乌鸦传媒視頻

Use of an artificial intelligence (AI)-based technique known as optimal policy trees (OPT) provided a potential solution to the ongoing debate surrounding the optimal margin width for patients with resectable colorectal cancer (CRC) liver metastases, according to a cohort study.

Among 386 patients with KRAS-variant tumors who underwent hepatectomy in internal training and testing cohorts, the area under the curve (AUC) of the random forest (RF) counterfactual model was 0.76, the "highest ever reported," said Georgios Antonios Margonis, MD, PhD, of Memorial Sloan Kettering Cancer Center in New York City, and colleagues.

In an external cohort of 375 patients with KRAS-variant tumors, use of the new RF model had an AUC of 0.78, "which allowed for a reliable external validation of the OPT-based optimal margin," confirming that the optimal margin width of 7 mm was associated with prolonged survival, the authors noted in .

"The more precise we can become with margins, the better outcomes for patients, as it will further decrease the morbidity of oncological procedures," said Allen Kamrava, MD, of Cedars-Sinai Medical Center in Los Angeles, who was not involved in this study.

Currently there is no consensus on the optimal margin width for CRC liver metastases, Margonis and team noted. Because it may impact surgical outcomes, researchers have debated whether the optimal margin width should be individualized based on patient characteristics.

The OPT technique, which was created by a team at the Massachusetts Institute of Technology, allows for "clinical interpretability" and assigns several different treatment recommendations to patient subgroups, the authors explained.

Margonis and colleagues examined data on 965 patients who underwent curative-intent hepatectomy for CRC liver metastases from January 2000 through December 2017 at Johns Hopkins Hospital in Baltimore, Memorial Sloan Kettering Cancer Center, and Charité-University of Berlin.

Additionally, 878 patients who were retrospectively identified from institutional databases in the U.S., Austria, Argentina, France, Japan, and Norway made up the external cohort. Both the internal and external cohorts were divided into those with KRAS-variant (n=386 and n=375, respectively) or wild-type tumors (n=579 and n=503).

For wild-type tumors, the researchers found an AUC of the RF counterfactual model of 0.79 in the training data set but 0.57 in the testing data set. "The AUC of 0.57 in the testing data set indicates poor discrimination of the model, whereas the higher AUC in the training data set suggests overfitting of the model," the authors noted. "We could not be confident about its predictions and elected not to further analyze patients with wild-type tumors."

Of those with KRAS-variant tumors, 52-58% were men, with a median age of 58 in the internal cohort and 61 in the external cohort. Among both cohorts, those with KRAS-variant tumors had a median of two CRC liver metastases, a margin width of 3-5 mm, and 41-53% had a right-sided primary tumor. Nearly all had advanced cancer (stage T3-T4).

The OPT technique was built on RF predictive regression models and was used to infer margin width in accordance with the maximal decrease in 5-year mortality probability. Margonis and team validated the RF model by calculating its AUC in the testing cohort, and the OPT component was validated by the Shapley Additive Explanations (SHAP), a game theory-based approach.

Among subgroups generated by the model, the recommended optimal margin widths for subgroups A, B, C, and D were 6, 7, 12, and 7 mm, respectively. These optimal margin widths were largely confirmed by SHAP analysis: 6-7 mm (subgroup A), 7 mm (B), 7-8 mm (C), 7 mm (D).

"Although these relatively wide margins should be attempted when possible, our findings should not be viewed as a recommendation to change operative indications, as submillimeter margin clearance is associated with much of the survival improvement," Margonis and colleagues concluded.

They acknowledged that "the surgical margin may also depend on variables, such as anatomical distribution of metastases and tumor growth pattern, which may have independent prognostic importance. Thus, it is possible that the observed differences in survival may also depend on other factors that we do not control for or know about."

Comment