Standard cardiovascular risk models no good in rheumatoid arthritis

Established cardiovascular risk models such as the Systematic Coronary Risk Evaluation and the Framingham risk score either underestimated or overestimated the risk of cardiovascular events in a retrospective analysis of prospectively collected data from a cohort of patients with early rheumatoid arthritis.

These findings illustrate that what is needed is an RA-specific CV risk model whose performance "should be compared with the performance of the current risk algorithms," according to lead investigator Elke E.A. Arts of Radboud University Nijmegen (Netherlands) Medical Centre and her colleagues.

The four different risk algorithms that failed to correctly estimate the 10-year risk of a CV event were the Framingham risk score, the Systematic Coronary Risk Evaluation (SCORE), the Reynolds risk score, and the QRisk II risk score.

Proposals to fix to these models suggest that patients at risk might be better identified by adjusting the cutoff points in CV risk that are used as indications for primary prevention for RA patients, but "this could also lead to overtreatment, as the majority of patients in the lower risk group do not develop events," cautioned Ms. Arts, a PhD student at the university, and her associates.

They noted that "alternatively, a correction factor could be used to adjust the CV risk in patients with RA, as was suggested by the European League Against Rheumatism (EULAR) recommendations for CV risk management," they added. The EULAR recommendations advise multiplying the results of the SCORE risk algorithm by 1.5 for RA patients when they fulfill two out of the following three criteria: disease duration greater than 10 years, rheumatoid factor or anti–cyclic citrullinated peptide positivity, and extra-articular manifestations. However, "there are no data supporting such a multiplicator; it was based on expert opinion," they wrote.

In the current study, the investigators applied the four risk algorithms to data from 1,050 patients who were originally enrolled in the Nijmegen Early RA Inception cohort. Patients with preexisting CV disease were excluded, and in cases with less than 10 years of follow-up, risk scores were "adjusted proportionally according to the length of actual follow-up and calculated as a proportion of 10 years" (Ann. Rheum. Dis. 2014 Jan. 3 [doi: 10.1136/annrheumdis-2013-204024]).

Overall, over the course of 9,957 patient-years available for analysis, there were 149 episodes of a first CV event, the primary endpoint of the study (1.14 events per 100 patient-years). This total included 67 cases of acute/unstable coronary syndrome (myocardial infarction or unstable angina), 24 cases of stable angina, 26 cerebrovascular accidents, 10 transient ischemic attacks, 18 cases of peripheral vascular disease, and 4 cases of heart failure. A total of 15 of these events were fatal.

The sensitivity of the models was 68%-87% for the risk score cutoff value of 10% (marking the difference between low risk and intermediate to high risk) and 40%-65% for the risk score cutoff value of 20% (marking the difference between low- to intermediate-risk and high-risk patients). The corresponding specificity ranges were 55%-76% and 77%-88%, respectively. Those cutoff points are recommended to be used as indicators for CV preventive treatment such as lifestyle adjustments and drug therapy interventions.

For the SCORE risk model, the researchers found that CV risk predictions deviated from observed CV events in all risk deciles, but especially in the middle and top deciles. Indeed, tests of the fit of the SCORE model to the data with the Hosmer-Lemeshow (H-L) test yielded a P value of less than .001, indicating poor model fit (where good fit corresponds to P values greater than .05).

Looking at the Framingham model, the number of CV events predicted was similar to the observed number of CV events, showing a modest difference in predicted and observed CV risk in the lower- and middle-risk deciles, according to the authors. However, the top two deciles decidedly under- and overestimated risk, respectively, with an H-L test indicating poor fit for the whole model (P =.024).

The next tested model, the Reynolds risk score, underestimated the number of CV events, with an overall P value of .020 on the H-L test.

The QRisk II had a moderately good fit on the H-L test (P = .20) but still mainly overestimated the observed CV risk.

Ms. Arts stated in an e-mail that this study was originally presented at the 2013 American College of Rheumatology annual meeting. She added that she is currently collaborating with a group on an RA-specific CV risk model.

Ms. Arts disclosed that the study was partially funded by the Rheumatology Research University Nijmegen foundation. Three of her coauthors disclosed financial relationships with multiple pharmaceutical companies, including the makers of drugs and therapies for RA.

Established cardiovascular risk models such as the Systematic Coronary Risk Evaluation and the Framingham risk score either underestimated or overestimated the risk of cardiovascular events in a retrospective analysis of prospectively collected data from a cohort of patients with early rheumatoid arthritis.

These findings illustrate that what is needed is an RA-specific CV risk model whose performance "should be compared with the performance of the current risk algorithms," according to lead investigator Elke E.A. Arts of Radboud University Nijmegen (Netherlands) Medical Centre and her colleagues.

The four different risk algorithms that failed to correctly estimate the 10-year risk of a CV event were the Framingham risk score, the Systematic Coronary Risk Evaluation (SCORE), the Reynolds risk score, and the QRisk II risk score.

Proposals to fix to these models suggest that patients at risk might be better identified by adjusting the cutoff points in CV risk that are used as indications for primary prevention for RA patients, but "this could also lead to overtreatment, as the majority of patients in the lower risk group do not develop events," cautioned Ms. Arts, a PhD student at the university, and her associates.

They noted that "alternatively, a correction factor could be used to adjust the CV risk in patients with RA, as was suggested by the European League Against Rheumatism (EULAR) recommendations for CV risk management," they added. The EULAR recommendations advise multiplying the results of the SCORE risk algorithm by 1.5 for RA patients when they fulfill two out of the following three criteria: disease duration greater than 10 years, rheumatoid factor or anti–cyclic citrullinated peptide positivity, and extra-articular manifestations. However, "there are no data supporting such a multiplicator; it was based on expert opinion," they wrote.

In the current study, the investigators applied the four risk algorithms to data from 1,050 patients who were originally enrolled in the Nijmegen Early RA Inception cohort. Patients with preexisting CV disease were excluded, and in cases with less than 10 years of follow-up, risk scores were "adjusted proportionally according to the length of actual follow-up and calculated as a proportion of 10 years" (Ann. Rheum. Dis. 2014 Jan. 3 [doi: 10.1136/annrheumdis-2013-204024]).

Overall, over the course of 9,957 patient-years available for analysis, there were 149 episodes of a first CV event, the primary endpoint of the study (1.14 events per 100 patient-years). This total included 67 cases of acute/unstable coronary syndrome (myocardial infarction or unstable angina), 24 cases of stable angina, 26 cerebrovascular accidents, 10 transient ischemic attacks, 18 cases of peripheral vascular disease, and 4 cases of heart failure. A total of 15 of these events were fatal.

The sensitivity of the models was 68%-87% for the risk score cutoff value of 10% (marking the difference between low risk and intermediate to high risk) and 40%-65% for the risk score cutoff value of 20% (marking the difference between low- to intermediate-risk and high-risk patients). The corresponding specificity ranges were 55%-76% and 77%-88%, respectively. Those cutoff points are recommended to be used as indicators for CV preventive treatment such as lifestyle adjustments and drug therapy interventions.

For the SCORE risk model, the researchers found that CV risk predictions deviated from observed CV events in all risk deciles, but especially in the middle and top deciles. Indeed, tests of the fit of the SCORE model to the data with the Hosmer-Lemeshow (H-L) test yielded a P value of less than .001, indicating poor model fit (where good fit corresponds to P values greater than .05).

Looking at the Framingham model, the number of CV events predicted was similar to the observed number of CV events, showing a modest difference in predicted and observed CV risk in the lower- and middle-risk deciles, according to the authors. However, the top two deciles decidedly under- and overestimated risk, respectively, with an H-L test indicating poor fit for the whole model (P =.024).

The next tested model, the Reynolds risk score, underestimated the number of CV events, with an overall P value of .020 on the H-L test.

The QRisk II had a moderately good fit on the H-L test (P = .20) but still mainly overestimated the observed CV risk.

Ms. Arts stated in an e-mail that this study was originally presented at the 2013 American College of Rheumatology annual meeting. She added that she is currently collaborating with a group on an RA-specific CV risk model.

Ms. Arts disclosed that the study was partially funded by the Rheumatology Research University Nijmegen foundation. Three of her coauthors disclosed financial relationships with multiple pharmaceutical companies, including the makers of drugs and therapies for RA.

Established cardiovascular risk models such as the Systematic Coronary Risk Evaluation and the Framingham risk score either underestimated or overestimated the risk of cardiovascular events in a retrospective analysis of prospectively collected data from a cohort of patients with early rheumatoid arthritis.

These findings illustrate that what is needed is an RA-specific CV risk model whose performance "should be compared with the performance of the current risk algorithms," according to lead investigator Elke E.A. Arts of Radboud University Nijmegen (Netherlands) Medical Centre and her colleagues.

The four different risk algorithms that failed to correctly estimate the 10-year risk of a CV event were the Framingham risk score, the Systematic Coronary Risk Evaluation (SCORE), the Reynolds risk score, and the QRisk II risk score.

Proposals to fix to these models suggest that patients at risk might be better identified by adjusting the cutoff points in CV risk that are used as indications for primary prevention for RA patients, but "this could also lead to overtreatment, as the majority of patients in the lower risk group do not develop events," cautioned Ms. Arts, a PhD student at the university, and her associates.

They noted that "alternatively, a correction factor could be used to adjust the CV risk in patients with RA, as was suggested by the European League Against Rheumatism (EULAR) recommendations for CV risk management," they added. The EULAR recommendations advise multiplying the results of the SCORE risk algorithm by 1.5 for RA patients when they fulfill two out of the following three criteria: disease duration greater than 10 years, rheumatoid factor or anti–cyclic citrullinated peptide positivity, and extra-articular manifestations. However, "there are no data supporting such a multiplicator; it was based on expert opinion," they wrote.

In the current study, the investigators applied the four risk algorithms to data from 1,050 patients who were originally enrolled in the Nijmegen Early RA Inception cohort. Patients with preexisting CV disease were excluded, and in cases with less than 10 years of follow-up, risk scores were "adjusted proportionally according to the length of actual follow-up and calculated as a proportion of 10 years" (Ann. Rheum. Dis. 2014 Jan. 3 [doi: 10.1136/annrheumdis-2013-204024]).

Overall, over the course of 9,957 patient-years available for analysis, there were 149 episodes of a first CV event, the primary endpoint of the study (1.14 events per 100 patient-years). This total included 67 cases of acute/unstable coronary syndrome (myocardial infarction or unstable angina), 24 cases of stable angina, 26 cerebrovascular accidents, 10 transient ischemic attacks, 18 cases of peripheral vascular disease, and 4 cases of heart failure. A total of 15 of these events were fatal.

The sensitivity of the models was 68%-87% for the risk score cutoff value of 10% (marking the difference between low risk and intermediate to high risk) and 40%-65% for the risk score cutoff value of 20% (marking the difference between low- to intermediate-risk and high-risk patients). The corresponding specificity ranges were 55%-76% and 77%-88%, respectively. Those cutoff points are recommended to be used as indicators for CV preventive treatment such as lifestyle adjustments and drug therapy interventions.

For the SCORE risk model, the researchers found that CV risk predictions deviated from observed CV events in all risk deciles, but especially in the middle and top deciles. Indeed, tests of the fit of the SCORE model to the data with the Hosmer-Lemeshow (H-L) test yielded a P value of less than .001, indicating poor model fit (where good fit corresponds to P values greater than .05).

Looking at the Framingham model, the number of CV events predicted was similar to the observed number of CV events, showing a modest difference in predicted and observed CV risk in the lower- and middle-risk deciles, according to the authors. However, the top two deciles decidedly under- and overestimated risk, respectively, with an H-L test indicating poor fit for the whole model (P =.024).

The next tested model, the Reynolds risk score, underestimated the number of CV events, with an overall P value of .020 on the H-L test.

The QRisk II had a moderately good fit on the H-L test (P = .20) but still mainly overestimated the observed CV risk.

Ms. Arts stated in an e-mail that this study was originally presented at the 2013 American College of Rheumatology annual meeting. She added that she is currently collaborating with a group on an RA-specific CV risk model.

Ms. Arts disclosed that the study was partially funded by the Rheumatology Research University Nijmegen foundation. Three of her coauthors disclosed financial relationships with multiple pharmaceutical companies, including the makers of drugs and therapies for RA.