Risk prediction models for malnutrition in patients with cancer: a systematic review

At present, many nutritional risk prediction models have been developed for cancer patients, but there is still uncertainty regarding the methodological quality and clinical applicability of these models.

To systematically review and critically appraise existing risk prediction models for malnutrition in cancer patients.

The PubMed, EBSCO, Medline, Web of Science, Scopus and Cochrane Library databases were systematically retrieved. The retrieval period was from the establishment of the databases to September 30, 2025. Based on the Predictive Model Bias Risk of Assessment Tool (PROBAST) and Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS), two independent authors conducted a rigorous assessment and data extraction of the study.

A total of 6,343 articles were retrieved, ultimately including 16 studies and 42 models. The sample size included in the studies ranged from 120 to 4,487. Models were developed with logistic regression or machine-learning algorithms. The area under the curve (AUC) for the development cohort ranged from 0.745 to 1.000, while for the validation cohort ranged from 0.687 to 0.982. Twelve studies (75.0%) evaluated model calibration using calibration curves, the Hosmer-Lemeshow test, and the Brier score, demonstrating good calibration performance. All risk prediction models have a relatively high risk of bias, primarily due to issues with the study population and analysis domain, two models (12.5%) raised high concern regarding applicability.

Research on prediction models for malnutrition in cancer patients is still in the development stage. The predictive performance of the developed models is generally acceptable, but there are still deficiencies in validation and evaluation.