Related Experiment Videos
Machine Learning Models for Predicting Pain, Fatigue, Depression, Anxiety, and Malnutrition in Cancer Patients: A
Maidenamu Reheman1, Yunhuan Li1,2, Yang Chen1
1Department of Nursing, West China Hospital, Sichuan University/West China School of Nursing, Sichuan University, Chengdu, Sichuan, People's Republic of China.
Summary
Machine learning (ML) models show acceptable performance in predicting cancer-related symptoms like pain and fatigue. Further prospective validation is needed for clinical use of these ML models in cancer care.
Area of Science:
- Oncology
- Biostatistics
- Medical Informatics
Background:
- Cancer-related symptoms (pain, fatigue, depression, anxiety, malnutrition) significantly impair quality of life and clinical outcomes.
- Machine learning (ML) models are being developed to predict these symptoms, but existing studies lack synthesis of performance, quality, and applicability.
- This study addresses the heterogeneity by comprehensively summarizing ML model characteristics and performance.
Purpose of the Study:
- To systematically review and meta-analyze ML models for predicting cancer-related symptoms.
- To evaluate the predictive accuracy, risk of bias, and clinical applicability of these models.
- To provide a quantitative synthesis of current ML model performance in cancer symptom prediction.
Main Methods:
- Systematic review and meta-analysis of 34 studies (11,217 records) published up to August 31, 2025.
- Data extraction using CHARMS; risk of bias and applicability assessed with PROBAST-AI.
- Evidence quality evaluated using GRADE; random-effects model for pooled analysis; subgroup analyses by cancer type, region, and algorithm.
Main Results:
- Pooled AUCs for predicting symptoms ranged from 0.76 (pain, depression) to 0.86 (malnutrition).
- Moderate certainty of evidence across all outcomes; no significant heterogeneity found in subgroup analyses.
- ML models demonstrated acceptable discriminative performance in available datasets.
Conclusions:
- ML models show promising, acceptable performance for predicting key cancer symptoms.
- Clinical utility necessitates further prospective validation and implementation studies.
- Future research should focus on theory-driven predictors and tailored algorithms for improved clinical translation.
