NEW YORK, July 12 — For all the progress in breast cancer treatment over the past few decades, the decision about how aggressive to be with therapy remains a clinical knife-edge. Too much treatment means chemotherapy side effects for patients who might never need it. Too little risks a recurrence that could have been prevented.
The dilemma is not new, but the tools to resolve it have been blunt. That is the technical reality that researchers at New York University are now trying to sharpen with a new artificial intelligence tool designed to provide more precise estimates of recurrence risk.
The work aims to give oncologists a faster, more accessible read on which patients genuinely need intensive treatment and which can safely avoid it. The detail that matters is that breast cancer is not a single disease. Krzysztof J.
Geras, one of the researchers involved, noted that “breast cancer is not a single disease,” underscoring the variability in patient outcomes. The NYU tool uses machine learning to generate risk estimates that the researchers say can match the performance of genetic tests.
Genetic tests are currently used for assessing recurrence risk, but they are expensive, take time, and are not universally available. The AI approach could change that calculus. If the tool proves out, it would offer a quicker and more accessible route to the same kind of risk stratification, potentially reducing both overtreatment and undertreatment.
Overtreatment is not a niche problem. Many patients receive chemotherapy that may not be necessary, exposing them to side effects without benefit. The flip side is equally real: some patients who need more aggressive treatment than they receive may not get it because their risk is underestimated.
The NYU team’s work is part of a broader push to bring precision medicine to oncology, where the goal is to match treatment intensity to actual risk rather than to population averages. The tool’s ability to match genetic test results from routine clinical data is the piece that could make it scalable.
As ever, the path from a research tool to a clinical standard is long. The NYU model will need validation in larger, diverse patient populations before it changes practice. But the direction of travel is clear: better data, better models, fewer unnecessary treatments.
What to watch next is whether the tool can be integrated into clinical workflows without adding friction. An AI that offers a faster, more accessible assessment of recurrence risk, without requiring a genetic test that may take time, is the kind of technical fix that could actually make a difference. Consult your doctor for medical advice.
