A research team at the University of Cincinnati Cancer Center has developed a breast cancer survival model that combines classical statistical methods with a deep learning framework, an approach the team reports achieved higher accuracy at predicting survival outcomes than comparison models when tested on a publicly available dataset of 1,980 breast cancer patients. The research, published in the Journal of Statistical Theory and Practice, describes the potential for a more robust and accurate survival prediction tool that could give physicians a better basis for personalized care decisions. The work was led by corresponding author Wedad Alatebi, a doctoral student in UC’s Department of Biostatistics, Health Information and Data Science, with collaborators including Anand Seth, Shesh Rai, Syed Ahmad and Marepalli B. Rao.

Each case of breast cancer is unique and depends on many variables, including clinical, biological and genetic factors. Traditional survival models are useful for estimating a patient’s risk and expected outcomes, but the team argues they cannot fully account for how the complex relationships between these factors affect survival prediction for a given patient. “We wanted to explore whether combining statistical methods with deep learning could capture these relationships and improve the overall survival prediction,” Alatebi said. “Deep learning can be described as teaching a computer to recognize complex patterns in patient data.” Seth, a coauthor and research manager in the same department, described the strategy as a “fusion” of multiple statistical approaches that allows a much more precise prediction outcome. The deep learning framework, he said, enables patients who appear similar on the surface to be stratified into more specific risk groups. “That way, you can individualize therapy for a specific group of patients,” Seth said.

A central technical feature of the model is how it handles tumor genetic information. Molecular features can improve survival models, but genetic testing is not always feasible or cost-effective in hospitals worldwide. To address this limitation, the team built simulated gene expression data into the model. “We first transform the clinical information into a better representation of patients’ risk,” Alatebi said. “We then simulate biologically plausible molecular features to explore how integrating molecular and clinical information may improve survival prediction. The model also considers similarities between patients and includes uncertainty in its predictions.” According to Rai, director of the Cancer Center’s Biostatistics and Informatics Shared Resource, the approach is fundamentally about personalization. “We’re either using or simulating the patients’ gene signatures in addition to other features, other demographics, combining all of this together, and from there predicting what could actually happen,” Rai said.

On the real-world evaluation dataset, the hybrid model achieved higher accuracy at predicting survival outcomes than the compared models in the study. “It was exciting to see how much information we could gain by combining different approaches rather than relying on one method alone,” Alatebi said. The team emphasizes that more research is needed to determine the model’s precise accuracy, but Seth noted that even a few percentage points of improvement could give physicians more confidence in treatment strategies and prognoses for their patients. The findings are model performance results on a single publicly available dataset rather than evidence of clinical benefit, and the simulated molecular features remain an approximation of real tumor gene expression.

As next steps, Alatebi and the team plan to further evaluate the model using larger and more diverse patient datasets. Their long-term goal is to develop a tool that could help physicians better identify patient risk and support more personalized treatment and follow-up decisions. “These deep learning-based models are not meant to replace physicians,” Alatebi said. “They are tools that could help physicians understand complex patient information and support their decisions. This combination of statistics, machine learning and cancer research has been exciting for me to help us improve how we understand patient outcomes.”

Subject of Research: A hybrid statistical deep learning framework for breast cancer survival prediction using covariate transformation and simulated gene expression data.

Article Title: AI breast cancer survival model built by University of Cincinnati Cancer Center team could provide improved accuracy and more personalized care

Article References: AI breast cancer survival model built by University of Cincinnati Cancer Center team could provide improved accuracy and more personalized care. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: breast cancer, survival prediction, deep learning, statistical modeling, personalized medicine, gene expression, University of Cincinnati Cancer Center, biostatistics, machine learning, risk stratification, Journal of Statistical Theory and Practice, hybrid model