A new paper from the Bioethics Research Center at WashU Medicine, published in Frontiers in Neurology and led by Tristan McIntosh, PhD, an assistant professor in the Division of General Internal Medicine, offers one of the first in-depth looks at how the people most affected by glioblastoma — patients, their caregivers, and the neurosurgeons who treat them — feel about using machine learning (ML) to help predict prognosis and guide surgical decisions.
Glioblastoma (GBM) is an aggressive brain cancer where treatment decisions often hinge on difficult judgment calls about prognosis and quality of life. As ML tools increasingly enter clinical practice to help synthesize complex patient data, questions about how these tools should be used and trusted have outpaced the research on their implementation and impact on important end-users like patients, families, and healthcare providers.
To close this gap, the research team conducted interviews with GBM patients, their caregivers, and neurosurgeons about a specific ML model designed to predict GBM prognosis and support surgical decision-making.
What they found
All three groups agreed on a key strength: the model’s ability to process large volumes of patient data in ways that could improve communication between patients and their care teams, particularly when navigating treatment planning or end-of-life decisions.
But participants also raised meaningful concerns, including:
- Worry about inaccuracies or hidden biases in the model’s output
- Discomfort with the idea of AI replacing a neurosurgeon’s clinical judgment
- Concerns that the technology is still early in development
- Fear that ML-informed prognoses could be delivered in ways that cause confusion or take away patients’ sense of hope
Why it matters
The study underscores a central theme in the advancement of medical technologies: building trustworthy artificial intelligence for medicine requires listening to the people who will use and be affected by it, not just optimizing for technical performance.
ML models hold real promise for supporting clinical decision-making, but McIntosh notes that biases in training data and over-reliance on automated predictions carry real risks for patient outcomes if these tools are deployed without careful attention to stakeholder perspectives.
“This research adds an important, human-centered voice to the fast-moving conversation about artificial intelligence in neuro-oncology and offers a roadmap for what responsible implementation should consider,” McIntosh said.