Breast cancer patients could be spared unnecessary chemotherapy with the use of a new artificial intelligence (AI) tool that helps predict survival risk, say researchers.
The research team says that the model could support pathologists in improving patient prognosis and adapting treatments, as well as “reduce disparities” for patients who are diagnosed in community settings.
Researchers from the Northwestern University Feinberg School of Medicine in Chicago have outlined their work using deep learning to develop the Histomic Prognostic Signature (HiPS) risk score, in a paper published in Nature Medicine. They describe HiPS as “a comprehensive, interpretable scoring of the survival risk incurred by breast tumor microenvironment morphology.”
In the study, HiPS “consistently outperformed pathologists in predicting survival outcomes.” It identified breast cancer patients who are currently classified as high or intermediate risk but who become long-term survivors, which could feasibly allow the duration or intensity of their chemotherapy to be reduced.
“Our study demonstrates the importance of non-cancer components in determining a patient’s outcome,” said corresponding study author Lee Cooper, associate professor of pathology at the Northwestern University Feinberg School of Medicine. “The importance of these elements was known from biological studies, but this knowledge has not been effectively translated to clinical use.”
This year, around 300,000 women in the U.S. will receive a diagnosis of invasive breast cancer. In diagnosis, pathologists provide a grading of cancer by reviewing breast tissue and focusing on the microscopic appearance of cancer cells, which helps determine treatment.
Cooper and colleagues built an AI model to evaluate breast cancer tissue from digital images based on the appearance of both cancerous and noncancerous cells and interactions between them. These are otherwise difficult for pathologists to discern or categorize with the naked eye.
The researchers trained the model using hundreds of thousands of human-generated annotations of cells and tissue structures within digital images of patient tissues provided by an international network of pathologists and medical students.
The team says that the model provides an “objective alternative” that mitigates the variability inherent in manual grading and captures latent features that cannot be reliably graded. The AI system analyzes 26 different properties of a patient’s breast tissue to generate an overall prognostic score. It also generates individual scores for the cancer, immune, and stromal cells to explain the overall score to the pathologist.
In some patients, a favorable prognosis score may be due to properties of their immune cells, whereas for others it may be due to properties of their cancer cells. This information may be used by a patient’s care team in creating an individualized treatment plan.
The authors write, “HiPS consistently outperformed pathologists in predicting survival outcomes, independent of tumor-node-metastasis stage and pertinent variables. This was largely driven by stromal and immune features … We show that the HiPS score is a strong, independent predictor of survival outcomes in nonmetastatic ER+ and HER2+ cancers, and that it is concordant with known epidemiologic and genomic risk profiles.”
The researchers say that adopting the model could allow treatment to be escalated or de-escalated depending on how the microscopic appearance of the tissue changes over time. The model may be able to recognize the effectiveness of a patient’s immune system in targeting the cancer during chemotherapy, which could be used to reduce the duration or intensity of chemotherapy.
For the study, the American Cancer Society created a “unique” dataset of breast cancer patients through their cancer prevention studies. Importantly, say the scientists, the dataset is representative of patients from over 423 U.S. counties, many of whom received a diagnosis or care at community medical centers.
Cooper said the model could “reduce disparities” for patients who are diagnosed in community settings who may not have access to a pathologist who specializes in breast cancer and help a generalist pathologist when evaluating breast cancers.
The scientists will evaluate the model prospectively to validate it for clinical use; they are also working to develop models for specific types of breast cancers, such as triple-negative or HER2-positive. They acknowledge the observational nature of their analysis and the numerous limitations of the study that need to be explored further.