AI Tool OncoNPC Predicts Cancer Origin, Guides Treatment: Study
A team of researchers at the Dana-Farber Cancer Institute has developed an AI-based tool called OncoNPC that can predict the primary source of a patient’s cancer using tumor gene sequencing data. The findings, published in Nature Medicine, show that OncoNPC could significantly improve outcomes in cases where the origin of the tumor cannot be determined using traditional diagnostic methods.
In approximately 3-5% of cancer cases, the primary source of the tumor remains unknown, leading to a diagnosis of cancer of unknown primary (CUP). These patients often face limited treatment options as most therapies are approved for specific types of cancer. This patient group has dismal outcomes, says senior author Alexander Gusev, a researcher at Dana-Farber.
The team trained and validated the AI model using the medical records of 36,445 patients with known primary tumors. The dataset included tumor genetic sequencing data and clinical information for each patient. The researchers opted for an interpretable machine learning model to make the reasoning behind predictions more transparent to clinicians and increase trust in the tool.
OncoNPC, the resulting classifier model, accurately predicted the origin of about 80% of tumors with known types, including metastatic tumors, using a subset of cases not used for training. The model made high-confidence predictions in 65% of the cases, with a 95% accuracy rate. When applied to a separate database of 971 CUP tumors, OncoNPC predicted the origin of the tumors with high confidence in 41.2% of the cases.
To validate these predictions, the researchers analyzed inherited germline risks of cancer and found that they aligned with the model’s predictions. They also examined patient outcomes and found that those who received treatments aligned with the predicted primary tumor site had longer survival compared to those who did not.
This could open the door to more precision treatment for these patients, says Gusev.
While the tool has been studied using retrospective data, further testing in a clinical trial will be necessary to determine its effectiveness in improving patient outcomes. Gusev and his team plan to expand OncoNPC by incorporating additional diagnostic information, such as pathology results, in future iterations. They also aim to collaborate with community cancer centers to assess how OncoNPC can complement existing diagnostic methods, as cases of CUP may be more common in smaller centers with limited resources.
The potential benefits of OncoNPC are promising, as it could provide valuable diagnostic information and help guide treatment decisions for patients with CUP. By leveraging AI technology and genetic sequencing data, doctors may be able to offer more targeted treatments, ultimately leading to improved outcomes for these challenging cases.
