Previous studies have indicated that patients with inflammatory arthritis (IA), such as rheumatoid arthritis (RA), psoriatic arthritis (PsA), and ankylosing spondylitis (AS), are at a higher risk of developing cardiovascular diseases (CVDs).¹ As a result, many international organizations such as the European Alliance of Associations for Rheumatology (EULAR) recommend CVD risk assessments to be conducted for patients with RA, AS or PsA.¹ QRisk3, Framingham, and Reynolds are the risk assessment tools available for use.¹ Unfortunately, these tools tend to underestimate the risk, and therefore the EULAR recommends a 1.5 times multiplier to be applied.¹ At the EULAR 2023 European Congress of Rheumatology, Dr. Sizheng Steven Zhao presented his group’s findings on the performance of these 3 risk prediction tools.

This study aimed to compare the performance of the abovementioned 3 risk prediction tools in predicting RA, AS, and PsA outcomes against a population of patients with osteoarthritis (OA).¹ QRisk3 captures RA, while Reynolds considers high-sensitivity C-reactive protein (CRP).¹ Additionally these 3 tools also define cardiovascular (CV) outcomes differently, which may lead to different calculated outcomes.¹ Thus, this study was designed to identify the most accurate tool for predicting these disease outcomes in OA patients.¹

In the study, primary care records in the United Kingdom (UK) were extracted from the Clinical Practice Research Datalink (CPRD) Aurum database and linked with the hospital admission statistics.¹ Patients with read codes corresponding to RA, PsA, AS, and OA were identified, whereas those taking a statin or those who had prior CVD incident before the index date were excluded.¹ The follow-up time ended either at data extraction (October 31, 2021), incident CVD, death, or 10 years.¹ Multiple imputation was used for missing data.¹ The observed 10-year CV risk scores of patients were calculated for each respective tool and compared with their respective actual outcomes.¹ This study investigated the capabilities of each tool to separate CVD cases and non-cases (i.e., discrimination) by time-dependent area under Receiver Operating Characteristic (ROC) curve (AUC), and calibration by measuring the observed and predicted risks in deciles of predicted risk.¹

Among the 93,028 patients with RA (mean age 57 years; 30% male), 44,511 with AS (47 years; 41% male), 26,375 with PsA (47 years; 48% male), and 1,079,815 with OA (63 years; 38% male) assessed, patients with AS and PsA had the lowest rates of CVD (6.8% each) likely due to their younger age group.¹ Conversely, RA patients had a higher rate of CVD (11.4%) whereas those with OA had the highest rate (12.5%).¹ The AUC results showed that across the 3 prediction tools, the performance was the highest for AS (QRisk3: 0.80; Framingham: 0.80; Reynolds: 0.72), followed by RA (QRisk3: 0.75; Framingham: 0.75; Reynolds: 0.71), and PsA (QRisk3: 0.76; Framingham: 0.76; Reynolds: 0.66).¹ Assessment of the calibration curves indicated that QRisk3 overpredicted the risks (observed CVD risk plotted against predicted CVD risk), while Framingham and Reynolds underpredicted the risks (observed CVD risk plotted to the left of the expected CVD risk).¹ The limitations of this study included the use of coded data and no death data, possibly contributing to underpredictions.¹

In summary, the study demonstrated that all 3 prediction tools are less accurate in RA, AS, and PsA.¹ The tools also underperformed in OA, which suggested factors other than inflammation impacting the performance.¹ Additionally, the results indicated that QRisk3 generally overpredicts, suggesting that the x1.5 multiplier may not be necessary.¹ However, on the other hand, the study found that underprediction of Framingham and Reynolds, supporting the use of this multiplier for these prediction tools.¹