Background
Cardiovascular disease (CVD) remains the leading cause of morbidity and mortality in developed countries, despite consistent improvement in outcomes [
1]. Modifiable risk factors are the major causes of atherosclerotic CVD (ASCVD), accounting for 90% of cardiovascular risk [
2,
3]. A key strategy in the primary prevention of ASCVD is the use of risk prediction algorithms, as the prophylactic interventions should target the people who should benefit from them most [
1,
4]. However, it is debatable which is the best algorithm for ASCVD risk estimation.
In 2003, the European Society of Cardiology (ESC) jointly with the European Atherosclerosis Society (EAS) published the SCORE (Systematic COronary Risk Estimation): a 10-year risk estimation system for ACSVD death [
5]. The SCORE Project pooled data from 12 European prospective cohort studies from eleven countries, including Southern European countries such as Spain and Italy, but not Portugal [
5]. External validation has been performed for the SCORE model in several Western countries and in the Asian population [
6‐
10]. However, calibration of the risk charts according to cardiovascular mortality levels of each country has been proposed for a more precise prediction of risk estimates [
5]. This is of utmost importance in countries not represented in the derivation cohorts, as Portugal.
In parallel, the American College of Cardiology (ACC) and American Heart Association (AHA) developed the Pooled Cohort Equations (PCE) that estimate a composite endpoint of 10-year ASCVD risk, instead of the SCORE-estimated 10-year fatal ASCVD risk [
11]. However, these equations were all derived from North American cohorts [
11], which limit their applicability to other populations. Further, the PCE have been controversial because of reports that they substantially overestimate risk in both American and European populations [
12‐
16].
Besides the different prediction models available to estimate the ASCVD risk, the decision thresholds recommended for drug therapy are different for, respectively, fatal ASCVD (SCORE) and any ASCVD (PCE) [
17]. Such conflicting recommendations may create confusion among physicians, potentially reflecting uncertainty about the external validity of different algorithms under different settings [
18], namely for low-risk countries, like those in Southern Europe [
19,
20]. In Portugal, a country with one of the lowest ASCVD event rates in Europe [
20], neither risk systems have been validated and doubts concerning their clinical applicability and predictive accuracy in this particular population are raised.
In this study, we aimed to determine the 10- and 15-year incidence of ASCVD events in a primary prevention cohort of a Southern European country. In addition, we aimed to compare the calibration and discrimination of both SCORE and PCE systems in this cohort at the predefined 10-year landmark and at a longer, 15-year follow-up, in order to estimate their external validity for clinical practice.
Discussion
In our cohort of Portuguese patients, the evaluation of PCE and SCORE risk prediction algorithms revealed that both performed at an acceptable level. The SCORE system performed better than the PCE system in discriminating their respective endpoints. Although calibration was similar, both risk systems markedly overestimated the risk.
While both risk systems address the same practical question, which individuals benefit from interventive measures, its equations differ significantly. The SCORE includes as predictors age, sex, smoking, systolic blood pressure, and TC; it is traditionally applicable in individuals from 40 to 65 years, and the predicted outcome is 10-year fatal ASCVD, disregarding non-fatal events [
5]. Besides the predictors included in SCORE, PCE also incorporates ethnicity, HDL-C, antihypertensive use, and diabetes. Furthermore, PCE has a larger age range (40–79 years) and predicts the 10-year risk of fatal and non-fatal ASCVD [
11].
Existing studies regarding the SCORE risk prediction system generally reported good discrimination, consistent with our findings [
5,
8,
9,
24]. Considering the SCORE system, our study reported a favorable C-statistic of 0.83. The discrimination performance of the SCORE risk to predict 10-year ASCVD mortality was evaluated in the SCORE project and the C-statistic values ranged from 0.71 to 0.84 [
5]. In the global population, we observed a systematic risk overestimation that has also been previously reported in low-risk populations, as in Spain [
8,
9], and high-risk populations [
24].
The external validation studies that evaluate the PCE system have generated controversy around the predictive accuracy of these equations in contemporary cohorts [
25]. The PCE models discriminated risk reasonably (C-statistic 0.62) in our cohort, but performing worse in comparison to other studies [
11,
15,
16]. The original PCE C-statistic ranged from 0.71 to 0.82 [
11] and 0.74 to 0.79 in contemporary cohorts [
16]. Similar to the SCORE, we observed a risk overestimation in our population, with an exception in predicting ASCVD risk in low-risk strata (< 5%). Previous literature established that the PCE generally overestimate risk among modern European and American populations [
13,
15,
16,
23,
26]. Although there are no reports of underestimation of ASCVD risk by the PCE, it has been widely recognized that a large number of people with ASCVD risk prediction < 7.5% will paradoxically experience ASCVD events [
27], reinforcing the value of additional risk markers to reclassify those patients [
4,
27].
Accurate calibration of the risk prediction models is crucial for the success of preventive measures. Although both models overestimated the risk, when observed risks were taken into account, the accuracy of risk prediction in women was less impressive. Actually, the risk was underestimated by the PCE risk system in the low-risk (< 5%) and intermediate-risk (7.5–20%) strata. Underestimation of ASCVD risk in the female gender has only been described once previously in a Korean cohort [
28].
Concerning the SCORE system, the interaction between sex and events should be analyzed as an exploratory and hypothesis-generating finding, due to the very low number of 10-year fatal ASCVD events (
n = 7), especially in men (
n = 2). This low ASCVDmortality rate is consistent with what has been found in a Spanish cohort of 608 non-diabetic patients on primary prevention with a similar moderate 10-year risk of fatal ASCVD (2.1%), where only 9 ASCVD deaths were reported (1.5%) [
8]. Nevertheless, women’s risk was underestimated in high-risk women (5–10%). An underestimation of the risk by the SCORE system in women has already been reported in Australian women below 50 years [
6], and in a Malaysian population [
7]. Conversely, in another Southern European cohort from Italy, the 10-year fatal ASCVD risk was properly predicted by SCORE [
29]. However, the low-risk (< 1%) women from this cohort had a 20-year rate of any ASCVD of 3.7, and 40% of ASCVD events occurred in this class, which signals potential problems with the calibration of this equation in women [
29].
The causes of the unexpected findings regarding the underprediction of fatal or any ASCVD risk in women and the very low number of ASCVD deaths in men are not clear. Although women from our cohort had higher levels of LDL-C, they were also treated more commonly with statins. In the past, numerous studies have shown excess mortality in women after myocardial infarction [
30]. In addition, women suffering an acute coronary syndrome are treated less intensively than men [
31], and, even on primary prevention, it has been observed a lower use of lipid-lowering drugs in women [
32]. Independently of the cause of the excess risk of ASCVD events in women, our results suggest that at least in our population, the risk stratification systems might need to be better calibrated among women, in order to reduce the proportion of women miscategorized and to avoid CVD events by implementing appropriate preventive measures [
5].
Lastly, the SCORE overestimated the 15-year risk in the overall population, probably reflecting the decline of CVD mortality in the last decades [
33]. Conversely, regarding PCE, we observed an important increase in any ASCVD events at 15-year follow-up. The observed 15-year any ASCVD incidence was 1.8 times the 10-year incidence among men and 1.4 times among women. This acceleration of ASCVD risk is even more relevant in the low- and borderline-risk strata in men and the low- and intermediate-risk strata in women, as the observed 15-year risk surpasses the 10-year predicted risk. A divergence between short-time and long-term risk has already been described in two Italian studies [
29,
34]. Among the relatively large group with low and intermediate short-term risk, differences in risk factor burden seem to translate into marked differences in the incidence of ASCVD events in the remaining lifespan [
35]. The added value of long-term risk estimation becomes even more relevant in women and younger men, more prone to be classified as being at low short-term risk [
36].
It is essential to evaluate the applicability of risk models to each population, as risk scores may perform worse in a different setting from the one they were originally derived [
37]. It is well-known that the incidence of CVD events is declining, and Portugal is no exception. Despite the high prevalence of hypertension and other chronic diseases, Portugal has one of the best indicators for cardiovascular mortality [
20], namely age- and risk-standardized mortality rates concerning coronary artery disease and cerebrovascular disease [
19]. This problem is well-recognized and the literature recommends that scores should be recalibrated if incidence rates differ substantially in the new settings [
5,
13], as we have previously discussed for women. Additionally, newer statistical methods could considerably improve the accuracy of ASCVD risk estimates [
16]. Another, more straightforward option to improve its predictive ability would be the incorporation of additional risk variables such as coronary artery calcium score and high-sensitivity C-reactive protein [
4,
27] or consider the lifetime risk for CVD in addition to short-time (10-year) risk [
4].
We acknowledge several limitations of our study. We have a limited sample size, representative of the Center region of Portugal, one of the regions with the lowest rate of CVD events in our country [
20]. This may explain partially the overestimation of the risk by both systems, and extrapolation of our results to other populations should be done cautiously. In addition, only half of the patients had a complete 15-year follow-up, as many patients were included in this cohort after 2003. Also, we cannot overlook that two-thirds of our cohort was under statin therapy; however, the majority of the patients were under low- and moderate- intensity statins as they were recruited before the era of high-intensity statins recommendation and are representative of a real-world population seen in everyday practice. Interestingly, in a sensitivity analysis of the Copenhagen General Population Study, including statin users at baseline (7% of the population) did not explain the observed mismatch between the predicted and observed event rates [
21]. Finally, we were also not able to adjust for prescription of other preventive medication during follow-up. This might also have contributed to some of the observed mismatch. A strength of our study is that the results originate from the largest contemporary Portuguese cohort evaluated so far, with no patient lost to follow-up.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.