Presentation (CTI)

OBJECTIVES: Kidney disease is prevalent and costly to treat. Risk prediction tools play an essential role in identifying at-risk individuals to facilitate early detection and management. Independent external validation of prediction models is needed to assess comparative performance and support clinical application. We compared the external validation performance of existing kidney risk prediction models using 0.5 million participants in the UK Biobank.
METHODS: We identified and compared 16 risk prediction models for chronic kidney disease from 3 recent systematic reviews (7 models for the whole population, 9 models specific for type 2 diabetes). We analysed 497,896 adults (age 40-73) in the UK Biobank data; of which 4.7% (n=23,298) had type 2 diabetes. Models were evaluated by discrimination and calibration performance with subgroup analyses by age, sex, ethnicity and pre-existing hypertension.
RESULTS: During a total follow-up of 5.95 million person-years (median: 12.2 years; IQR: 1.4), predictive models for people without diabetes exhibited fair-to-excellent discrimination performance (c-indices: 0.69-0.81) but severely overpredicted risk. The O'Seaghdha model demonstrated the best overall performance for discrimination (c-index: 0.81 [0.81-0.81]) and calibration (slope: 0.69, intercept: -0.011; integrated calibration index: 0.03 [0.02,0.04]). Performance for individuals with type 2 diabetes varied considerably (c-indices: 0.60-0.76 for chronic kidney disease; 0.67-0.88 for kidney failure) though models including medications for diabetes showed superior performance. Discriminative performance was poorer for people with hypertension. Severe miscalibration occurred for many models.
CONCLUSIONS: The O'Seaghdha risk prediction model for chronic kidney disease developed in the US population demonstrated excellent potential for clinical application in this large UK cohort. Model performance in individuals with diabetes or hypertension was poorer warranting further development. Recalibration prior to clinical application is needed for most models.