Risk prediction models for patients with chronic kidney disease: a systematic review.

Background: Patients with chronic kidney disease (CKD) are at increased risk for kidney failure, cardiovascular events, and all-cause mortality. Accurate models are needed to predict the individual risk for these outcomes.

Purpose: To systematically review risk prediction models for kidney failure, cardiovascular events, and death in patients with CKD.

Data Sources: MEDLINE search of English-language articles published from 1966 to November 2012.

Study Selection: Cohort studies that examined adults with any stage of CKD who were not receiving dialysis and had not had a transplant; had at least 1 year of follow-up; and reported on a model that predicted the risk for kidney failure, cardiovascular events, or all-cause mortality.

Data Extraction: Reviewers extracted data on study design, population characteristics, modeling methods, metrics of model performance, risk of bias, and clinical usefulness.

Data Synthesis: Thirteen studies describing 23 models were found. Eight studies (11 models) involved kidney failure, 5 studies (6 models) involved all-cause mortality, and 3 studies (6 models) involved cardiovascular events. Measures of estimated glomerular filtration rate or serum creatinine level were included in 10 studies (17 models), and measures of proteinuria were included in 9 studies (15 models). Only 2 studies (4 models) met the criteria for clinical usefulness, of which 1 study (3 models) presented reclassification indices with clinically useful risk categories.

Limitation: A validated risk-of-bias tool and comparisons of the performance of different models in the same validation population were lacking.

Conclusion: Accurate, externally validated models for predicting risk for kidney failure in patients with CKD are available and ready for clinical testing. Further development of models for cardiovascular events and all-cause mortality is needed.

Primary Funding Source: None.