If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Corresponding author at: Julius Centre University of Malaya, Department of Social and Preventive Medicine, Faculty of Medicine, University of Malaya, 50603 Kuala Lumpur, Malaysia. Tel.: +60 3 7967 7895; fax: +60 3 7967 4975.
Affiliations
Clinical Research Centre, Kuala Lumpur Hospital, Jalan Pahang, 50586 Kuala Lumpur, MalaysiaJulius Centre University of Malaya, Department of Social and Preventive Medicine, University of Malaya, 50603 Kuala Lumpur, MalaysiaJulius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Utrecht, The Netherlands
Four cardiovascular risk prediction models were assessed in an Asian population.
•
The WHO/ISH model performed poorly for cardiovascular risk stratification.
•
The Framingham and SCORE models could stratify risk in Asian men and women.
•
The SCORE-high model accurately predicted risk for men, but not women.
•
The Framingham model stratified risk better than the SCORE models in women.
Abstract
Background
Cardiovascular risk-prediction models are used in clinical practice to identify and treat high-risk populations, and to communicate risk effectively. We assessed the validity and utility of four cardiovascular risk-prediction models in an Asian population of a middle-income country.
Methods
Data from a national population-based survey of 14,863 participants aged 40 to 65 years, with a follow-up duration of 73,277 person-years was used. The Framingham Risk Score (FRS), SCORE (Systematic COronary Risk Evaluation)-high and -low cardiovascular-risk regions and the World Health Organization/International Society of Hypertension (WHO/ISH) models were assessed. The outcome of interest was 5-year cardiovascular mortality. Discrimination was assessed for all models and calibration for the SCORE models.
Results
Cardiovascular risk factors were highly prevalent; smoking 20%, obesity 32%, hypertension 55%, diabetes mellitus 18% and hypercholesterolemia 34%. The FRS and SCORE models showed good agreement in risk stratification. The FRS, SCORE-high and -low models showed good discrimination for cardiovascular mortality, areas under the ROC curve (AUC) were 0.768, 0.774 and 0.775 respectively. The WHO/ISH model showed poor discrimination, AUC = 0.613. Calibration of the SCORE-high model was graphically and statistically acceptable for men (χ2 goodness-of-fit, p = 0.097). The SCORE-low model was statistically acceptable for men (χ2 goodness-of-fit, p = 0.067). Both SCORE-models underestimated risk in women (p < 0.001).
Conclusions
The FRS and SCORE-high models, but not the WHO/ISH model can be used to identify high cardiovascular risk in the Malaysian population. The SCORE-high model predicts risk accurately in men but underestimated it in women.
Cardiovascular risk prediction models are important in the prevention and management of cardiovascular diseases. These models are used in clinical practice to identify and treat high-risk populations as well as to communicate risk effectively [
]. Currently, there are three cardiovascular risk prediction models recommended in the Malaysian clinical practice guidelines for the prevention of cardiovascular diseases; the Framingham Risk Score (FRS), SCORE (Systematic COronary Risk Evaluation) and the World Health Organization/International Society of Hypertension (WHO/ISH) models [
There are various concerns when adopting a risk prediction model for the clinical assessment of a patient to determine treatment options. First, is the risk score applicable to the local patient setting? It is well known that the underlying incidence of disease and prevalence of its risk factors determines the suitability of any risk prediction model. Secondly, can the risk prediction model be calibrated? In other words, can it be fully assessed of its clinical utility to predict risk accurately in the local patient setting? Unfortunately, developing countries often lack the information on cardiovascular events that are required for a full calibration of cardiovascular risk prediction models.
In Malaysia, a middle-income multi-ethnic developing country, both factors discussed above are applicable. Thus, these questions remain; 1) Are all the recommended cardiovascular risk-prediction scores applicable in the Asian patient setting? 2) Do they accurately stratify risks? With these concerns in mind, we sought to compare the cardiovascular risk prediction models that are recommended in local clinical practice guidelines; the FRS, high and low risk SCORE and WHO/ISH risk-prediction models.
2. Methods
The population dataset from the 2006 National Health and Morbidity Survey (NHMS) was used in this study. The NHMS was a nationwide cross-sectional population-based survey that assessed cardiovascular risk factors among other health and social indicators. Details of the study have been published elsewhere and the measurement of cardiovascular risk factors has been described in detail [
]. For this study, all NHMS participants aged between 40 and 65 years were selected to ensure comparability between the different risk models assessed. Those with known history of cardiovascular disease were excluded.
Ethics approval was obtained from the Malaysian Medical Research and Ethics Committee (NMRR ID-10-731-6916). This study complies with the Declaration of Helsinki.
2.1 Cardiovascular risk prediction models
Four cardiovascular risk prediction models were assessed; the FRS for global cardiovascular risk [
] models use information on age, sex, systolic blood pressure and smoking. The SCORE and WHO/ISH models include information on total cholesterol whereas the FRS model in the present study uses body mass index instead of the ratio of total and high-density lipoprotein (HDL) cholesterol. This is because the NHMS did not collect information on HDL cholesterol. Both the FRS and WHO/ISH include diabetes in the model.
For both the FRS and SCORE models, the originally developed and validated predictors and coefficients were used to calculate the predicted cardiovascular risk in the present study. The cardiovascular risk prediction for the FRS was calculated using the Framingham equations for general cardiovascular risk provided online (http://www.framinghamheartstudy.org/risk-functions/cardiovascular-disease/10-year-risk.php). The FRS equation used a Cox proportional hazards model [
]. The original SCORE models do not take diabetes status into account but it was recommended that risks be multiplied two-fold for men and four-fold for women with diabetes [
]. This is because cardiovascular risk in diabetics is higher than non-diabetics; almost in parallel for every conventional cardiovascular risk factor [
]. In our study, the estimated risk in diabetics was multiplied by three in men, and by five in women as this is the current estimated impact of diabetes on cardiovascular risk [
]. Both SCORE models were used, that for populations at high and low cardiovascular risk populations. This is because it is unknown which model would perform better in Malaysia. The WHO/ISH risk prediction model does not provide regression equations that can estimate absolute risk for individuals [
]. The WHO/ISH risk prediction model has risk charts with five categories of risk. In this study, each subject's risk category was calculated.
2.2 Cardiovascular risk stratification
For each model, cardiovascular risk was stratified into three categories; low, intermediate and high cardiovascular risk. High cardiovascular risk was defined as ten-year risk of ≥20%, ≥5% and ≥30% for the FRS, SCORE and WHO/ISH models, respectively [
]. All other values were in the intermediate risk group. Spearman's correlation coefficient was used to assess the correlation between the rankings of each subject's absolute cardiovascular risk. Agreement between different models was determined by lower numbers of population misclassification in extremes of risk categories.
2.3 Outcome of interest for model performance
The 5-year risk of cardiovascular mortality was the outcome of interest in this study. Non-fatal cardiovascular events were not considered, because despite a relatively good healthcare system, estimates of incidence of coronary heart disease, stroke and other cardiovascular events assessed by many cardiovascular risk-prediction models cannot be obtained. Cardiovascular causes of death were fatal events described in the International Classification of Diseases (ICD)-10 codes I10–I15 (hypertensive diseases), I20–25 (ischemic heart diseases), I60–I69 (cerebrovascular diseases), I70 and I71 (other atherosclerosis), which were used in the SCORE models [
Mortality data from 1st January 2006 till 31st December 2010 for the NHMS population were obtained via record linkages with the Malaysian National Registration Department. All Malaysians have a unique numerical identification number given at birth by the National Registration Department. This unique identification number is used for all official matters, including death registrations. Death registrations are compulsory in the country. The IBM® InfoSphere® QualityStage (http://www-01.ibm.com/software/data/infosphere/qualitystage/) was used for record matching purposes.
Kaplan–Meier curves were plotted for 5-year cardiovascular mortality risk, by risk stratification for the four risk prediction models.
2.4 Statistical methods
Missing data was reviewed and determined if it was missing at random. Imputation was performed since several studies have indicated that complete case analyses leads to biased results [
]. Continuous variables with ≤2% missing were imputed using mean or median values where applicable. Single imputation using a linear regression with a random error term was done for total cholesterol (7% missing).
2.5 Model performance
Validity of prediction models was assessed based on discrimination and calibration of the models. Discrimination is the ability to categorize those with and without disease based on predictive values. Calibration is the measure of how accurately the predicted risk matched the observed risk. Utility of the models in this study was based on discrimination and calibration except for two models, the FRS and WHO/ISH. Calibration was not assessed for the FRS and WHO/ISH models because these models estimate the 10-year risk of fatal and non-fatal cardiovascular events. In our study, only fatal cardiovascular events were available.
2.5.1 Discrimination
Discrimination was assessed using the area under the receiver operating characteristic (ROC) curve. A value of >0.75 was considered good discrimination. Model comparisons were statistically tested for differences in the area under the ROC (AUC). This method has been proven acceptable for comparing ordinal tests (WHO/ISH) with continuous tests (FRS and SCORE) [
Model calibration was tested for two models; SCORE-high and SCORE-low. As the cardiovascular mortality events were available for five years in this study, and not 10 as in the original SCORE model, calibration was tested for 5-year mortality. Only for model calibration, all relevant regression equations by Conroy et al. [
] were modified to estimate the 5-year risk of fatal cardiovascular disease. E.g., the underlying survival probabilities were calculated separately for each person's current age and for their age in 5 years time [
Calibration was assessed statistically using a chi-square goodness-of-fit test to determine if the observed 5-year cardiovascular mortality rates differed significantly from the expected [
]. Calibration was also determined graphically by plotting the observed and expected 5-year mortality events, grouped according to deciles of predicted probabilities.
For all analyses, p values less than 0.05 were considered statistically significant. Analyses were performed using IBM SPSS Statistics for Windows, Version 20.0 (IBM Corp., Armonk, NY, USA) and Stata Statistical Software: Release 11.0 (College Station, TX: Stata Corporation LP).
3. Results
There were a total of 14,983 participants aged between 40 and 65 years. 120 participants with existing cardiovascular disease were excluded. Of the remaining 14,863 participants, almost 50% were between the ages 40 and 49 (Table 1). 45.3% were men. Cardiovascular risk factors were highly prevalent; more in women than in men. There were a total of 73,277 person-years of follow-up with a mean duration of 4.93 years. The 5-year cardiovascular and all-cause mortality rate was 1% (148 events) and 3.2% (475 events) respectively. There were 98 cardiovascular mortality events in men, with a follow-up duration of 33,088 person-years and 50 events in women, with a follow-up duration of 40,189 person-years. Among those who died, 82.4% had hypertension, 44.6% had hypercholesterolemia, 39.9% had diabetes mellitus, 4.7% had impaired glucose tolerance and 74.3% were overweight or obese. Of the 148 cardiovascular mortality events, 65.5% were due to ischemic heart disease, 22.3% due to cerebrovascular disease and 12.2% due to other cardiovascular causes of death. 73% of cardiovascular causes of death were medically certified; 78% for men and 64% for women.
Table 1Characteristics of study participants.
Variables
Overall
Men
Women
p value
n
14,863
6739
8124
Age (years)
50.4 (7)
50.6 (7.1)
50.1 (7)
<0.001
Age groups (years)
<0.001
40–49
49.7
48.1
51.1
50–59
37.2
38
36.6
60–65
13
13.9
12.3
Male sex
45.3
100
NA
Race
0.046
Malay
55.1
54.6
55.6
Chinese
23.6
24.2
23.1
Asian Indian
8.4
7.9
8.9
Others
12.8
13.2
12.5
Residence
0.04
Urban
58.5
57.6
59.3
Rural
41.5
42.4
40.7
Education (years of schooling)
<0.001
No schooling
12.1
6.4
16.9
Primary (≤6 years)
40.3
39.2
41.1
Secondary (7–12 years)
41.1
45.4
37.5
Tertiary (≥13 years)
6.5
9
4.5
Household income
<0.001
<MYR2000 (USD 634)
63.1
61.3
64.6
MYR2000–3999 (USD 635–1267)
24
24.8
23.3
≥MYR4000 (USD 1268)
12.9
13.9
12.2
Smoking
20
42.1
1.7
<0.001
Physical examination & investigations
Systolic blood pressure (mm Hg)
139 (22.2)
138.6 (20.7)
139.3 (23.4)
0.04
Diastolic blood pressure (mm Hg)
84.2 (13)
83.7 (12.6)
84.6 (13.4)
<0.001
Body mass index (kg/m2)
25.8 (4.8)
25.1 (4.3)
26.4 (5.2)
<0.001
Waist circumference (cm)
85.1 (12.4)
86.7 (11.8)
83.8 (12.6)
<0.001
Total cholesterol (mmol/l)
4.8 (1.1)
4.7 (0.9)
4.9 (1.1)
<0.001
Hypertension
54.9
52.8
56.7
<0.001
Hypercholesterolemia
33.8
30.4
36.6
<0.001
Diabetes
17.1
16.7
17.5
0.03
Overweight (BMI 23.5 < x < 25)
40.4
44.1
37.3
<0.001
Obesity (BMI > =25)
32
25.3
37.5
<0.001
Data are % for categorical variables and mean (sd) for continuous variables.
3.1 Cardiovascular risk stratification and mortality distribution
Fig. 1 depicts the distribution of the cardiovascular risk categories for the FRS, SCORE and WHO/ISH risk-prediction models. In men, all score models except the WHO/ISH model showed similar trends in risk stratification. Despite the high proportions of cardiovascular risk factors prevalent in this study population, the WHO/ISH model classified almost 90% in the low cardiovascular risk category. Among the women, all models showed similar trends. However, as with men, the WHO/ISH model had the highest classification of low risk populations.
Fig. 1Comparison of cardiovascular risk categories for the Framingham Risk Score (FRS), SCORE and WHO/ISH risk prediction models (y-axis reflect percentage of individuals).
The 5-year cardiovascular mortality rates for all models increased as cardiovascular risk increased (Fig. 2). All showed similar trends and were able to distinguish risk categories in men. However, in women, only the FRS model was able to distinguish differences in mortality risk for the three cardiovascular risk categories. With the SCORE and WHO/ISH models, cardiovascular mortality rates among the intermediate and high risk categories overlapped substantially.
Fig. 25-year cardiovascular mortality rate (y-axis, percentage) by the Framingham Risk Score, SCORE and WHO/ISH risk prediction models.
3.2 Comparison of cardiovascular risk-prediction models
The agreement for risk categorization and correlation of scores between the FRS and SCORE models was good for both men and women. There was hardly any misclassification between the extremes of risk categories in these models. There was slightly better correlation between the FRS and the SCORE-high model (Table 2). There was poor correlation between all models with the WHO/ISH model.
Table 2Agreement of cardiovascular risk categorization and correlation of scores for the Framingham Risk Score, SCORE and WHO/ISH risk prediction models.
The FRS, SCORE-high and SCORE-low models showed good discrimination for cardiovascular mortality (Table 3), with slightly better performance in the SCORE models. The WHO/ISH model had poor discrimination for men and women. Model comparisons for area under the curve showed a p value of <0.0001 for men and women (Fig. 3). The WHO/ISH model was statistically significantly different from the FRS, SCORE-high and SCORE-low models.
Table 3Sensitivity, specificity and discriminative ability for the FRS, SCORE and WHO/ISH models for 5-year cardiovascular mortality.
Models
Cut-off
Sensitivity
Specificity
AUC (95% CI)
FRS
Overall
>20%
61.8
76.8
0.768 (0.734, 0.802)
Men
>20%
72.6
63.6
0.751 (0.708, 0.795)
Women
>20%
42.4
87.7
0.758 (0.702, 0.815)
SCORE-high
Overall
>5%
59.4
79.4
0.774 (0.741, 0.807)
Men
>5%
74.5
67.4
0.768 (0.726, 0.809)
Women
>5%
32.2
89.3
0.763 (0.711, 0.815)
SCORE-low
Overall
>5%
38.2
88.5
0.775 (0.742, 0.807)
Men
>5%
46.2
82.8
0.768 (0.726, 0.81)
Women
>5%
23.7
93.1
0.761 (0.709, 0.813)
WHO/ISH
Overall
>30%
13.3
96.9
0.613 (0.564, 0.662)
Men
>30%
16.0
96.4
0.617 (0.556, 0.678)
Women
>30%
8.5
97.3
0.597 (0.516, 0.678)
AUC, area under the receiver operating characteristic curve.
Graphically, the SCORE-high model for males showed good agreement between the predicted and observed mortality events, except for the highest decile of probabilities (Fig. 4). There was poor agreement for women. Statistically, calibration of the SCORE-high model was acceptable for men but poor for women, with a χ2 goodness-of-fit = 14.79, p value of 0.097 and χ2 = 38.07, p value < 0.001 respectively. Graphically, there was less agreement between observed and predicted risk for men in the SCORE-low model as compared to the SCORE-high model, but calibration was acceptable with a χ2 goodness-of-fit = 16.03, p value 0.066. For women, there was poor calibration visually and statistically with a χ2 goodness-of-fit = 43.62, p value < 0.001. Both SCORE models underestimated the risk in women.
Fig. 4Observed and predicted cardiovascular mortality events for SCORE-high and SCORE-low cardiovascular risk prediction models, by deciles of probabilities.
Our study confirmed that the FRS and both SCORE models, but not the WHO/ISH model can stratify cardiovascular risk in the Malaysian population. The SCORE-high model could accurately predict mortality risk in men, suggesting that Malaysia is a high cardiovascular risk country. The findings of this study provide evidence that despite having similar variables in the model, not all cardiovascular risk-prediction models can accurately identify high-risk individuals. This confirms that the performance of prediction models have to be assessed in the population of interest prior to adoption into clinical practice [
Cardiovascular risk-prediction models in limited resource settings have a very important role. The model cut-off point should sufficiently distinguish between the high and low-cardiovascular risk so as to optimize treatment for those who will benefit the most [
World Health Organization (WHO) and International Society of Hypertension (ISH) risk prediction charts: assessment of cardiovascular risk for prevention and control of cardiovascular disease in low and middle-income countries.
] is that it provides charts with and without total cholesterol. The WHO/ISH model used in this study included total cholesterol levels, yet it could not accurately stratify cardiovascular risk. Our study underscored that unless the WHO/ISH model has been validated, it should not be used for cardiovascular risk stratification.
The implications of using the WHO/ISH risk prediction model without prior validation are worrying. In the Malaysian population, the WHO/ISH model incorrectly categorized most people into the low cardiovascular risk group. This is detrimental to the prevention and control of cardiovascular disease since resources would be spent on screening, yet high-risk individuals would be under-identified, leading to higher rates of under-treatment and subsequently more complications. As the burden of cardiovascular morbidity and mortality is disproportionately higher in developing countries, misclassification of high cardiovascular risk at the crucial period for initiating or optimizing treatment strategies is unacceptable. It may derail any preventive effort.
To our knowledge, this is the first study that has evaluated the performance of the WHO/ISH model. However, other studies that have used the WHO/ISH model for cardiovascular risk stratification in the general population have similarly shown that the WHO/ISH risk prediction model identifies most people as having low cardiovascular risk. Dugee et al. showed that the proportion of low cardiovascular risk as categorized by the WHO/ISH model were 97% (95% CI 96.4, 97.7) for Cambodia, 89.6% (95%CI 86.8, 92.2) for Mongolia and 94.4% (95%CI 91, 97.8) for Malaysia [
Excess cardiovascular risk burden in Jamaican women does not influence predicted 10-year CVD risk profiles of Jamaica adults: an analysis of the 2007/08 Jamaica Health and Lifestyle Survey.
], as determined by the WHO/ISH risk-models for their respective regions. All these studies noted that the prevalence of cardiovascular risk factors was high but it did not translate into high cardiovascular risk categorization.
The FRS model showed good discrimination for both sexes at the quantitative level and at the recommended cut-off of 20% for high cardiovascular risk in our study. The ability of D'Agostino's FRS model to accurately stratify risk has been proven in other studies. In Australia, the FRS had an AUC of 0.73 (95%CI 0.69, 0.77) for men and 0.76 (95%CI 0.72, 0.80) for women [
]. However, our study used the BMI version of the 2008 FRS equation (instead of total cholesterol/HDL cholesterol ratio). The only study validating the discriminatory performance of the BMI model showed a c-statistic of 0.747 in the multi-ethnic Women's Health Initiative Observational Cohort [
], which was lower than the 0.758 found in our study.
The SCORE models have been extensively assessed and externally validated. Most studies prove good discrimination, but some require recalibration for optimal performance. Our study findings are similar to that of Jorstad et al. who found that in the UK–Norfolk population both SCORE models had good discrimination, AUC of 0.78 (95%CI 0.75, 0.81), but only one of the models was calibrated, the SCORE-low model χ2 21.60 (p = 0.02) [
]. In Iceland, similar to the UK–Norfolk study, both SCORE models had good discrimination; AUC 0.80 (95%CI 0.78, 0.82) for SCORE-high and AUC 0.80 (95%CI 0.77, 0.82) for the SCORE-low model, with the SCORE-low model being calibrated [
Estimation of 10-year risk of fatal cardiovascular disease and coronary heart disease in Iceland with results comparable with those of the Systematic Coronary Risk Evaluation project.
]. In Australia, the SCORE-high model had an AUC of 0.75 (95%CI 0.68, 0.82) for men and 0.70 (95%CI 0.62, 0.79) for women, and the SCORE-low model had an AUC of 0.75 (95%CI 0.68, 0.83) for men and 0.70 (95%CI 0.62, 0.79) for women [
]. Again the SCORE-low model performed better, χ2 4.4, p = 0.36 for men, χ2 12.92, p = 0.01 for women. The SCORE-low models were also calibrated for the Dutch population [
]. The SCORE-high model performed poorly in Norway despite it being categorized as a high cardiovascular risk country; AUC range was 0.65–0.68 for men and 0.68–0.72 for women. Instead the SCORE-low model performed better, reiterating the need for each country to validate any recommended risk prediction model [
So far, no SCORE model has been validated for an Asian population. Our study provides the first evidence that the currently available SCORE-high model accurately predicts risk in men. Although both the SCORE models showed good discrimination for women when absolute risks were taken into account, the SCORE models were unable to distinguish mortality risks between the intermediate and high cardiovascular risk categories for women. This suggests that the cut-off values for intermediate and high cardiovascular risk need to be re-adjusted for women, to enable clear distinctions in mortality risks for the low, intermediate and high cardiovascular risk categories, like the FRS. The better discriminatory performance of the SCORE models compared to the FRS seen in this study may be due two reasons; 1) the inclusion of total cholesterol levels in the model, or 2) the use of the same end-points for assessment. In the original SCORE cohort, the authors used the ‘hard’ endpoint of fatal cardiovascular disease, which was the same outcome of interest in our study. The poor calibration seen for the SCORE models in women may be due to the small number of cardiovascular mortality events. Women in South East Asian countries have shown low cardiovascular causes of mortality unlike other countries [
] and this may be similar for Malaysia. An alternative explanation is that cardiovascular causes of mortality are under-recognized in Malaysian women, as in other South East Asian countries [
The cardiovascular risk scores assessed in this study are contemporary models used in other countries. The FRS was developed from a single cohort, all of whom were Caucasians [
]. The SCORE models were developed from a variety of cohort studies from different European countries, also in individuals mostly of Caucasian origin [
]. Despite the mainly Caucasian ethnicity in the development of these models, they were able to discriminate cardiovascular risk in the multi-ethnic Malaysian population. This is most likely because the FRS and SCORE models were developed from real population cohorts. The WHO/ISH risk prediction model was not based on actual cohorts [
]. The model was developed based on a hypothetical cohort that was assigned values of cardiovascular risk factors using estimates of risk-factor prevalence of the various regions. Calculation of absolute risk of cardiovascular events was based on incidence rates estimated from other WHO studies. Despite having charts specifically developed for various regions, the WHO/ISH model for the Malaysian region performed poorly. It is possible that a WHO/ISH model for a different region may work better. However, in light of different WHO/ISH regions categorizing most of their population as having low cardiovascular risk [
Excess cardiovascular risk burden in Jamaican women does not influence predicted 10-year CVD risk profiles of Jamaica adults: an analysis of the 2007/08 Jamaica Health and Lifestyle Survey.
], it would be more prudent to recommend the use of the FRS with a BMI model or the SCORE models for cardiovascular risk prediction.
Our study used data from a cohort that is representative of the whole population. However, some methodological issues remain. Although death registrations are compulsory and each citizen can be traced using a unique identification number, there is a low percentage of medically certified deaths in the country, 57% in 2010 [
]. The other 43% are certified by the police. In these instances, cause of death is determined through interviews with family members regarding the individual's medical history and hospitalizations. However, in this study, 73% of the cardiovascular causes of death were medically certified. Thus, it is likely that the findings of this study remain unchanged.
In this study, diabetes status was determined by self-report or a single hyperglycemic reading by glucometer. There may be misclassification of diabetes status based on a single hyperglycemic reading, but it is likely to be non-differential. Hence a possible overestimation of risk in newly diagnosed diabetics may be balanced by an underestimation of risk in undetected diabetics.
The WHO/ISH risk prediction model has five points that were used for model discrimination. This is a less than ideal comparison. However, unlike the FRS and SCORE models, the WHO/ISH model regression equations are unpublished [
] and calculation of absolute risk is not available for clinical use. Thus, the discriminative ability shown in this study represents actual clinical practice.
Calibration of the SCORE models was based on 5-year cardiovascular mortality events. Actual observed numbers of 10-year events will be essential for adequate calibration of the SCORE models. Aside from this, the FRS model, which includes the total cholesterol/HDL cholesterol ratio, is more widely used in cardiovascular risk-prediction. It is unknown how the more commonly used FRS model would have performed in our population but it is likely to be similar. One study assessing the discriminatory performance of the FRS model using BMI showed a c-statistic of 0.747, which was approximately the same as 0.750 of the FRS model using the total cholesterol/HDL cholesterol ratio [
Our study highlighted that it is crucial to assess cardiovascular risk-prediction models prior to clinical use, as not all are created equal. Only the FRS and SCORE models are applicable for use in clinical practice for the identification of patients at high cardiovascular risk in Malaysia. However, their performance differs by gender. For men, the SCORE-high model should be recommended because of its performance, and ease of recalibration when required in the future. For women, the FRS should be recommended until the SCORE models are recalibrated and new high cardiovascular risk cut-offs can be identified.
Conflicts of interest
The authors report no relationships that could be construed as a conflict of interest.
Acknowledgments
We would like to thank the Director of Health Malaysia for permission to publish this paper.
Appendix A. Characteristics of the FRS, SCORE and WHO/ISH risk prediction models
Tabled
1
Characteristics
Framingham Risk Score (FRS)
SCORE (Systematic COronary Risk Evaluation)
World Health Organization/International Society of Hypertension (WHO/ISH)
Data source
Cohort studies: Framingham Heart Study and Framingham Offspring Study.
Pooled cohort studies
Hypothetical cohort s for different regions
Population
General population in Framingham, Massachusetts, United States of America
12 cohort studies (population and occupational cohorts) from 11 European countries
Not applicable
Age range
30–75
40–65
40–79
Variables
Age, gender, body-mass-index OR total cholesterol & HDL cholesterol, SBP, smoking status, diabetes, hypertensive treatment.
Age, gender, total cholesterol, SBP, smoking status.
Age, gender, with or without total cholesterol, SBP, smoking status, diabetes.
World Health Organization (WHO) and International Society of Hypertension (ISH) risk prediction charts: assessment of cardiovascular risk for prevention and control of cardiovascular disease in low and middle-income countries.
Excess cardiovascular risk burden in Jamaican women does not influence predicted 10-year CVD risk profiles of Jamaica adults: an analysis of the 2007/08 Jamaica Health and Lifestyle Survey.
Estimation of 10-year risk of fatal cardiovascular disease and coronary heart disease in Iceland with results comparable with those of the Systematic Coronary Risk Evaluation project.
☆Funding: This work was supported by a grant from the Ministry of Health, Malaysia (NMRR 10-731-6916) and is part of the MyCARDIO project (STeMM Programme) supported by the University of Malaya/Ministry of Higher Education (UM/MOHE) High Impact Research Grant (Grant number E000010-20001).
01258-3&id=gr1.jpg){kind=link}
01258-3&id=gr2.jpg){kind=link}
01258-3&id=gr3.jpg){kind=link}
01258-3&id=gr4.jpg){kind=link}