Prediction Models for Psychological Distress in Patients With Malignant Tumors: A Scoping Review.

Abstract

Objective: Psychological distress is common among patients with malignant tumors and adversely affects treatment adherence and quality of life. Numerous prediction models have been developed to identify high-risk patients, yet few have been implemented clinically. This scoping review synthesizes the development methods, performance, validation methods, and limitations of existing models to inform future research and support clinical translation.

Methods: Following the Joanna Briggs Institute (JBI) methodology, eight databases were searched from inception to June 10, 2026. Two reviewers independently conducted screening, data extraction, and quality assessment.

Results: Thirteen studies involving 26 prediction models were included. Logistic Regression (LR), Random Forests (RF), eXtreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN) were the most common development methods. Reported sensitivities ranged from 0.518 to 0.968, specificities from 0.651 to 1.000, and Areas Under the Curve (AUCs) from 0.673 to 1.000. Thirteen studies underwent internal validation only; none underwent external validation, and all were rated as high risk of bias. Frequently included predictors were tumor stage, sleep quality, pain degree, age, financial problems, and coping style. Nomograms and web-based calculators were the predominant presentation formats.

Conclusion: Although models developed using XGBoost, RF, and ANN reported high performance, these findings are likely inflated due to small sample sizes, low Events Per Variable (EPV), and lack of external validation. Future research should strengthen methodological rigor, increase sample sizes, and conduct external validation to support clinical adoption.