The predictive accuracy of cardiovascular risk prediction tools in inflammatory arthritis and psoriasis: an observational validation study using the Clinical Practice Research Datalink

Abstract

Abstract Objectives Cardiovascular risk prediction tools developed for the general population often underperform for individuals with rheumatoid arthritis (RA), and their predictive accuracy are unclear for other inflammatory conditions that also have increased cardiovascular risk. We investigated performance of QRISK-3, Framingham Risk Score (FRS) and Reynolds Risk Score (RRS) in RA, psoriatic disease (psoriatic arthritis (PsA) and psoriasis) and ankylosing spondylitis (AS). We considered osteoarthritis as a non-inflammatory comparator. Methods We utilised primary care records from the Clinical Practice Research Datalink (CPRD) Aurum database to identify individuals with each condition and calculated 10-year cardiovascular risk using each prediction tool. Discrimination and calibration of each tool in each disease was assessed. Results Time-dependent AUC for QRISK3 was 0.752 for RA (95% CI 0.734–0.777), 0.794 for AS (95% CI 0.764–0.812), 0.764 for PsA (95% CI 0.741–0.791),0.815 for psoriasis (95% CI 0.789–0.835), and 0.698 for osteoarthritis (95% CI 0.670–0.717) indicating reasonably good predictive performance. AUC for FRS were similar, and slightly lower for RRS. FRS was reasonably well calibrated for each condition but underpredicted risk for patients with RA. RRS tended to underpredict CVD risk, whilst QRISK3 overpredicted CVD risk, especially for the most high-risk individuals. Conclusions CVD risk for individuals with RA, AS and psoriatic disease were generally less accurately predicted using each of the 3 CVD risk prediction tools than reported accuracies in the original publications. Individuals with osteoarthritis also had less accurate predictions suggesting inflammation is not the sole reason for underperformance. Disease specific risk prediction tools may be required.