Con: Most clinical risk scores are useless.

Area of Science:

• Nephrology
• Medical Informatics
• Clinical Epidemiology

Background:

• Prediction model development is prevalent in nephrology but clinical implementation is limited.
• Many published models suffer from poor reporting and inappropriate methodologies, hindering their practical use.

Purpose of the Study:

• To highlight the shortcomings in the development and validation of prediction models in nephrology.
• To advocate for increased focus on external validation and impact assessment of existing models.

Main Methods:

• Review of common issues in prediction model development, including predictor selection and sample size.
• Emphasis on the critical role of external validation for assessing model performance in new populations.
• Discussion of the scarcity of impact studies demonstrating clinical outcome improvements.

Main Results:

• Many prediction models are developed using flawed methods, such as inappropriate predictor selection or inadequate sample sizes.
• External validation is rarely performed for nephrology prediction models, limiting their generalizability.
• Few studies report on the clinical impact of prediction models in routine practice.

Conclusions:

• Most nephrology prediction models are inadequately reported and lack external validation, preventing clinical adoption.
• Future research should prioritize validating existing models and assessing their real-world impact over developing new ones.