Titles and Affiliations
Assistant Attending, Computational Oncology Service
Department of Epidemiology and Biostatistics
American Association for Cancer Research
Research area
Developing AI tools with more accurate predictive and prognostic capabilities for racially underrepresented breast cancer patients.
Impact
Artificial intelligence (AI) offers a unique opportunity to improve therapeutic outcomes for cancer patients. Whereas traditional clinical models for outcome prediction have focused on a narrow set of interpretable biomarkers such as tumor size, grade, hormone receptor status, gene mutations, or overexpression of a small number of genes, AI-based approaches can integrate millions of data points from multiple data modalities including radiology images, pathology slides, comprehensive genomic sequencing and clinical variables for improved prediction and prognostication. Unfortunately, most of these new tools continue to be developed in a race-agnostic manner and do not account for race-specific differences in tumor biology. AI tools trained and validated using racially homogeneous cohorts exacerbate existing racial disparities in clinical outcomes, where age-adjusted breast-cancer mortality is about 40 percent higher among Black women than among non-Hispanic White women.
Progress Thus Far
Dr. Sanchez-Vega and his team analyzed data from nearly 1,200 breast cancer patients across a racially and clinically diverse cohort to build computational tools that predict which patients will fully respond to chemotherapy before surgery. Using advanced machine learning methods, they found important differences in how clinical, genomic, pathology, and radiology data predict treatment response depending on race and tumor subtype. For example, Hispanic patients were diagnosed at a younger age on average, and predictors performed differently across groups, such as clinical data being especially effective for Hispanic patients with hormone receptor (HR)-negative/HER2-positive tumors and genomic data performing particularly well for Asian patients across subtypes. The study also revealed variation in response rates by both tumor type and race—highest among Asian patients and lowest among Black patients. These findings provide critical insight into r inherent bias in predictive models and the need to ensure that new AI tools will work equitably for all breast cancer patients.
What’s next
In the coming year, Dr. Sanchez-Vega and his team will focus on developing and refining new AI algorithms that bring together multiple types of patient data—such as clinical, genomic, pathology, and imaging information—to better predict which patients will respond to chemotherapy before surgery. The researchers are testing different strategies to combine data to ensure these models work equally well across patients of different racial backgrounds, and ultimately aim to create more robust, inclusive predictors that could ultimately guide more personalized and equitable breast cancer treatment.
Biography
Francisco Sanchez-Vega, PhD is an Assistant Attending in the Computational Oncology Service of the Department of Epidemiology and Biostatistics at Memorial Sloan Kettering. He has a PhD in Applied Mathematics and Statistics with an area of specialization in Computational Medicine from The Johns Hopkins University. His research focuses on translational applications of machine learning, statistical modeling and computational methods to the field of cancer genomics and precision oncology. His group is also interested in the use and implementation of novel computational approaches for multimodal integration of genomic sequencing data and orthogonal sources of biological and clinical information.
