image: An MSK researcher works in the lab.
Credit: Memorial Sloan Kettering Cancer Center
New research from Memorial Sloan Kettering Cancer Center (MSK) investigates whether introducing new mutations could make immunotherapy more effective against some cancers; shows that people in their 90s who underwent lung cancer surgery had positive outcomes; shares lessons on the responsible governance of artificial intelligence in oncology; studies AI-assisted biomarker assessment in lung cancer; and demonstrates a cancer-trained large language model has strong predictive value.
Intentionally introducing new mutations could make immunotherapy more effective for some cancers
Previous clinical research has shown that tumors with specific gene mutations or changes are more sensitive to immunotherapy, a treatment that helps the immune system recognize and attack cancer cells. Tumors that are mismatch repair-deficient have gene mutations that prevent cells from fixing errors in their DNA, leading to an accumulation of damage that triggers the body’s immune response. Most solid tumors don’t respond to immunotherapy, but a breakthrough clinical trial at MSK in 2022 showed that people with mismatch repair-deficient rectal cancer saw their tumors disappear after treatment with PD-1 blockade immunotherapy. Now, MSK investigators are developing a new approach to induce these mutations in resistant cancer cells, a strategy that could lead to new treatment options.
Research led by MSK’s Benoît Rousseau, MD, PhD, Neil Segal, MD, PhD, and Luis Diaz, MD, from the Division of Solid Tumor Oncology in the Department of Medicine, showed that combining temozolomide and cisplatin can induce mismatch repair deficiency in colon cancer cells in mice, making them sensitive to PD-1 blockade. Building on these findings, Dr. Rousseau and Dr. Diaz led a clinical trial with 18 participants with metastatic colorectal cancer to see whether the same drug combination, along with PD-1 blockade, could induce mismatch repair deficiency and shrink tumors. While no responses were observed, mutational profiles compatible with mismatch repair deficiency developed in patients’ tumor DNA and correlated with improved outcomes, suggesting the potential for this strategy to make resistant tumors more responsive to immunotherapy. Read more in Cancer Cell.
Dr. Rousseau and Dr. Diaz also contributed to another recent study in Cancer Cell, alongside colleagues at the University of Torino, that evaluated this drug combination in colorectal and breast cancer cells in mice.
Data shows people in their 90s who underwent lung cancer surgery at MSK had positive outcomes
As people continue to live longer, more individuals in their 90s are being diagnosed with lung cancer. (At MSK, the number of nonagenarians seen in our Thoracic Oncology clinics increased nine-fold from 2001 to 2021.) Surgical removal of the part of the lung bearing lung cancer is a potential treatment option, but nationally, a lack of information about short- and long-term outcomes for older patients has made it challenging for individuals and their doctors to make informed decisions about surgery.
A new retrospective study looked at outcomes among 118 MSK patients in their 90s who were diagnosed with non-small cell lung cancer during that period, 17 of whom underwent surgery.
The patients who elected to have surgery had a median survival of 43 months, which is close to the 53-month life expectancy for people in the age group without cancer. There were no major complications, their cognitive functioning remained stable after surgery, and no patients died within 90 days of being operated on.
“These findings reflect multidisciplinary care of these patients and can help patients, families, and doctors make informed decisions when considering whether a patient in their 90s should undergo lung cancer surgery,” says thoracic surgeon Prasad Adusumilli, MD, the corresponding author of the study. Read more in Lung Cancer.
Responsible governance is critical to the success of AI in oncology
As artificial intelligence (AI) becomes more integral to oncology — and to healthcare more generally — the need for effective governance models is becoming increasingly important. Recently, a team from MSK published a study reporting on the first year of our comprehensive cancer center’s responsible AI governance model for clinical programs, operations, and research — one of the first such published reports in the field.
The study covers the registration and monitoring of 26 AI models (including large language models), two ambient AI pilots, and a review of 33 nomograms (a graphical calculation tool).
“Our analysis shows governance and quality assurance of AI models are feasible at scale, but require key components for success,” says MSK hospitalist Peter Stetson, MD, MA, the paper’s corresponding author. The article outlines novel management tools, including for risk assessment and lifecycle management, as well as two case studies illustrating lessons learned.
“Taken together, we hope that these real-world insights will be useful to others in oncology as these technologies continue to evolve and shape the future of cancer care,” says senior study author Anaeze Offodile II, MD, MPH, MSK’s Chief Strategy Officer. Read more in npj Digital Medicine.
Real-world test shows value of AI-assisted biomarker assessment in lung cancer
AI holds promise for helping to diagnose cancer from pathology slides — but real-world applications are still being proven.
A team from MSK, the Icahn School of Medicine at Mount Sinai, and their collaborators are demonstrating its effectiveness for assessing EGFR mutations in lung cancer. Mutations in the EGFR gene can make cells grow too much, leading to the development of cancer.
Current PCR testing protocols are fast but less accurate, and next-generation sequencing tests require additional tissue samples.
To assess their computational tool — dubbed EAGLE for EGFR AI Genomic Lung Evaluation — the researchers assembled a clinical dataset of more than 8,000 lung cancer slides, the largest international dataset to date.
They found that the AI-assisted analysis cut the number of molecular tests needed by more than 40% while maintaining current clinical standards for performance.
“Our findings demonstrate the real-world clinical utility of this computational biomarker,” says MSK pathologist Chad Vanderbilt, MD, a co-senior author of the study alongside Thomas Fuchs, DrSc, of Mount Sinai (now chief AI officer for Eli Lilly). The project was supported by the Warren Alpert Center for Digital and Computational Pathology at MSK.
First author Gabriele Campanella, PhD, an assistant professor in the Windreich Department of AI and Human Health at Mount Sinai adds, “The insights from this study additionally provide a roadmap for integrating AI into clinical pathology and highlight the potential of computational biomarkers to improve precision oncology.” Read more in Nature Medicine.
Cancer-trained LLM shows strong predictive value
Artificial intelligence, including large language models (LLMs), holds promise for transforming cancer care. However, a lack of the specialized knowledge required to interpret complex oncology data has limited their adoption. To address this, a team of researchers from MSK and the University of California, San Francisco (UCSF), has created Woollie — an open-source, cancer-specific LLM trained on real-world data that can help doctors understand and predict cancer progression.
Woollie was developed using thousands of radiology reports from MSK across five major cancer types: lung, breast, prostate, pancreatic, and colorectal cancers. The model was then independently validated on data from UCSF to test its real-world performance.
“Woollie successfully brings together oncology-specific knowledge with the advanced reasoning and conversational abilities of a modern LLM,” says Anyi Li, PhD, Chief of Computer Service in the Department of Medical Physics and a senior author of the study. “And our analysis also shows both the feasibility and importance of cross-institutional validation, which increases general, real-world applicability.”
The team, which included 17 researchers from across MSK, analyzed nearly 40,000 impression notes from over 4,000 patients. These notes provide information about tumor location and size, as well as changes over time — critical information for assessing and guiding treatment.
An analysis of Woollie showed strong capabilities, achieving a score of 97 out of 100 for predicting cancer progression from MSK data, and 98 specifically for pancreatic cancer. On UCSF data, it achieved an overall score of 88, and 95 for lung cancer. (A score of 50 would be the equivalent of a coin flip, and 100 would indicate perfect prediction.)
Beyond guiding treatment, Woollie could also help uncover broader insights into cancer biology, such as metastatic pathways and patterns of disease progression. “This represents an important step toward AI tools that can reliably support oncologists and researchers,” Dr. Li adds. Read more in npj Digital Medicine.