Diagnostic prediction models show moderate-to-good discrimination for cancer-related fatigue

Photo by Google DeepMind / Unsplash

Supportive care in cancer : official journal of the Multinational Association of Supportive Care in Cancer Published June 6, 2026 Medically reviewed June 6, 2026 Study authors: Liang Fanqi, Wang Menglu, Li Keke, Lv Deliang, Qian Taoming, Bu Zhijun PubMed ↗ DOI ↗ By Dr. Julia Lee, PhD · Oncology, Genomics & Drug Development

Key Takeaway

Interpret pooled AUC of 0.83 for cancer-related fatigue prediction models with caution due to high heterogeneity and bias risk.

This systematic review and meta-analysis evaluated diagnostic prediction models for cancer-related fatigue. The analysis included 13 studies with a total of 444,447 cancer patients from research settings. The primary outcome was the pooled area under the receiver operating characteristic curve (AUC), which measures a model's ability to discriminate between patients with and without fatigue.

The pooled AUC was 0.83 (95% CI = 0.78-0.87), indicating moderate-to-good discrimination. However, the authors caution that this estimate should be interpreted with caution due to substantial heterogeneity (I > 94%) and high risk of bias in most studies, particularly related to statistical analysis and reporting. Possible funnel plot asymmetry was noted but considered exploratory given the small number of studies and high heterogeneity.

Limitations include the substantial heterogeneity and risk of bias, which reduce confidence in the pooled estimate. The authors note that the findings provide evidence for clinical practice and future research, but the models require further validation before widespread use.

Study Details

Study typeMeta analysis

EvidenceLevel 1

PublishedJun 2026

PMID42230345

View Original Abstract ↓

PURPOSE: To evaluate the performance and methodological quality of published diagnostic prediction models for cancer-related fatigue (CRF), and to provide evidence for clinical practice and future research. METHODS: A systematic review and meta-analysis were conducted. PubMed, Web of Science, the Cochrane Library, Embase, and Scopus were searched from inception to October 18, 2024, for studies developing or validating diagnostic prediction models for CRF. The pooled area under the receiver operating characteristic curve (AUC) and 95% confidence interval (CI) were calculated using R. Heterogeneity was assessed using the I statistic and Cochran's Q test, publication bias was explored using funnel plots and Egger's test, and risk of bias was evaluated with the Prediction Model Risk of Bias Assessment Tool (PROBAST). RESULTS: A total of 8418 records were identified, of which 13 studies met the inclusion criteria. These studies included 23 cohorts, 444,447 cancer patients, and 11 diagnostic prediction models. The pooled AUC was 0.83 (95% CI = 0.78-0.87), indicating moderate-to-good discrimination. However, substantial heterogeneity was observed (I > 94%), suggesting that the pooled estimate should be interpreted with caution. PROBAST assessment indicated a high risk of bias in most studies, mainly related to statistical analysis and reporting. Egger's test suggested possible funnel plot asymmetry, but this finding should be considered exploratory because of the small number of studies and high heterogeneity. CONCLUSION: Existing CRF diagnostic prediction models show moderate-to-good discrimination in research settings, but their performance varies across populations, outcome definitions, and modeling approaches. Future studies should prioritize large-scale, multi-center, multiethnic, and externally validated models to improve early identification and precise management of CRF.