A systematic review finds prediction models for chronic kidney disease were poorly reported and often developed using inappropriate methods

Background: Chronic kidney disease (CKD) is a global health concern that is increasing mainly as the result of increasing incidences of diabetes and hypertension. Furthermore, if left untreated, individuals with CKD may progress to end-stage kidney failure. Identifying individuals with undiagnosed CKD or those who are at an increased risk of developing CKD or progressing to end-stage kidney disease (ESKD) is therefore an important challenge. We sought to systematically review and critically assess the conduct and reporting of methods used to develop risk prediction models for predicting the risk of having undiagnosed (prevalent) or future risk of developing (incident) CKD or end-stage kidney failure in adults.

Methods: We conducted a systematic search of PubMed database to identify studies published up until September 2011 that describe the development of models combining two or more variables to predict the risk of prevalent or incident CKD or ESKD. We extracted key information that describes aspects of developing a prediction model, including the study design, data quality, sample size and number of events, outcome definition, risk predictor selection and coding, missing data, model-building strategies, and aspects of performance.

Results: Eleven studies describing the development of 14 prediction models were included. Eight studies reported the development of 11 models to predict incident CKD or ESKD, whereas 3 studies developed models for prevalent CKD. A total of 97 candidate risk predictors were considered, and 43 different risk predictors featured in the 14 prediction models. A method, not recommended to select risk predictors for inclusion in the multivariate model, using statistical significance from univariate screening was carried out in six studies. Missing data were frequently poorly handled and reported with no mention of missing data in four studies; 4 studies explicitly excluded individuals with missing data, and only 2 studies used multiple imputation to replace missing values.

Conclusion: We found that prediction models for chronic kidney were often developed using inappropriate methods and were generally poorly reported. Using poor methods can affect the predictive ability of the models, whereas inadequate reporting hinders an objective evaluation of the potential usefulness of the model.