Development and validation of an ensemble machine learning model to predict survival in locally advanced rectal cancer: A multicenter, retrospective study

Comprehensive Summary

Local advanced rectal cancer (LARC) is one of the most common types of malignant and colorectal tumors globally. Treated with neoadjuvant chemoradiotherapy (nCRT) and total mesorectal excision, the variability in patient responses affects long-term prognoses. This led the researchers to develop a machine learning tool that can accurately asses patients' prognoses after receiving nCRT. They developed the model using 1119 eligible LARC patients who were diagnosed with LARC (confirmed by pathology) and then received complete nCRT followed by radical surgery (TME). All patients went to appointments comprised of physical checkups and biochemical testing at 3-month intervals during years 1–2, 6-month intervals during years 3–5, and once a year subsequently. The researchers found their prognostic model GRM was extremely accurate with a high level of agreement between the model's predicted outcomes and actual results. Moreover, the model used key predictors such as postoperative systemic inflammatory markers to better predict the body's overall response to the tumor and its treatment.

Outcomes and Implications

The model allows for more precise risk stratification after nCRT, allowing patients to be divided in low, intermediate, and high-risk groups to allow for individualized prognoses. In addition, the model allows for better postoperative treatment plan, especially for high-risk patients. The model can also guide how aggressive patients are treated and monitored after surgery to improve early detection of recurrence in high-risk patients and patient burden reduction in low-risk patients.