"The clinical and biological features proposed here in conjunction with ML can improve the interpretation of CRC mechanisms and predict patient survival."
BUFFALO, NY – December 17, 2025 – A new research paper was published in Oncotarget (Volume 16) on December 15, 2025, titled " Machine learning-based survival prediction in colorectal cancer combining clinical and biological features ."
In this study, led by Lucas M. Vieira from the University of Brasília and the University of California San Diego , researchers used machine learning to predict survival in patients with colorectal cancer. They built a model by combining biological markers with clinical data. This approach could help improve prognosis and guide treatment strategies for one of the world's most common and deadly cancers.
The team analyzed data from over 500 patients, using clinical details such as age, chemotherapy status, and cancer stage, along with molecular features like gene expression and microRNAs. Their goal was to improve how clinicians identify high-risk patients and make outcome predictions more precise. Researchers evaluated three different patient data scenarios using different machine learning techniques. The best-performing was an adaptive boosting model, which achieved 89.58% accuracy. This approach showed that integrating clinical and biological data led to significantly better predictions than using either data type alone.
Among the biological markers, the gene E2F8 was consistently influential in all patient groups and is known to play a role in tumor growth. Other important markers included WDR77 and hsa-miR-495-3p, which are also associated with cancer development. Key clinical predictors included cancer stage, patient age, lymph node involvement, and whether chemotherapy was administered.
"The proposed method combines biological and clinical features to predict patient survival, using as input data from patients from the United States, available in the TCGA database."
Unlike earlier models that relied on either clinical or molecular data alone, this study demonstrates the added value of combining both. Ensemble methods, which merge multiple learning algorithms, provided more stable and consistent results across all patient groups tested.
These research findings could lead to new tools that help clinicians better predict how a patient's disease might progress or respond to treatment. The study also highlights the importance of collecting complete clinical information, such as lifestyle factors, which were missing from the dataset but could enhance future predictions.
Overall, the study demonstrated how machine learning can support more accurate and personalized survival predictions in colorectal cancer. It also points to potential future research on markers like E2F8, which may be useful for monitoring or targeted therapy.
DOI: https://doi.org/10.18632/oncotarget.28783
