Curr Atheroscler Rep (2014) 16:427 DOI 10.1007/s11883-014-0427-z
CARDIOVASCULAR DISEASE AND STROKE (P PERRONE-FILARDI AND S. AGEWALL, SECTION EDITORS)
New Risk Markers for Cardiovascular Prevention Guy G. De Backer
# Springer Science+Business Media New York 2014
Abstract The importance of total cardiovascular (CV) risk estimation before management decisions are taken is well established. Models have been developed that allow physicians to stratify the asymptomatic population in subgroups at low, moderate, high, and very high total CV risk. Most models are based on classical CV risk factors: age, gender, smoking, blood pressure, and lipid levels. The impact of additional risk factors is discussed here, looking separately at the predictive increments of novel biomarkers and of indicators of subclinical atherosclerotic disease. The contribution of biomarkers to the total CV risk estimation is generally modest, and their usage should be limited to subjects at intermediate total CV risk. Detection of subclinical vascular damage may improve total CV risk estimation in asymptomatic subjects who are close to a threshold that could affect management decisions and in whom the chances of re-classification in a different risk category are great. There is, however, an urgent need for trials in which the value of using total CV risk estimation models is tested. Keywords Cardiovascular disease . Prevention . Risk factors . Prediction
Introduction For many years, different expert committees have recommended using estimates of total cardiovascular (CV) risk for the prevention of cardiovascular disease (CVD)[1, 2, 3•, 4, 5•]. The reason for that is to tailor the preventive strategy in This article is part of the Topical Collection on Cardiovascular Disease and Stroke G. G. De Backer (*) Ghent University, Ghent, Belgium, University Hospital, De Pintelaan 185 B, 9000 Gent, Belgium e-mail: [email protected]
accordance with the total CV risk of the individual: the higher the patient’s total CV risk, the more intense should be the preventive approach. For particular patient groups it is known that the CV risk is high or very high and no special risk estimation models are needed to know that; this is the case for patients with established CVD, for patients with diabetes mellitus or with chronic kidney disease, or for subjects with severe elevation of single CV risk factors, such as familial hypercholesterolaemia or severe arterial hypertension. But a majority of the new CV events that occur in the community comes from people free of all these conditions. In them, the total CV risk can be elevated because of the combination of different CV risk factors that interact with each other in a complex way. Reasons why total CV risk should be estimated in them are summarized in Table 1. To estimate the total CV risk in these apparently healthy asymptomatic people, risk estimation models are needed and have been developed on the basis of observations in large cohort studies. Tools based on results from the Framingham study in the US [6, 7] have been developed first, but have some limitations related to numbers, to the study population on which the models are based, and in some models to the ‘soft’ character of some outcome measures. Several other models have been developed in recent years [8–13], among which the SCORE model [8] in Europe, with the advantage that it is based on a very large population base and that it estimates the chances of developing fatal CV events which can be reproduced, and allows the calibration of the model at the level of the nation on the condition of the existence of reliable mortality statistics. All these models include non-modifiable CV risk factors (age and gender) and modifiable risk factors such as smoking habits, blood pressure, and total cholesterol; in other models additional factors have been included such as family history, diabetes, HDL cholesterol ( HDL-C), high-sensitivity C-reactive protein (hs-CRP), or indicators of social deprivation [9–11].
427, Page 2 of 8 Table 1 Why should one estimate total CV risk?
Curr Atheroscler Rep (2014) 16:427 Table 2 Quality criteria to evaluate new risk markers for the estimation of total CV risk in the asymptomatic population
Total CV risk estimation is of fundamental importance because: CVD is multifactorial in origin CV risk factors interact synergistically Clinicians treat patients, not isolated risk factors Preventive actions should be guided in accordance to the total CV risk level.
The question can be raised if the risk estimation that is obtained with these models is fit for purpose, should we aim for better predictive tools using additional risk markers? What “added value” can we expect by including novel risk markers in the existing risk scores? How to address all these questions has been well summarized by M.Romanens et al [14]. In approaching these questions we should not lose sight of the original aim of using these risk estimation models: to stratify the apparently healthy population in subgroups according to their total CV risk; preventive strategies should be different for people at low, moderate, high, or very high total CV risk. But these risk estimation models should not be used to estimate the total CV risk at the level of an individual person. CV risk estimation of a given person and expressed in terms of absolute risk remains hazardous and very approximate; risk models are not crystal balls for prophesying [15]. To improve total CV risk estimation, one could start by including in the model an average of multiple measures of the traditional risk factors (total cholesterol and blood pressure) rather than results based on a single measurement [16, 17]. Among the novel risk markers that could improve risk estimation are several candidates related to lifestyles, to lipid metabolism, inflammation, thrombosis, genetic predisposition, vascular function, or neurohumoral activity. One could also include in these models markers that reflect the existence of subclinical atherosclerotic disease that can be detected with different methods. What is the value of introducing these ‘novel’ risk markers in existing risk estimation models? Criteria that can be used to evaluate the quality of new risk markers for the estimation of total CV risk in the asymptomatic population are summarized in Table 2. It is insufficient to demonstrate the predictive value of these factors, independent of the conventional CV risk factors. This on its own is no proof of the incremental value of the marker in the risk estimation tool. Novel biomarkers should add reliably to the predictive power of traditional CV risk factors; it should also lead to a better calibration with no significant differences between predicted and observed CVD event rates. It should result in a significant net reclassification improvement index (NRI), particularly in the large group of subjects at intermediate CV risk in whom the clinical NRI (CNRI) can be estimated. Particular interest should
- The measurement of the risk marker should be easy, accurate and reliable - The risk marker should predict CVD independently of other risk factors in persons free of CVD - The risk marker should provide incremental information beyond what can be obtained with existing risk estimation models. - Adding the risk marker should influence management strategies in a way that CVD prevention becomes more efficacious and more cost-effective.
go into modifiable novel risk markers that could become part of an intervention program; the latter should then be tested in a randomized controlled trial (RCT)regarding its potential in preventing clinical outcomes, safety as well as cost-effectiveness. Although it must be admitted that the application of the existing risk estimation tools has also never been tested in an RCT as to its value in preventing ‘hard’ CVD events compared to standard of care.
Predictive Increments of Novel Biomarkers at the Level of the Community For the estimation of the total CV risk, the incremental value of new biomarkers that can be measured on easily obtainable material such as blood or urine has been examined in numerous large population studies. This has been studied by adding new markers to different existing risk estimation models that have been reviewed by G. Siontis et al.[18] . One can reasonably assume that biomarkers improving the predictive power of one model will also do so for other models that are based on similar traditional CV risk factors. For appraising the improved predictive value of adding a biomarker to an established risk estimation model, different statistics should be used as proposed by others [14, 19••]. These include changes in C-statistics derived from the area under the receiver operating curve (AUROC), indices of calibration, the NRI, and the CNRI. The Cstatistic on its own is of limited value in models that already have a substantial AUROC. The SCORE model with age, gender, smoking, systolic blood pressure, and total cholesterol resulted already in a C-statistic of 0.81. This leaves little room for improvement through the introduction of new biomarkers. It has been suggested that a reasonable improvement in discrimination can only be expected for variables that are strongly associated with CVD with an odds ratio of 16 or more [20].
Curr Atheroscler Rep (2014) 16:427
Page 3 of 8, 427
The Case of HDL-C The predictive increment of HDL-C in risk estimation has been re-examined using the SCORE database [21]. This study has shown that HDL-C can contribute to risk estimation if entered as an independent variable. HDL-C levels modified risk at all levels of SCORE risk and this effect was seen in both sexes and in all age groups. It resulted in a modest improvement of the risk estimation with a statistically significant, but very modest change in C-statistic of 0.006 points. The addition of HDL-C to the SCORE model did, however, improve the NRI (see Table 3). In women from high-risk countries, a significant proportion was correctly re-classified by introducing HDL-C; the NRI in them was 11.5 % (p=0.015), and this was mainly due to a reclassification towards higher risk categories [21]. In other studies of the predictive increment of HDL-C, the NRI was consistently reported different from zero with large variations in the size of the increment from 1.7 % [22] to 12.1 % [23]. From this exercise with HDL-C in SCORE it looks as if including novel risk markers may be useful in estimating the total CV risk in the subgroup at intermediate risk just below the threshold for intensive risk modification of the middleaged population. SCORE charts incorporating different levels of HDL-C are included in the EAS/ESC guidelines on the management of dyslipidaemias [4].
Other Biomarkers a) The role of elevated triglycerides as a predictor of CVD has been discussed for many years. Fasting triglycerides relate to risk in univariate analyses but the effect is attenuated by adjustment for other factors, especially HDL-C [24]. More recently, attention has focused on non-fasting triglycerides, which may be more strongly related to risk independent of the effects of HDL-C [25]. Table 3 Net reclassification improvement (% reclassified into high [ >= 5 %] or low [=30 kg/m2) and a parental history of myocardial infarction to a model based on classical risk factors, yielded in a significant NRI of 5.5 % in men and a nonsignificant NRI of 3.3 % in women when a 10-year estimated risk of total CV events was categorized using 10 % cut-off values; the improvements were mainly due to obesity [27]. Therefore, it may well be that by including BMI as a continuous variable does not improve the NRI while it does with the BMI dichotomized into those who are obese or not. The Belgian Health Care Knowledge Centre (KCE) published a report in 2013 on ‘novel serum biomarkers for the prediction of CV risk’ [28]. Based on a systematic review of the literature from 2008 to 2012, 167 references were identified and 16 studies were included in the review. Studies that were published before 2008 had been discussed in the excellent overview published by MJ. Pencina et al. in 2008 [23]. Based on the results in these 16 studies, the predictive increments of various biomarkers were tested among which markers of inflammation: (hs-CRP, fibrinogen), markers of lipid metabolism (apolipoprotein A1, apolipoprotein B100, phospholipase A2, HDL cholesterol), and a variety of markers such as N-terminal proB-type natriuretic peptide (NT-proBNP), uric acid, homocysteine, and troponin. c) hs-CRP had been investigated in a majority of identified studies. Two of these were systematic reviews of the addition of hs-CRP to risk estimation models which partially overlapped [29, 30]; the first included 23 studies the other 31. In addition, they identified 12 original studies published since 2008. In the first review [29], it was demonstrated that hsCRP was associated with CVD, independently from the risk factors that were included in the Framingham model and there was evidence of a dose-response gradient. But in only one study risk reclassification was calculated. In that study [31], 14 % of all participants originally stratified into the Framingham risk group of 10 - 20 % chances of developing CVD/10 years, were re-classified as low-risk (20 %). In the second review [30], 13 studies reported on the effect on the AUROC of adding hs-CRP to the Framingham model; five reported no change and eight reported an improvement of the C-statistic varying from + 0.01 to +0.15 points. From the 12 original studies published since 2008 it could be concluded that adding hs-CRP to the Framingham or the ASSIGN models resulted in a
427, Page 4 of 8
significant but modest reclassification of participants (NRI ranging from 1.52 to 11.8 %). CNRI was consistently greater than NRI. This may partially be explained by the fact that, in intermediate-risk persons, other risk factors such as smoking or advanced age are less prevalent leaving more room for a larger contribution of hs-CRP. The NRI was also consistently higher when coronary heart disease (CHD) was considered an outcome compared with CVD. In the intermediate-risk participants of the MESA study, the addition of hs-CRP to the FRS resulted in an increase of the C statistic from 0.623 to 0.640(p=0.03) and in an NRI of 8 % for the incidence of coronary events [32•]. d) In the report of the KCE [28], two studies were identified regarding the predictive increments of fibrinogen [22, 33]; the findings were to a certain extent comparable to what was observed for hsCRP: a modest overall NRI and a greater CNRI, although the latter was mainly due to downreclassification. Adding fibrinogen to the model on top of hs-CRP did not result in a change in C-statistic more than when either marker was used alone [22]. e) Among other biomarkers the systematic review of the KCE [28] retained six original studies which assessed the predictive increment of NT-proBNP, homocysteine, uric acid, and troponin. NT-proBNP was the only one to improve substantially discrimination and reclassification when added to the Framingham model in five studies out of six [33–35]. In the cardiovascular cohort of the Malmo Diet and Cancer study [35], the NT-proBNP was added to a model with the conventional CV risk factors; NT-proBNP predicted CV and coronary events but the changes in Cstatistics were small; the NRIs were not significant. By limiting the analyses to participants at moderate CV risk the NRI was 7 % for CV events and 15 % for coronary events, but this was mainly a re-classification towards lower risk groups with limited clinical relevance . In a systematic overview published in 2009, of the predictive increments of nine emerging risk factors, among which there were five biomarkers (hs-CRP, Lp(a), homocysteine, leukocyte count, fasting glycaemia), it was concluded that the current evidence did not support the routine use of any of these risk factors for further stratification of intermediate-risk persons [36]. f) In the Scottish Heart Health Extended Cohort (SHHEC) study it was possible to measure troponin I levels in nearly 75 % of the study population using a high-sensitivity assay. In models adjusted for sex, cohort, and all risk factors from the ASSIGN model, troponin I improved the prediction of new CV events significantly over a 20-
Curr Atheroscler Rep (2014) 16:427
year period; the NRI was estimated at 3.2 % (p=0.0005) for CV events and at 4.5 % (p=0.012) for coronary death. The C-statistics improved also significantly [37•].
Predictive Increments of Multiple Novel Biomarkers Together But what if multiple novel risk markers are introduced together on top of the traditional risk factors? Does the inclusion of several new biomarkers simultaneously have clinical utility? The results from studies on this subject are rather inconsistent: a) In the Women’s Health Initiative, a moderate improvement in CHD risk prediction was observed when an 18biomarker panel was added to different prediction models, including the Framingham Risk Score (FRS), in postmenopausal women; five out of the 18 new risk markers improved the C-statistic with 0.02 and an NRI of 6.5 % [38]. b) In the Framingham Heart Study, a multimarker score including NT-proBNP and urinary albumin/creatinine ratio moderately improved the C-statistic by 0.01 in the prediction of CVD events [39]. c) In the Cardiovascular Health Study, the addition of six biomarkers – CRP, fibrinogen, factor VIIIc, interleukin-6, Lp(a), and Hgb – did not improve discrimination beyond established risk factors [40]. d) In the prospective Atherosclerosis Risk in Communities ( ARIC) Study, the basic model included traditional risk factors; the hs-CRP did not add significantly to the AUROC and neither did most of the 19 other novel markers that were tested [41]. e) In the Uppsala Longitudinal Study of adult men, the C statistic for CVD death prediction increased by 0.06 when four markers – troponin, N-BNP, cystatin-C and CRP– were added to established markers in the subgroup free of CVD at baseline [42]. f) In the MORGAM study, 30 novel biomarkers were examined in two cohort studies; several of them were associated with the incidence of CVD, but this was only consistent in both cohorts and in men and women for NT-proBNP, hs-CRP, and troponin I; with these three markers, a score was developed and the addition of this score to a conventional risk model resulted in improved Cstatistics (from 0.67 to 0.69, p=0.004) and in an NRI of 11 % ( p
CARDIOVASCULAR DISEASE AND STROKE (P PERRONE-FILARDI AND S. AGEWALL, SECTION EDITORS)
New Risk Markers for Cardiovascular Prevention Guy G. De Backer
# Springer Science+Business Media New York 2014
Abstract The importance of total cardiovascular (CV) risk estimation before management decisions are taken is well established. Models have been developed that allow physicians to stratify the asymptomatic population in subgroups at low, moderate, high, and very high total CV risk. Most models are based on classical CV risk factors: age, gender, smoking, blood pressure, and lipid levels. The impact of additional risk factors is discussed here, looking separately at the predictive increments of novel biomarkers and of indicators of subclinical atherosclerotic disease. The contribution of biomarkers to the total CV risk estimation is generally modest, and their usage should be limited to subjects at intermediate total CV risk. Detection of subclinical vascular damage may improve total CV risk estimation in asymptomatic subjects who are close to a threshold that could affect management decisions and in whom the chances of re-classification in a different risk category are great. There is, however, an urgent need for trials in which the value of using total CV risk estimation models is tested. Keywords Cardiovascular disease . Prevention . Risk factors . Prediction
Introduction For many years, different expert committees have recommended using estimates of total cardiovascular (CV) risk for the prevention of cardiovascular disease (CVD)[1, 2, 3•, 4, 5•]. The reason for that is to tailor the preventive strategy in This article is part of the Topical Collection on Cardiovascular Disease and Stroke G. G. De Backer (*) Ghent University, Ghent, Belgium, University Hospital, De Pintelaan 185 B, 9000 Gent, Belgium e-mail: [email protected]
accordance with the total CV risk of the individual: the higher the patient’s total CV risk, the more intense should be the preventive approach. For particular patient groups it is known that the CV risk is high or very high and no special risk estimation models are needed to know that; this is the case for patients with established CVD, for patients with diabetes mellitus or with chronic kidney disease, or for subjects with severe elevation of single CV risk factors, such as familial hypercholesterolaemia or severe arterial hypertension. But a majority of the new CV events that occur in the community comes from people free of all these conditions. In them, the total CV risk can be elevated because of the combination of different CV risk factors that interact with each other in a complex way. Reasons why total CV risk should be estimated in them are summarized in Table 1. To estimate the total CV risk in these apparently healthy asymptomatic people, risk estimation models are needed and have been developed on the basis of observations in large cohort studies. Tools based on results from the Framingham study in the US [6, 7] have been developed first, but have some limitations related to numbers, to the study population on which the models are based, and in some models to the ‘soft’ character of some outcome measures. Several other models have been developed in recent years [8–13], among which the SCORE model [8] in Europe, with the advantage that it is based on a very large population base and that it estimates the chances of developing fatal CV events which can be reproduced, and allows the calibration of the model at the level of the nation on the condition of the existence of reliable mortality statistics. All these models include non-modifiable CV risk factors (age and gender) and modifiable risk factors such as smoking habits, blood pressure, and total cholesterol; in other models additional factors have been included such as family history, diabetes, HDL cholesterol ( HDL-C), high-sensitivity C-reactive protein (hs-CRP), or indicators of social deprivation [9–11].
427, Page 2 of 8 Table 1 Why should one estimate total CV risk?
Curr Atheroscler Rep (2014) 16:427 Table 2 Quality criteria to evaluate new risk markers for the estimation of total CV risk in the asymptomatic population
Total CV risk estimation is of fundamental importance because: CVD is multifactorial in origin CV risk factors interact synergistically Clinicians treat patients, not isolated risk factors Preventive actions should be guided in accordance to the total CV risk level.
The question can be raised if the risk estimation that is obtained with these models is fit for purpose, should we aim for better predictive tools using additional risk markers? What “added value” can we expect by including novel risk markers in the existing risk scores? How to address all these questions has been well summarized by M.Romanens et al [14]. In approaching these questions we should not lose sight of the original aim of using these risk estimation models: to stratify the apparently healthy population in subgroups according to their total CV risk; preventive strategies should be different for people at low, moderate, high, or very high total CV risk. But these risk estimation models should not be used to estimate the total CV risk at the level of an individual person. CV risk estimation of a given person and expressed in terms of absolute risk remains hazardous and very approximate; risk models are not crystal balls for prophesying [15]. To improve total CV risk estimation, one could start by including in the model an average of multiple measures of the traditional risk factors (total cholesterol and blood pressure) rather than results based on a single measurement [16, 17]. Among the novel risk markers that could improve risk estimation are several candidates related to lifestyles, to lipid metabolism, inflammation, thrombosis, genetic predisposition, vascular function, or neurohumoral activity. One could also include in these models markers that reflect the existence of subclinical atherosclerotic disease that can be detected with different methods. What is the value of introducing these ‘novel’ risk markers in existing risk estimation models? Criteria that can be used to evaluate the quality of new risk markers for the estimation of total CV risk in the asymptomatic population are summarized in Table 2. It is insufficient to demonstrate the predictive value of these factors, independent of the conventional CV risk factors. This on its own is no proof of the incremental value of the marker in the risk estimation tool. Novel biomarkers should add reliably to the predictive power of traditional CV risk factors; it should also lead to a better calibration with no significant differences between predicted and observed CVD event rates. It should result in a significant net reclassification improvement index (NRI), particularly in the large group of subjects at intermediate CV risk in whom the clinical NRI (CNRI) can be estimated. Particular interest should
- The measurement of the risk marker should be easy, accurate and reliable - The risk marker should predict CVD independently of other risk factors in persons free of CVD - The risk marker should provide incremental information beyond what can be obtained with existing risk estimation models. - Adding the risk marker should influence management strategies in a way that CVD prevention becomes more efficacious and more cost-effective.
go into modifiable novel risk markers that could become part of an intervention program; the latter should then be tested in a randomized controlled trial (RCT)regarding its potential in preventing clinical outcomes, safety as well as cost-effectiveness. Although it must be admitted that the application of the existing risk estimation tools has also never been tested in an RCT as to its value in preventing ‘hard’ CVD events compared to standard of care.
Predictive Increments of Novel Biomarkers at the Level of the Community For the estimation of the total CV risk, the incremental value of new biomarkers that can be measured on easily obtainable material such as blood or urine has been examined in numerous large population studies. This has been studied by adding new markers to different existing risk estimation models that have been reviewed by G. Siontis et al.[18] . One can reasonably assume that biomarkers improving the predictive power of one model will also do so for other models that are based on similar traditional CV risk factors. For appraising the improved predictive value of adding a biomarker to an established risk estimation model, different statistics should be used as proposed by others [14, 19••]. These include changes in C-statistics derived from the area under the receiver operating curve (AUROC), indices of calibration, the NRI, and the CNRI. The Cstatistic on its own is of limited value in models that already have a substantial AUROC. The SCORE model with age, gender, smoking, systolic blood pressure, and total cholesterol resulted already in a C-statistic of 0.81. This leaves little room for improvement through the introduction of new biomarkers. It has been suggested that a reasonable improvement in discrimination can only be expected for variables that are strongly associated with CVD with an odds ratio of 16 or more [20].
Curr Atheroscler Rep (2014) 16:427
Page 3 of 8, 427
The Case of HDL-C The predictive increment of HDL-C in risk estimation has been re-examined using the SCORE database [21]. This study has shown that HDL-C can contribute to risk estimation if entered as an independent variable. HDL-C levels modified risk at all levels of SCORE risk and this effect was seen in both sexes and in all age groups. It resulted in a modest improvement of the risk estimation with a statistically significant, but very modest change in C-statistic of 0.006 points. The addition of HDL-C to the SCORE model did, however, improve the NRI (see Table 3). In women from high-risk countries, a significant proportion was correctly re-classified by introducing HDL-C; the NRI in them was 11.5 % (p=0.015), and this was mainly due to a reclassification towards higher risk categories [21]. In other studies of the predictive increment of HDL-C, the NRI was consistently reported different from zero with large variations in the size of the increment from 1.7 % [22] to 12.1 % [23]. From this exercise with HDL-C in SCORE it looks as if including novel risk markers may be useful in estimating the total CV risk in the subgroup at intermediate risk just below the threshold for intensive risk modification of the middleaged population. SCORE charts incorporating different levels of HDL-C are included in the EAS/ESC guidelines on the management of dyslipidaemias [4].
Other Biomarkers a) The role of elevated triglycerides as a predictor of CVD has been discussed for many years. Fasting triglycerides relate to risk in univariate analyses but the effect is attenuated by adjustment for other factors, especially HDL-C [24]. More recently, attention has focused on non-fasting triglycerides, which may be more strongly related to risk independent of the effects of HDL-C [25]. Table 3 Net reclassification improvement (% reclassified into high [ >= 5 %] or low [=30 kg/m2) and a parental history of myocardial infarction to a model based on classical risk factors, yielded in a significant NRI of 5.5 % in men and a nonsignificant NRI of 3.3 % in women when a 10-year estimated risk of total CV events was categorized using 10 % cut-off values; the improvements were mainly due to obesity [27]. Therefore, it may well be that by including BMI as a continuous variable does not improve the NRI while it does with the BMI dichotomized into those who are obese or not. The Belgian Health Care Knowledge Centre (KCE) published a report in 2013 on ‘novel serum biomarkers for the prediction of CV risk’ [28]. Based on a systematic review of the literature from 2008 to 2012, 167 references were identified and 16 studies were included in the review. Studies that were published before 2008 had been discussed in the excellent overview published by MJ. Pencina et al. in 2008 [23]. Based on the results in these 16 studies, the predictive increments of various biomarkers were tested among which markers of inflammation: (hs-CRP, fibrinogen), markers of lipid metabolism (apolipoprotein A1, apolipoprotein B100, phospholipase A2, HDL cholesterol), and a variety of markers such as N-terminal proB-type natriuretic peptide (NT-proBNP), uric acid, homocysteine, and troponin. c) hs-CRP had been investigated in a majority of identified studies. Two of these were systematic reviews of the addition of hs-CRP to risk estimation models which partially overlapped [29, 30]; the first included 23 studies the other 31. In addition, they identified 12 original studies published since 2008. In the first review [29], it was demonstrated that hsCRP was associated with CVD, independently from the risk factors that were included in the Framingham model and there was evidence of a dose-response gradient. But in only one study risk reclassification was calculated. In that study [31], 14 % of all participants originally stratified into the Framingham risk group of 10 - 20 % chances of developing CVD/10 years, were re-classified as low-risk (20 %). In the second review [30], 13 studies reported on the effect on the AUROC of adding hs-CRP to the Framingham model; five reported no change and eight reported an improvement of the C-statistic varying from + 0.01 to +0.15 points. From the 12 original studies published since 2008 it could be concluded that adding hs-CRP to the Framingham or the ASSIGN models resulted in a
427, Page 4 of 8
significant but modest reclassification of participants (NRI ranging from 1.52 to 11.8 %). CNRI was consistently greater than NRI. This may partially be explained by the fact that, in intermediate-risk persons, other risk factors such as smoking or advanced age are less prevalent leaving more room for a larger contribution of hs-CRP. The NRI was also consistently higher when coronary heart disease (CHD) was considered an outcome compared with CVD. In the intermediate-risk participants of the MESA study, the addition of hs-CRP to the FRS resulted in an increase of the C statistic from 0.623 to 0.640(p=0.03) and in an NRI of 8 % for the incidence of coronary events [32•]. d) In the report of the KCE [28], two studies were identified regarding the predictive increments of fibrinogen [22, 33]; the findings were to a certain extent comparable to what was observed for hsCRP: a modest overall NRI and a greater CNRI, although the latter was mainly due to downreclassification. Adding fibrinogen to the model on top of hs-CRP did not result in a change in C-statistic more than when either marker was used alone [22]. e) Among other biomarkers the systematic review of the KCE [28] retained six original studies which assessed the predictive increment of NT-proBNP, homocysteine, uric acid, and troponin. NT-proBNP was the only one to improve substantially discrimination and reclassification when added to the Framingham model in five studies out of six [33–35]. In the cardiovascular cohort of the Malmo Diet and Cancer study [35], the NT-proBNP was added to a model with the conventional CV risk factors; NT-proBNP predicted CV and coronary events but the changes in Cstatistics were small; the NRIs were not significant. By limiting the analyses to participants at moderate CV risk the NRI was 7 % for CV events and 15 % for coronary events, but this was mainly a re-classification towards lower risk groups with limited clinical relevance . In a systematic overview published in 2009, of the predictive increments of nine emerging risk factors, among which there were five biomarkers (hs-CRP, Lp(a), homocysteine, leukocyte count, fasting glycaemia), it was concluded that the current evidence did not support the routine use of any of these risk factors for further stratification of intermediate-risk persons [36]. f) In the Scottish Heart Health Extended Cohort (SHHEC) study it was possible to measure troponin I levels in nearly 75 % of the study population using a high-sensitivity assay. In models adjusted for sex, cohort, and all risk factors from the ASSIGN model, troponin I improved the prediction of new CV events significantly over a 20-
Curr Atheroscler Rep (2014) 16:427
year period; the NRI was estimated at 3.2 % (p=0.0005) for CV events and at 4.5 % (p=0.012) for coronary death. The C-statistics improved also significantly [37•].
Predictive Increments of Multiple Novel Biomarkers Together But what if multiple novel risk markers are introduced together on top of the traditional risk factors? Does the inclusion of several new biomarkers simultaneously have clinical utility? The results from studies on this subject are rather inconsistent: a) In the Women’s Health Initiative, a moderate improvement in CHD risk prediction was observed when an 18biomarker panel was added to different prediction models, including the Framingham Risk Score (FRS), in postmenopausal women; five out of the 18 new risk markers improved the C-statistic with 0.02 and an NRI of 6.5 % [38]. b) In the Framingham Heart Study, a multimarker score including NT-proBNP and urinary albumin/creatinine ratio moderately improved the C-statistic by 0.01 in the prediction of CVD events [39]. c) In the Cardiovascular Health Study, the addition of six biomarkers – CRP, fibrinogen, factor VIIIc, interleukin-6, Lp(a), and Hgb – did not improve discrimination beyond established risk factors [40]. d) In the prospective Atherosclerosis Risk in Communities ( ARIC) Study, the basic model included traditional risk factors; the hs-CRP did not add significantly to the AUROC and neither did most of the 19 other novel markers that were tested [41]. e) In the Uppsala Longitudinal Study of adult men, the C statistic for CVD death prediction increased by 0.06 when four markers – troponin, N-BNP, cystatin-C and CRP– were added to established markers in the subgroup free of CVD at baseline [42]. f) In the MORGAM study, 30 novel biomarkers were examined in two cohort studies; several of them were associated with the incidence of CVD, but this was only consistent in both cohorts and in men and women for NT-proBNP, hs-CRP, and troponin I; with these three markers, a score was developed and the addition of this score to a conventional risk model resulted in improved Cstatistics (from 0.67 to 0.69, p=0.004) and in an NRI of 11 % ( p
