Cardiovascular Risk Models Still Fall Short for Arthritis Patients

MILAN -- Despite longstanding awareness that standard cardiovascular prediction models have not been very accurate for people with inflammatory arthritis, a new study reported here shows that it remains the case.

The most famous of these prediction models, the Framingham Risk Score, substantially underestimated true cardiovascular event risks in real-world arthritis patients with records in Great Britain's Clinical Practice Research Datalink, according to Sizheng Steven Zhao, MBChB, of the University of Liverpool in England, as did the Reynolds Risk Score.

Meanwhile, the British-developed QRISK3 tool launched in 2017 overestimated risk for inflammatory arthritis patients, Zhao told attendees at the European Alliance of Associations for Rheumatology (EULAR) annual meeting.

To be fair, these tools were developed from data drawn from the general population, and patients with inflammatory diseases are known to be at heightened risk for cardiovascular disease and events.

Nevertheless, predictive models specific for these patients do not exist. In 2009, a EULAR task force recommended simply multiplying by 1.5 whatever degree of risk was produced by the Framingham calculator, but a 2015/2016 update refrained from offering anything so specific, merely urging rheumatologists to be aware of the increased risk.

The data now reported by Zhao indicated that a one-size-fits-all approach is not appropriate for different types of inflammatory arthritis, a family that includes rheumatoid arthritis (RA), psoriatic arthritis (PsA), and axial spondyloarthritis (AS).

He and lead investigator David M. Hughes, PhD, also of the University of Liverpool, analyzed records from the Clinical Practice Datalink's Aurum database, which covers more than 40 million Britons in all, to determine the actual 10-year incidence of major cardiovascular events in inflammatory arthritis patients for comparison with predictions from the Framingham, Reynolds, and QRISK3 tools from baseline characteristics. Their numbers were as follows:

• RA: 93,028; mean age 57, 30% male
• PsA: 26,375; 47, 48%
• AS: 26,375; 47, 41%

Zhao and Hughes also looked at data for just over 1 million patients with osteoarthritis (mean age 63, 38% male), a non-inflammatory condition, to serve as a control. They also checked results for psoriasis patients (N not reported) as a different sort of inflammatory condition.

The three tools approach risk prediction differently. QRISK3 is the most comprehensive, basing its estimates on a dozen different factors -- one of which is a diagnosis of RA or systemic lupus erythematosus, which one might think would make it the most accurate. Reynolds uses seven of these same factors (not RA or lupus, however) and the Framingham score is based on eight, including high-sensitivity C-reactive protein levels, a marker of systemic inflammation.

Nevertheless, none of the three tools showed accuracy as good as that reported in their original publications for the general population. Moreover, the accuracy varied according to the specific type of arthritis.

As evaluated by area under the receiver operating characteristic curve (AUC), the accuracies were not bad in absolute terms, ranging from 0.70 to 0.82 for QRISK3, 0.69 to 0.84 for Framingham, and 0.64 to 0.75 for Reynolds. The greatest accuracy was seen for psoriasis in all three models and the worst, perhaps surprisingly, was in osteoarthritis patients. Accuracy in RA patients was in the middle of the range for all three tools.

Zhao also noted that the under- and overpredictions seen in the models were consistent across specific diagnoses. That is, Reynolds and Framingham provided estimates for RA, PsA, AS, and osteoarthritis that were substantially lower than the actual experience, while QRISK3 overestimated the risk. This latter finding is particularly important, Zhao suggested, because clinicians might assume that all models underpredict the risk and exaggerate it even further for patients they evaluate with QRISK3.

He also highlighted the "underperformance" of the three tools for osteoarthritis patients, as a topic for further research to understand how the disease could affect cardiovascular risk in the absence of inflammation.

Unfortunately, the study did not examine mortality risk, Zhao acknowledged, and obviously it was based on clinical records that may not have been 100% accurate and did not cover all potential influences on cardiovascular risk.

Comment