Machine Learning-Based Prediction of Fear of Cancer Recurrence in Gastrointestinal Cancer Survivors: A Cross-Sectional Study

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Aims: To explore the current status of fear of cancer recurrence (FCR) in gastrointestinal (GI) cancer survivors, and analyze its influencing factors based on random forest algorithms. Design: A cross-sectional study design. Method: A convenient sampling method was used to select GI cancer survivors who were hospitalized in a tertiary Grade A hospital in Guangdong Province from April 2024 to May 2025. Patients were surveyed using the General Information Questionnaire, Fear of Progression Questionnaire-Short Form (FoP-Q-SF), Brief Illness Perception Questionnaire (BIPQ), Herth Hope Index (HHI), Comprehensive Score for Financial Toxicity–Patient-Reported Outcome Measure (COST-PROM), and Social Support Rating Scale (SSRS). A two-stage analytical strategy was used for prediction: five classifiers (logistic regression, ridge regression, Bayesian logistic regression, support vector machine, and random forest) were compared; the optimal model was then further optimized and interpreted via variable importance ranking and partial dependence plots. Results: A total of 572 GI cancer survivors were recruited. The incidence of FCR (FoP-Q-SF score ≥ 34) was 75.0%. The random forest model showed better predictive performance (AUC = 0.904) than other algorithms. The top five variables in order of importance were advanced cancer stage (39.79), illness perception (31.79), financial toxicity (24.87), body mass index (21.19), and age (14.04). On the test set, the model achieved an AUC of 0.884, an accuracy of 0.824, a sensitivity of 0.643, and a specificity of 0.883. Conclusion: The incidence of FCR is high among GI cancer survivors. Clinical nurses can identify high-risk patients early and implement effective nursing interventions based on the influencing factors of FCR in GI cancer survivors.