Abstract
RATIONALE & OBJECTIVE: Hospitalization is common in patients with advanced chronic kidney disease (CKD). Predicting hospitalization and related outcomes would be beneficial for hospitals and patients. Therefore, we aimed to (1) give an overview of current prediction models for hospitalization, length of stay, and readmission in patients with advanced CKD; (2) externally validate these models; and (3) develop a new model if no valid models were identified. STUDY DESIGN: Systematic review, development, and external validation study. SETTING & PARTICIPANTS: We were interested in prediction models of hospitalization, length of stay, or readmission for patients with advanced CKD. Our available development and validation data consisted of hemodialysis, peritoneal dialysis, and advanced CKD patients not receiving dialysis from a Dutch dialysis and European advanced CKD cohort. SELECTION CRITERIA FOR STUDIES: We systematically searched PubMed. Studies had to intentionally develop, validate, or update a prediction model in adults with CKD. ANALYTICAL APPROACH: We used the PROBAST for risk of bias assessment. Identified models were externally validated on model discrimination (C-statistic) and calibration (calibration plot, slope, and calibration-in-the-large). We developed a Fine-Gray model for hospitalization within 1 year in patients initiating hemodialysis, accounting for the competing risk of death. RESULTS: We identified 45 models in 8 studies. The majority were of low quality with a high risk of bias. Due to underreporting and population-specific predictors, we could only validate 3 models. These were poorly calibrated and had poor discrimination. Using multiple modeling strategies, an adequate new model could not be developed. LIMITATIONS: The outcome hospitalization might be too heterogeneous, and we did not have all relevant predictors available. CONCLUSIONS: Hospitalizations are important but difficult to predict for patients with advanced CKD. An improved prediction model should be developed, for example, using a more specific outcome (eg, cardiovascular hospitalizations) and more predictors (eg, patient-reported outcome measures).