Am J Kidney Dis. 2026 May 15:S0272-6386(26)00919-4. doi: 10.1053/j.ajkd.2026.03.038. Online ahead of print.
ABSTRACT
RATIONALE & OBJECTIVE: The performance of clinical risk prediction scores for ischemic stroke and bleeding are not well established in those with CKD and atrial fibrillation (AF). We aimed to evaluate the performance of three risk scores for ischemic stroke and intracranial bleeding in patients across a broad range of kidney function.
STUDY DESIGN: Retrospective study.
SETTING & PARTICIPANTS: The study included two community-based cohorts of adults with incident AF (Kaiser Permanente and Ontario, Canada).
EXPOSURE: Baseline estimated glomerular filtration rate (eGFR) was calculated using outpatient serum creatinine measures, excluding those with a kidney transplant or on dialysis.
OUTCOMES: Three risk scores for ischemic stroke (ATRIA, CHA2DS2-VASc and R2CHADS2) and one for major bleeding (HAS-BLED) were calculated across eGFR categories (≥60, 45-59, 30-44, and <30 mL/min/1.73 m2). Outcomes included ischemic stroke and intracranial hemorrhage.
ANALYTICAL APPROACH: C-statistics were calculated and calibration plots generated within eGFR strata.
RESULTS: There were 101,360 adults with incident AF in the Kaiser Permanente cohort and 33,200 adults with incident AF in the Ontario cohort. Mean risk scores for stroke and bleeding were higher with lower eGFR across all four measures. The c-statistics for the stroke prediction scores were lower in those with an eGFR <60 mL/min/1.73 m2 compared with eGFR ≥ 60 mL/min/1.73 m2 in both cohorts. The lowest discrimination for ischemic stroke was seen with the R2CHADS2 risk score (ranging from 0.53-0.61) with a c-statistic of 0.57 for the eGFR <30 mL/min/1.73m2 category in both cohorts. The c-statistics for ATRIA (range 0.59 to 0.64) and CHA2DS2-VASc (range 0.55 to 0.61) were also modest across lower eGFR categories compared with eGFR >60 mL/min/1.73 m2. The discrimination for the HAS-BLED bleeding risk score for intracranial bleeding was also low in those with lower eGFR categories (ranging from 0.51 to 0.56). Calibration varied by risk score and eGFR level.
LIMITATIONS: Observational study.
CONCLUSIONS: Performance of clinically used stroke and bleeding risk prediction scores was modest among those with CKD, particularly at advanced CKD. More accurate risk scores are needed to improve decision-making.
PLAIN-LANGUAGE SUMMARY: This retrospective analysis evaluated the predictive accuracy of standard clinical risk scores for ischemic stroke (ATRIA, CHA2DS2-VASc and R2CHADS2) and intracranial hemorrhage (HAS-BLED) in patients with atrial fibrillation (AF) and varying degrees of kidney function. Utilizing data from two large cohorts, Kaiser Permanente and an Ontario-based registry, the study focused on how baseline estimated glomerular filtration rate (eGFR) influenced the discrimination and calibration of these models. The study revealed a significant inverse relationship between kidney function and model reliability. While mean risk scores were higher in patients with lower eGFR, the c-statistics and calibration decreased significantly in more advanced stages of kidney disease. The results suggest that the pathophysiological complexities of advanced kidney disease limit the utility of tools to predict stroke and bleeding designed for the general population with AF. Development of kidney-specific risk models is needed to better inform anticoagulation strategies and clinical decision-making.
PMID:42142610 | DOI:10.1053/j.ajkd.2026.03.038