Home›Oncology› Machine learning models predict cancer symptoms with AUC values ranging from 0.76 to 0.86
Machine learning models predict cancer symptoms with AUC values ranging from 0.76 to 0.86Machine Learning Models Predict Common Symptoms in Cancer Patients
Journal of nursing scholarship : an official publication of Sigma Theta Tau International Honor Society of NursingPublished September 2, 2026Study authors: Reheman Maidenamu, Li Yunhuan, Chen Yang, Yan Qianwen, Hu XiaolinPubMed ↗DOI ↗Editorial oversight: Dr. Julia Lee, PhD · Oncology, Genomics & Drug Development
AI-generated summary of the cited source, checked by automated accuracy review.
How we work
Share
Key Takeaway
Note that machine learning models provide a preliminary foundation for predicting cancer symptoms with moderate certainty.
This meta-analysis synthesized 34 studies to evaluate the predictive accuracy of machine learning (ML) models for identifying symptoms in cancer patients. The primary outcomes measured were the Area Under the Curve (AUC) for predicting pain, fatigue, depression, anxiety, and malnutrition.
The analysis reported AUC values of 0.76 for pain and 0.76 for depression. Fatigue prediction showed an AUC of 0.82, while anxiety prediction yielded an AUC of 0.78. Malnutrition prediction demonstrated the highest accuracy with an AUC of 0.86. The authors noted moderate certainty of evidence across all reported outcomes.
Limitations noted by the authors include a predominance of internal validation and observed heterogeneity among the studies. While these results provide a preliminary foundation for clinical translation, the authors emphasize that clinical utility requires further prospective validation and implementation studies. The findings suggest that ML models can serve as a tool for symptom prediction, but their practical application in clinical workflows remains to be fully established.
How this fits prior evidence
This meta-analysis addresses a gap in the technological tools available for managing cancer symptoms. While previous evidence highlighted the impact of multidisciplinary palliative rehabilitation on quality of life and the prevalence of oral frailty in cancer patients, this study provides a technical basis for identifying those patients at risk for specific symptoms like fatigue and malnutrition using machine learning.
Researchers analyzed 34 different studies to see how well machine learning models could predict common symptoms in cancer patients. The study looked at five specific areas: pain, fatigue, depression, anxiety, and malnutrition. The results showed that these computer models had a moderate level of accuracy in predicting these conditions.
Specifically, the models showed high accuracy for predicting malnutrition and fatigue. They also showed consistent results for predicting pain, depression, and anxiety. While the results are promising, the researchers noted that many of the studies used internal data to test the models, which can sometimes limit how well the results apply to every patient.
Because this is a meta-analysis of existing studies, the findings are currently a foundation for future work. More large-scale, real-world studies are needed to see how these tools can be used in daily clinical practice. For now, these results suggest that machine learning could eventually help doctors better anticipate and manage patient needs.
What this means for you:
Machine learning models show promise in predicting cancer symptoms like pain and fatigue, but more research is needed.
Common questions
What symptoms can machine learning help predict in cancer patients?
The study looked at five main symptoms: pain, fatigue, depression, anxiety, and malnutrition. The machine learning models showed varying levels of predictive accuracy for each. Specifically, the models showed an accuracy of 0.86 for malnutrition and 0.82 for fatigue, while pain, depression, and anxiety had accuracy scores of 0.76 and 0.78.
How accurate were the models for predicting pain and anxiety?
The models showed a predictive accuracy of 0.76 for pain and 0.76 for depression. For anxiety, the models showed a predictive accuracy of 0.78. These scores indicate a moderate level of certainty across all outcomes studied in the analysis.
Can these tools be used in clinics right now?
The results are currently considered a preliminary foundation for clinical use. Because many studies used internal validation and there was some variation in the data, more prospective studies are needed to confirm how these tools work in real-world medical settings before they can be widely used by doctors.
INTRODUCTION: Cancer-related symptoms including pain, fatigue, depression, anxiety, and malnutrition drive poor quality of life and adverse clinical outcomes in cancer patients. While machine learning (ML) models are increasingly developed to predict these symptoms, existing studies are marked by significant heterogeneity in algorithms, sample sizes, and predictors, and lack quantitative synthesis of model performance, methodological quality, and clinical applicability. This study aimed to comprehensively summarize the characteristics of models and predictors, evaluate the predictive accuracy, risk of bias, and clinical applicability of ML prediction models.
DESIGN: Systematic review and meta-analysis.
METHODS: A comprehensive literature search was conducted in PubMed, Web of Science, the Cochrane Library, CINAHL, PsycINFO, CNKI, WanFang, VIP, and SinoMed, from database inception to August 31, 2025. Data were extracted in accordance with the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS), and the risk of bias and applicability of included models were assessed using the Prediction Model Risk of Bias Assessment Tool and Artificial Intelligence (PROBAST-AI). The quality of evidence was evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework. A random-effects model was employed for pooled analysis. Subgroup analyses were stratified by cancer type, geographic region, and algorithm type.
RESULTS: A total of 11,217 records were retrieved, and 34 studies were included in the analysis. The pooled AUCs for predicting pain, fatigue, depression, anxiety, and malnutrition were 0.76 (95% CI: 0.69-0.83, I = 97.7%), 0.82 (95% CI: 0.76-0.88, I = 98.5%), 0.76 (95% CI: 0.70-0.82, I = 98.4%), 0.78 (95% CI: 0.69-0.86, I = 37.9%), and 0.86 (95% CI: 0.80-0.91, I = 94.9%), respectively. Subgroup analyses across cancer type, geographical region, and algorithm type revealed no statistically significant sources of heterogeneity. The certainty of evidence was moderate across all outcomes.
CONCLUSION: This systematic review and meta-analysis showed that ML models achieved acceptable discriminative performance for predicting pain, anxiety, depression, fatigue, and malnutrition in patients with cancer in available datasets. Given predominant internal validation and observed heterogeneity, clinical utility requires further prospective validation and implementation studies. Future research may consider theory-driven predictors and clinically tailored algorithms to improve model performance.
CLINICAL RELEVANCE: These pooled findings provide a preliminary foundation for the clinical translation of ML models to predict pain, anxiety, depression, fatigue, and malnutrition in cancer patients. Further prospective validation in diverse clinical settings and randomized controlled trials evaluating the effectiveness of model-guided symptom management strategies are needed to improve patient outcomes.
PROSPERO REGISTRATION: CRD420251130183.
