Assessing risk in chronic kidney disease: a methodological review

Area of Science:

• Nephrology
• Epidemiology
• Biostatistics

Background:

• Chronic kidney disease (CKD) presents a growing public health challenge, marked by significant morbidity and mortality.
• Risk prediction models are crucial for forecasting adverse events and stratifying CKD patients, but require precise specification for accuracy.

Purpose of the Study:

• To outline essential considerations for accurate absolute risk prediction in chronic kidney disease (CKD).
• To emphasize the importance of careful model specification, including competing events, predictor variable forms, and interactions.

Main Methods:

• Discusses the necessity of accounting for competing events (e.g., death) that may preclude the event of interest.
• Highlights the need for accurate specification of functional forms, nonlinearity, and interactions of predictor variables.
• Addresses the potential impact of measurement error on risk prediction accuracy.

Main Results:

• Model misspecification in any component can significantly impact absolute risk prediction.
• Recommends evaluating prognostic models using traditional metrics (e.g., Hosmer-Lemeshow, AUC) and newer measures (e.g., risk reclassification tables, net reclassification indices).
• Stresses the utility of newer metrics for assessing the addition of novel predictors to existing models.

Conclusions:

• Absolute risk prediction models necessitate rigorous internal and external validation.
• Generalizability of models is typically limited to populations with similar baseline characteristics and competing event rates.
• Accurate risk prediction is vital for effective clinical decision-making and patient management in CKD.