Machine learning-based clinical decision support tool for advanced ESCC in the immunotherapy era: a multi-center...

Area of Science:

• Oncology
• Machine Learning in Medicine
• Clinical Decision Support

Background:

• Advanced esophageal squamous cell carcinoma (ESCC) presents a significant challenge in treatment selection.
• Personalized prognostic assessment is crucial for optimizing treatment strategies in treatment-naïve advanced ESCC patients.

Purpose of the Study:

• To develop and validate clinical decision support tools for predicting overall survival (OS) in treatment-naïve advanced ESCC patients.
• To leverage machine learning for enhanced prognostic accuracy using routine clinical variables.

Main Methods:

• A cohort of 1,048 patients receiving first-line immunochemotherapy was analyzed.
• Feature selection was performed using Boruta, followed by model development with a random survival forest (RSF) algorithm.
• Model performance was evaluated using time-dependent area under the receiver operator curve (tAUC), concordance index (C-index), Brier score, calibration plots, and decision curve analysis (DCA).

Main Results:

• The Boruta-RSF model demonstrated superior predictive performance, with tAUCs for 6-, 12-, and 18-month OS of 0.886, 0.775, and 0.772, respectively, in the training set.
• The model's superiority was confirmed in an external validation cohort (all tAUCs > 0.750).
• A publicly accessible web calculator was developed for individualized OS prediction, showing excellent calibration and clinical utility.

Conclusions:

• The Boruta-RSF model effectively predicts prognosis for advanced ESCC patients using readily available clinical data.
• The developed web calculator facilitates personalized prognostic assessment and treatment strategy optimization.
• This tool enhances clinical decision-making for patients with treatment-naïve advanced ESCC.