Machine Learning-Based Prediction of Distant Recurrence in Invasive Breast Carcinoma Using Clinicopathological Data:
Shrey S Sukhadia1,2, Kristen E Muller2, Adrienne A Workman2
1Centre for Genomics and Personalised Health, Queensland University of Technology, Brisbane, QLD 4059, Australia.
Cancers
|August 12, 2023
Summary
Machine learning models can predict invasive breast carcinoma (IBC) distant recurrence using pre- and post-therapy tumor and lymph node staging. Tumor response to neoadjuvant therapy was the most significant predictor.
Area of Science:
- Oncology
- Biostatistics
- Machine Learning
Background:
- Breast cancer is the most common cancer globally, with approximately 30% of cases recurring distantly after treatment.
- Distant recurrence is more prevalent in specific subtypes like invasive breast carcinoma (IBC).
- Current prediction methods for IBC distant recurrence rely on clinicopathological measurements but do not integrate machine learning with pre- and post-therapy evaluations.
Purpose of the Study:
- To evaluate the efficacy of machine learning (ML) models in predicting distant recurrences in invasive breast carcinoma (IBC) patients.
- To combine clinicopathological evaluations of IBC tumors (pre- and post-therapy) with ML techniques for enhanced predictive accuracy.
- To identify key clinicopathological measurements that significantly impact the prediction of distant recurrences.
Main Methods:
- Trained and tested four ML models (random forest, C-support vector classifier, multilayer perceptron, logistic regression) using clinicopathological data from 144 training and 17 testing patients.
- Validated the best-performing model on an external dataset of 8 patients.
- Included pathological staging (tumor and lymph nodes), therapy response (imaging), and adjuvant therapy status as predictive features.
Main Results:
- The random forest model demonstrated superior performance, achieving an Area Under the Curve (AUC) of 1.0 in the testing set and 0.75 in the validation set (p < 0.002).
- The model showed cross-institutional portability and validity across Duke University and Dartmouth Hitchcock Medical Center datasets.
- Tumor response to neoadjuvant therapy (via imaging and pathology, including tumor and node staging) was the most critical factor in predicting distant recurrences.
Conclusions:
- Machine learning models, particularly random forest, can effectively predict distant recurrences in invasive breast carcinoma (IBC) patients.
- Integrating pre- and post-therapy clinicopathological data with ML offers a promising approach for personalized risk assessment in IBC.
- Tumor response to neoadjuvant therapy is a key determinant for predicting distant recurrence in IBC, highlighting the importance of early treatment response evaluation.
