Risk prediction models for malnutrition in dialysis patients in China: a systematic review and meta-analysis
Clinical Snapshot
PICO Framework
| P — Population | Adult dialysis patients (haemodialysis and/or peritoneal dialysis) in China with chronic kidney failure |
| I — Intervention | Risk prediction models for malnutrition (clinical prediction rules, nomograms, logistic regression-based scoring tools) |
| C — Comparator | No comparator (systematic review of model development and validation studies); reference standard for malnutrition diagnosis varies across included studies |
| O — Outcomes | Primary: discriminatory performance of prediction models (pooled AUC/C-statistic); Secondary: pooled prevalence of malnutrition, identification of statistically significant predictors, risk of bias assessment via PROBAST |
Bottom Line
This systematic review and meta-analysis of 12 Chinese studies evaluating risk prediction models for malnutrition in dialysis patients reports a pooled malnutrition prevalence of 41% and a pooled AUC of 0.83 across nine internally validated models — suggesting adequate discrimination on paper. Eight clinically plausible predictors were identified: age, sex, serum calcium, Kt/V, triglycerides, vitamin D, NT-proBNP, and comorbid diabetes. However, the critical finding is that every included model was rated at high risk of bias by PROBAST, primarily due to inappropriate data sources and poor analytical reporting. No externally validated models were identified. Calibration data and confidence intervals for pooled estimates are absent from reporting. These methodological deficiencies mean that no existing model can be recommended for routine clinical use. For Australian nephrologists and renal nurses, the practical implication is clear: continue using validated nutritional screening tools such as the Malnutrition-Inflammation Score or Subjective Global Assessment while awaiting the development of rigorously designed, externally validated prediction models. The predictor set identified here provides a useful evidence-informed starting point for future model development in Australian and other non-Chinese dialysis cohorts.
Key Findings
P Value: Not reported for pooled AUC; predictors reported as 'statistically significant'
Effect Size: Pooled AUC/C-statistic = 0.83 (nine internally validated models); pooled malnutrition prevalence = 41%
Primary Outcome: Discriminatory performance of risk prediction models for malnutrition in dialysis patients in China
Nnt Or Sensitivity: Sensitivity and specificity of individual models not reported in abstract; calibration metrics (Hosmer-Lemeshow, calibration plots) not synthesised; no NNT or decision curve analysis reported
Confidence Interval: Not reported in abstract
Clinical Application
The identified predictors are largely routinely collected in dialysis care (Kt/V, serum calcium, triglycerides, vitamin D, NT-proBNP, diabetes status, age, sex), making future model implementation feasible in principle. However, no existing model can be recommended for clinical deployment given universal high risk of bias and absence of external validation. Clinicians should continue using validated nutritional screening tools (e.g., Malnutrition-Inflammation Score, Subjective Global Assessment) pending development of externally validated prediction models. In Australia, malnutrition in dialysis patients is a recognised clinical priority. ANZDATA registry data indicate substantial nutritional morbidity in the Australian dialysis population. The RACGP and Kidney Health Australia guidelines emphasise regular nutritional assessment in CKD and dialysis patients. The Malnutrition-Inflammation Score and Subjective Global Assessment are currently recommended screening tools. The predictors identified in this review (Kt/V, serum calcium, vitamin D, NT-proBNP, diabetes) align with parameters routinely monitored under Australian dialysis care standards. PBS-listed nutritional supplements and dietitian-led interventions are available for malnourished dialysis patients. However, the Chinese-specific models reviewed cannot be directly adopted in Australian practice without external validation in Australian cohorts. TGA approval would be required for any digital clinical decision support tool incorporating these models. Future Australian-specific model development could leverage ANZDATA and state-based renal registries. Adult patients receiving haemodialysis or peritoneal dialysis for chronic kidney failure, particularly in Chinese healthcare settings. Findings regarding predictor variables (age, serum calcium, Kt/V, triglycerides, sex, vitamin D, NT-proBNP, diabetes) may inform clinical vigilance in dialysis populations globally, but direct model application requires external validation in the target population.
Abstract
Although multiple risk prediction models have been developed to identify malnutrition in dialysis patients, their quality and performance remain unclear, limiting their practicality in current clinical practice and future research. Therefore, we conducted a systematic review and meta-analysis to evaluate these models. Searches were conducted in PubMed, Embase, Web of Science, The Cochrane Library, CINAHL, SinoMed, CNKI, Wanfang, and VIP Database from inception to January 26, 2026. Two investigators independently screened the literature, extracted data, and assessed quality using the Prediction model Risk of Bias Assessment Tool (PROBAST). Meta-analyses of the prevalence of malnutrition, common predictors and model performance were performed using Stata 18.0 and R 4.5.1. A total of 12 eligible studies conducted in China were included, and the pooled prevalence of malnutrition in dialysis patients was 41%. Meta-analysis identified age, serum calcium, Kt/V, triglycerides, sex, vitamin D, NT-proBNP, and comorbid diabetes as statistically significant predictors. The pooled effect of the nine internal validated models was 0.83, indicating good discriminatory performance. However, all included models were rated at high risk of bias, primarily due to inappropriate data sources and poor reporting of the analysis. The current analysis reveals a high prevalence of malnutrition among dialysis patients. Eight significant predictors were identified, guiding future selection for constructing predictive models of malnutrition risk in this population. Although existing models demonstrate adequate discriminatory performance, their methodological limitations constrain clinical applicability. Future studies should prioritize the development of standardized, externally validated models to enable early identification and intervention, thereby improving outcomes in this vulnerable group.
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service