. CCs and ISK allow insight into interpretive variation and may be essential for the next generation in quality.

Understanding risk factors in grade 3 breast cancer in NZ

Background Breast cancer is the most common cancer in New Zealand women, accounting for approximately 3000 new registrations per year, affecting one in nine women and resulting in more than 600 deaths annually. This study analysed data of patients selected with prognostic factor of Nottingham grade 3 tumours over a specified five- year period. These represent a heterogeneous group of cancers with variable survival rates. Method All women diagnosed with Nottingham grade 3 invasive breast cancer between 1 st January 2011 to 31 st December 2015, from four Breast Cancer Registries in New Zealand (Auckland, Waikato, Christchurch, and Wellington) were studied. Results Applying Fine-Gray analyses, the study of 2,493 women found that subjects in the older age group (>70 years) had a higher five-year mortality risk (SHR: 1.74 to 2.25, p: 0.053 to <0.001). Analysis of hormonal receptors showed that tumours with hormonal profile ER-positive, PR negative and ER-negative, PR negative subjects were at increased mortality risk (SHR: 3.56, p: <0.001) and (SHR: 2.67, p: <0.001) respectively. Molecular subtypes TNBC and Luminal B subjects were at increased risk of five-year mortality (SHR: 3.01 and 3.35 respectively, both p: <0.001). HER2 enriched subjects, were at elevated risk (SHR: 1.66, p: 0.11). Women identifying as NZ European ethnicity were at elevated risk of mortality overall (SHR: 1.70, p: 0.11), and they presented with the highest CIF across ethnicities. The NZ Europeans represented the largest proportion of HER2 enriched and TNBC subjects; however, Pacific Islanders experienced the highest HER2 CIF. Conclusion The survival rates for grade 3 breast cancer vary across the selected prognostic factors and ethnicity. Although grade 3 breast cancer is considered as high grade heterogeneous cancer, this study showed that not every patient has a poor outcome. NZ Europeans are worst affected followed by Pacific Islanders. Biology of the cancer and ethnicity needs to be looked at as a possible factor associated with this disease for survival differences. The results of this study make an initial contribution to the understanding of high-grade malignancy and other prognostic factors must be included in order to get a better understanding of survival differences.

CACTUS: cancer image annotating, calibrating, testing, understanding and sharing in breast cancer histopathology

Objective Develop CACTUS (cancer image annotating, calibrating, testing, understanding and sharing) as a novel web application for image archiving, annotation, grading, distribution, networking and evaluation. This helps pathologists to avoid unintended mistakes leading to quality assurance, teaching and evaluation in anatomical pathology. Effectiveness of the tool has been demonstrated by assessing pathologists performance in the grading of breast carcinoma and by comparing inter/intra-observer assessment of grading criteria amongst pathologists reviewing digital breast cancer images. Reproducibility has been assessed by inter-observer (kappa statistics) and intra-observer (intraclass correlation coefficient) concordance rates. Results CACTUS has been evaluated using a surgical pathology application—the assessment of breast cancer grade. We used CACTUS to present standardized images to four pathologists of differing experience. They were asked to evaluate all images to determine their assessment of Nottingham grade of a series of breast carcinoma cases. For each image, they were asked for their overall grade impression. CACTUS helps and guides pathologists to improve disease diagnosis with higher confidence and thereby reduces their workload and bias. CACTUS can be useful for both disseminating anatomical pathology images for teaching, as well as for evaluating agreement amongst pathologists or against a gold standard for evaluation or quality assurance.

Canine Mammary Tumors: Comparison of Classification and Grading Methods in a Survival Study

Histopathology remains the cornerstone for diagnosing canine mammary tumors (CMTs). Recently, 2 classification systems (the World Health Organization [WHO] classification of 1999 and the proposal of 2011) and 2 grading methods based on the human Nottingham grade have been used by pathologists. Despite some evidence that the histological subtype and grade are prognostic factors, there is no comprehensive comparative study of these classification and grading systems in the same series of CMTs. In this study, the 2 classifications and the 2 grading methods were simultaneously applied to a cohort of 134 female dogs with CMTs. In 85 animals with malignant tumors, univariable and multivariable survival analyses were performed. Using the 2 systems, the proportion of benign (161/305, 53%) and malignant (144/305, 47%) tumors was similar and no significant differences existed in categorization of benign tumors. However, the 2011 classification subdivided malignant tumors in more categories—namely, those classified as complex, solid, and tubulopapillary carcinomas by the WHO system. Histological subtype according to both systems was significantly associated with survival. Carcinomas arising in benign tumors, complex carcinomas, and mixed carcinomas were associated with a better prognosis. In contrast, carcinosarcomas and comedocarcinomas had a high risk of tumor-related death. Slight differences existed between the 2 grading methods, and grade was related to survival only in univariable analysis. In this cohort, age, completeness of surgical margins, and 2 index formulas adapted from human breast cancer studies (including tumor size, grade, and vascular/lymph node invasion) were independent prognostic factors.