2019MAJORRWLPhD.pdf (7.07 MB)
Download file

Predicting Cardiovascular Disease Risk in Chronic Kidney Disease

Download (7.07 MB)
posted on 19.11.2019, 13:30 by Rupert W. L. Major
Chronic kidney disease (CKD) is a long term condition in which glomerular filtration is reduced and/or proteinuria occurs. Cardiovascular risk factors in CKD are different to the general population and overall risk is higher too. Therefore, risk prediction tools for cardiovascular disease require specific validation in CKD.
The Leicester City and County CKD (LCC) cohort of 17,248 anonymised individuals with CKD from 44 general practices was established. Cardiovascular events were identified from general practice and hospital records, and 2,072 cardiovascular events occurred during five years of follow-up.
Risk factors for cardiovascular events in CKD were identified in a systematic review and a second systematic review updated a previous systematic review of risk prediction tools for cardiovascular events in CKD. Albumin, haemoglobin and phosphate were identified as risk factors to be consider for risk prediction tools in addition to factors included in general population risk prediction tools. Seven CKD-specific and six general population risk prediction models were identified. All models were developed using the Cox proportional hazards (‘Cox’) model.
The LCC cohort was used to externally validate these models. Discrimination was worse and calibration suggested overprediction of risk in all models. The latter worsened as predicted risk increased. Some calibration improvement was achieved through Cox model baseline risk function re-estimation. There was no significant risk prediction improvement by including the variables identified in the systematic review. Sensitivity analysis suggested that the Cox model’s censoring at random assumption may have been violated in the risk prediction models due to the competing risk from death.
Risk prediction models for cardiovascular events in people with CKD require improvement and updating to optimise risk prediction accuracy. Alternative methods, such as multi-state models, should be considered in future model development.



Laura Gray; Nigel Brunskill

Date of award


Author affiliation

Department of Health Sciences

Awarding institution

University of Leicester

Qualification level


Qualification name




Usage metrics