Integration of RA-specific factors (DAS28-CRP score, anti-CCP antibody, RF, MTX use, and disease duration) into a stroke risk model significantly improved stroke risk identification in RA patients, demonstrating superior discrimination over traditional cardiovascular risk assessment tools including the Framingham risk score.
Key Findings
Results
A basic stroke risk model using 11 traditional risk factors identified via LASSO regression achieved moderate discrimination in both derivation and external validation cohorts.
The basic model was derived from NHANES 2011-2020 data (n=1366) and externally validated in a Beijing Tiantan Hospital RA cohort (n=774).
The basic model achieved an AUC of 0.712 (95% CI: 0.665-0.759) in the derivation cohort.
External validation of the basic model yielded an AUC of 0.716 (95% CI: 0.674-0.758).
Neutrophil count, hypertension, and coronary heart disease showed the strongest associations among the 11 traditional risk factors identified.
Results
Incorporating five RA-specific factors into the basic model significantly improved stroke risk discrimination compared to the basic model.
The five RA-specific factors added were DAS28-CRP score, anti-CCP antibody, RF, MTX use, and disease duration.
The enhanced model achieved an AUC of 0.829 (95% CI: 0.794-0.864), compared to 0.716 for the externally validated basic model.
Both the likelihood ratio test (χ²=137.26, P<0.001) and DeLong test (Z=-5.751, P<0.001) confirmed the enhanced model's superiority over the external validation model.
The enhanced model achieved sensitivity of 70.0%, specificity of 81.6%, and accuracy of 78.9%.
Results
The enhanced RA-specific model substantially outperformed the Framingham risk score for stroke risk identification in RA patients.
The Framingham risk score achieved an AUC of 0.611 in the RA cohort.
The enhanced model AUC of 0.829 was significantly superior to the Framingham risk score AUC of 0.611 (DeLong Z=8.210, P<0.001).
This finding highlights the poor performance of existing cardiovascular risk assessment tools in the RA population.
Methods
A two-step modelling approach was employed using two distinct cohorts to develop and validate the stroke risk identification model.
Step one used NHANES 2011-2020 data (n=1366) with LASSO regression to identify traditional stroke risk factors for the basic model.
Step two used a Beijing Tiantan Hospital RA cohort (n=774) for external validation and incorporation of RA-specific factors to develop the enhanced model.
The study design was cross-sectional.
Background
Patients with RA face significantly elevated stroke risk, and existing cardiovascular risk assessment tools perform poorly in this population.
This limitation of traditional tools motivated the development of an RA-specific integrated stroke risk model.
The Framingham risk score, a commonly used cardiovascular risk tool, achieved only an AUC of 0.611 when applied to the RA cohort.
The authors note the model warrants further prospective validation before clinical implementation.
What This Means
This research suggests that people with rheumatoid arthritis (RA) have a higher-than-average risk of stroke, but standard tools doctors use to estimate stroke risk — like the Framingham risk score — do not work well for this group. Researchers developed a new two-part model: first identifying 11 traditional stroke risk factors (with high neutrophil count, high blood pressure, and coronary heart disease being the most important), and then adding five factors specific to RA, including a measure of disease activity (DAS28-CRP), two antibody markers (anti-CCP and rheumatoid factor), use of the drug methotrexate, and how long someone has had RA. Adding the RA-specific factors dramatically improved the model's ability to identify patients at risk for stroke, raising the accuracy measure (AUC) from about 0.72 to 0.83, and far outperforming the Framingham score (AUC 0.61).
The enhanced model correctly identified about 70% of stroke cases (sensitivity) while correctly ruling out stroke risk in about 82% of lower-risk patients (specificity), with an overall accuracy of nearly 79%. The model was developed using a large U.S. national health survey database and then tested on a separate group of RA patients from a Chinese hospital, suggesting it may have broader applicability. This research suggests that using RA-specific disease characteristics alongside traditional cardiovascular risk factors provides a meaningfully better picture of stroke risk in people with RA, and could help clinicians identify which RA patients need more intensive monitoring or preventive care. The authors caution that the model should be tested in future prospective studies before widespread clinical use.
Wan J, Wang J, Cao X, Yang Y. (2026). Cross-sectional stroke risk identification in RA: integrating traditional and disease-specific factors.. Rheumatology (Oxford, England). https://doi.org/10.1093/rheumatology/keag471