Keywords
INTRODUCTION
Cardiovascular disease (CVD), especially ischemic heart disease (IHD), is one of the most frequent causes of death in Spain.1 The high level of fatality is similar to that in other industrialized countries.2,3 However, the incidence of CVD in general and of IHD in particular is among the lowest in the world.4
The Framingham algorithms have proved extremely useful tools for screening patients and in primary prevention in countries with a high incidence of coronary and cardiovascular disease.5 However, they overestimate the risk of coronary heart disease to a greater or lesser extent in several countries.6-10 The Third Joint Task Force of European and Other Societies on Cardiovascular Disease Prevention in Clinical Practice recommends using the Systematic COronary Risk Evaluation (SCORE) algorithm, and suggests cardiovascular prevention guidelines are adapted to the national population characteristics of the country of use.11
One part of the suggested adaptation includes function of risk. A validated methodology to adapt the Framingham algorithm to the reality of each country has been developed and this means we can reliably predict coronary events with precision.12 The REGICOR (REgistre GIroní del COR) group used this method to adapt the algorithm to population-based data on the prevalence of risk factors and rate of coronary events observed in Spain.10,13,14 The VERIFICA study (Validation of the adapted Framingham individual coronary incident risk algorithm) demonstrated the reliability and precision of the REGICOR adaptation in Spain.15
The range of cardiovascular risk algorithms available means we need to compare their operational performance (sensibility, specificity, and predictive values) and the percentages of the population classified as high-risk in order to select the most suitable algorithm for clinical practice in Spain.
We compare the performance of the original Framingham algorithm16 and the REGICOR and SCORE17 adaptations13 for countries with low CVD risk, in terms of their ability to classify patients; ie, sensitivity, specificity, and predictive values. We also evaluate the distinctive characteristics of patients considered high-risk by the different algorithms and the percentage of the population classified as high-risk.
METHODS
Design
This cohort study enrolled 5732 participants aged 35-74 (Figure 1): 4427 were retrospectively selected at random in 67 health centers in the autonomous Spanish regions of Andalusia, Aragón, Catalonia, the Basque country, Extremadura, Galicia, Balearic Islands, Madrid, and Navarre, from patients for whom baseline data were available (1996-1998); 1305 came from a selected, representative, population-based cohort (1995) with prospective follow-up.18
Figure. Flowchart on participant enrolment in the combined cohorts. HDL-C indicates high-density lipoprotein cholesterol; CVD, cardiovascular disease; BP, blood pressure.
All patients were initially free of IHD, stroke, or peripheral arterial disease and contributed reliable data from January 1995 thru December 1998 on blood pressure (BP), total cholesterol, high-density lipoprotein cholesterol (HDL-C), tobacco use, and diabetes.
We only included patients if we had guaranteed access to information about their condition at the end of the follow-up period (5 years) and were able to conduct further follow-up.
Sample Size
The number of coronary events recorded at 5 years in the VERIFICA study (n=180) enables us to estimate the sensitivity of the tests with 95% confidence interval (CI) (7)%, assuming sensitivity was 40%. Specificity can be estimated with CI 95% lower than (2)%, assuming this was around 80%, given that 5552 patients experienced no coronary events.15
Measurements
In the population-based cohort, all cardiovascular risk factors were measured according to standard procedures.18 We considered patients to have diabetes when it was previously diagnosed, they were receiving treatment for diabetes, or had glycemia >125 mg/dL on initial examination. We considered patients hypertensive when it was previously diagnosed, they were receiving antihypertensive treatment, or they presented systolic blood pressure (SBP) ≥140 mm Hg, or diastolic blood pressure (DBP) ≥90 mm Hg on initial examination. Patients consuming at least 1 cigarette/day during the previous year were considered as smokers. At 5 years, we conducted a telephone follow-up using a structured questionnaire to determine the appearance of events of interest, and patients underwent an electrocardiogram. Moreover, we reviewed primary care and in-hospital clinical records of participants hospitalized during the follow-up.
All primary care cohort participants contributed reliable data on BP, total cholesterol, HDL-C, tobacco use, diabetes, and drug regimens administered in the 5 years prior to inclusion. To establish the follow-up start date and, therefore, the date of risk factor baseline values, we used the date when total cholesterol was measured. If data on BP or lipids included more than 1 measurement, we calculated the mean as follows: we noted the mean of the first 2 consecutive measurements of total cholesterol and HDL-C in the earliest 3-month period or the earliest isolated measurement from January 1996 thru December 1998. If total cholesterol had been measured alone and total cholesterol and HDL-C later, we recorded the mean of the 2 total cholesterol measurements and the HDL-C value at the second measurement. We recorded BP values measured closest to the date of the cholesterol means used as follows: the mean of the 2 or 3 consecutive measures in the 6 months closest to the date of the first measurement of cholesterol, described above, or the closest value if we had no access to consecutive values over 6 months. We assumed all patients following drug regimens had received the corresponding dietary advice.15 The family physician was put in contact with all patients during data collection to determine the precise nature of their situation and determine whether they had experienced CVD that had did not appear in clinical records.
For the Framingham and REGICOR algorithms, we evaluated fatal and non-fatal coronary events (including initial angina pectoris); and for the SCORE algorithm, fatal cardiovascular events (coronary events, strokes, heart failure, and peripheral arterial disease). Diagnosis of events of interest was by standard criteria.15
Quality Control of Information
We randomly sampled 15% of patients selected for 3 independent monitors (previously trained physicians) to verify data gathered from the corresponding clinical records. Concordance was very good.15
Follow-Up
We present data from the follow-up conducted during 2003, censored at 5 years.
Events of Interest
We considered as events of interest the appearance of angina pectoris of any type with electrocardiogram (ECG) changes during pain, exercise test with or without isotopes, or positive coronary angiography, non-fatal, symptomatic, acute myocardial infarction, non-fatal stroke, death from CVD, including congestive heart failure, and other causes of death.
Ethical Issues
The study was approved by the ethics committees of the Fundación Jordi Gol i Gurina and Barcelona's Instituto Municipal de Asistencia Sanitaria.
Statistical Analysis
We calculated 5-year risk for each patient using 3 equations: the original Framingham16 algorithm and REGICOR13 and SCORE17 algorithms adapted for low-risk countries. We used Kaplan-Meier survival tables to calculate rate of events censored at 5 years.
The algorithm design enabled us to estimate risk in the population aged 35-74 with the original Framingham and adapted REGICOR algorithms, and in the population aged 35-65 with SCORE.
For the original Framingham algorithm, we used a 10-year risk limit of 20% to determine high risk. With REGICOR, we used 10%, 15%, and 20% limits; with SCORE, we used the recommended 5% limit. For patients <60 years who did not reach the 5% limit, we also calculated SCORE extrapolated to age 60, as recommended in European guidelines.11 We calculated 5-year rates of incidence of IHD and CVD. We present 95% CI.
We conducted a descriptive study of the cohort and data are expressed as mean (SD) for quantitative variables and as percentages for categorical variables. In comparisons between 2 groups, for presence of events or for classification as high-risk, we used Student t test and χ2, respectively.
A P value less than .05 was considered significant. Analysis was with S-Plus 2000 (Insightful Corporation, Seattle, WA, U.S.A.) and SAS 8.2 (SAS Institute, Cary, NC, U.S.A.).
RESULTS
Of 5732 patients enrolled, 3285 were women (57.3%). Table 1 describes characteristics of the sample by gender.
Concordance between data collected and data recorded by monitors was high: demographic data and risk factors, κ=0.75; events recorded during follow-up, κ=0.84. Intra-class correlation coefficient for continuous variables (age, total cholesterol, HDL-C, SBP, and DBP) was ≥0.90 in all cases.
During the 5-year follow-up, 180 patients presented IHD episodes (3.1%; 95% CI, 2.7%-9.7%): 112 men and 68 women; and 43 patients (0.75%; 95% CI, 0.53%-0.97%) died from CVD, 25 (1.0%) men, and 18 (0.5%) women. Patients aged 65-74 experienced 46.6% of all cardiovascular events recorded, and 69.8% of fatal events. Moreover, we observed, 24 (0.4%; 95% CI, 0.2-0.6) noncoronary cardiovascular events and 107 (1.9%; 95% CI, 1.5-2.2) deaths from noncardiovascular causes.
Table 2 shows the differences between participants according to appearance of IHD episodes. Among patients presenting IHD, we found a greater percentage of men and a higher prevalence of the risk factors studied, except mean DBP and total cholesterol, with differences between groups. Despite finding no significant differences between groups in mean total cholesterol, we did find differences in HDL-C values, which were lower in patients with IHD. Table 3 shows differences between patients who died from CVD and the rest. Variables with significant differences were the same as those found for coronary heart disease, except the percentage of smokers and of patients receiving treatment for hypercholesterolemia, which are similar in both groups.
Table 4 describes sensitivity, specificity, and positive predictive value for the 3 algorithms, and the percentage of population classified as high-risk by each of them, with different risk limits for REGICOR. Sensitivity and specificity for predicting coronary, and cardiovascular events were similar for all the algorithms. REGICOR, with 10% and 15% limits, had the highest positive predictive value for both types of event.
Table 5 compares distinctive characteristics of patients classified as high-risk by the 3 algorithms. REGICOR identified greater percentages of women and of patients with HDL-C <45 mg/dL. Extrapolated SCORE and Framingham classified a lower percentage of patients with diabetes in the high-risk group.
Table 6 compares patients aged 35-64 classified with disquieting high-risk by REGICOR and SCORE. Patients classified as high-risk by REGICOR but not by SCORE, presented higher total cholesterol and lower HDL-C than those classified as high-risk by SCORE and not REGICOR.
DISCUSSION
Our study shows that more than half of the coronary events occur in patients who do not classify as high-risk according to these algorithms. The sensitivity of all the algorithms and limits studied is low (<60%) when detecting patients with IHD in the 5-year follow-up. Depending on the algorithm and limits used, percentages of patients, and characteristics of possible candidates for lipid-lowering drug treatment vary considerably. The percentage of the population aged 35-74 selected was 0.8%-22.4%, depending on the algorithm. A balance between sensitivity and specificity in the screening tests is required to select candidates for pharmacologic intervention while avoiding unnecessary treatment. Treatments available produce a relative reduction in risk limited to approximately 30%.14 In this context, the REGICOR algorithm, with a 10-year risk limit of 10%, can be used in the oldest age groups and provides a good balance between sensitivity, specificity, and positive predictive value. Moreover, it classifies as high-risk a reasonable proportion of the population. As clinical practice guidelines indicate,19 in the population it is applicable to, the extrapolated SCORE algorithm classifies the same percentage of patients as high-risk as the original Framingham algorithm does. Both algorithms overestimate the number of events observed in the Spanish population in a similar way.15,20
It is fundamental to adapt risk algorithms in countries like Spain where incidence of coronary heart disease is low.11,15,17 In these countries, most of the population is at a very low level of risk, although this is never absent. Consequently, most cardiovascular events will occur in the low-risk population.
By comparison with patients who presented no coronary event, those who did, initially presented more cardiovascular risk factors, including a lower concentration of HDL-C, but not higher total cholesterol values. This coincides with data reported elsewhere.21,22 Consequently, it seems important that the algorithm used to calculate risk in Spain should consider the concentration of HDL-C.
Whether patients with diabetes and without IHD present coronary risk similar to that of patients with previous IHD19,23 remains controversial. Coronary risk in patients with diabetes but without coronary heart disease is approximately half that of patients without diabetes who present previous IHD.24,25 Unarguably, diabetes is a cardiovascular risk factor that algorithms must take into account. The Framingham original and REGICOR algorithms include diabetes, whereas SCORE does not differentiate these patients.
In patients with diabetes, REGICOR, when compared with UKPDS (United Kingdom Prospective Diabetes Study) and Framingham original, is the only algorithm that does not present significant differences between patients observed and expectations.26 Risk associated with diabetes should probably include time of evolution as well as the presence of complications like albuminuria.
The different equations do not agree on all patients classified as at-risk. The principal difference in the factors susceptible to modification in the population rests in the greater number of patients with abnormal cholesterol (high total cholesterol and/or low HDL-C) among those only classified as at-risk by REGICOR versus those only classified as at-risk by SCORE.
Characteristics and Limitations of the Study
Although it is not representative of the population of each of the autonomous Spanish regions, the sample contains participants from the most populous areas of Spain and the sample size was large enough to obtain conclusive results.
Follow-up of the retrospective part of our cohort could imply some bias in information that we attempted to minimize by selecting in health centers with reliable clinical record systems. Our inclusion criteria required that data on patient health status be available via personal or telephone contact and that verification of supposed cardiovascular events and cause of death, if necessary, should be communicated by patients or their families if it was not included in clinical records.
The retrospective cohort consisted of a patient population presenting a greater prevalence of cardiovascular risk factors than that found in the general population. This prevents us from calculating population-based incidence of the disease. Our study design was not intended to obtain a representative sample of the population, as validating a risk algorithm requires a representative sample of each risk category.
One inevitable limitation of contemporary cohort studies is that we cannot avoid the intervention (pharmacologic or not) during the follow-up that may modify risk factors in the participants. The population studied that experienced a coronary event more frequently received hypertensive and hypercholesterolemia treatment at the start of the follow-up in 1995. In contrast, we did not detect differences in prior hypercholesterolemia treatment among patients with and without fatal cardiovascular events. However, adjustment for treatment in the models did not affect results.15
Clinical Implications
Primary prevention of CVD has been controversial.27-33 Choice of the most adequate algorithm for each country should be based on scientific tests and adapted to preventative needs and policies.11 In Spain, the REGICOR algorithm has been adapted to the characteristics of the population, validated for that population15 and can be applied in those aged 65-74, who experience almost half of all cardiovascular events, and more than two-thirds of fatal cardiovascular events. REGICOR enables us separately to analyze risk in patients with diabetes and provides a better predictive capacity to evaluate patients with abnormal cholesterol.13,14
To understand the usefulness of cardiovascular risk algorithms we have to distinguish between diagnostic tests of disease and screening tests. Cardiovascular risk is not a disease to be treated but a way to rationalize the selection of patients to formulate the best possible primary prevention interventions. An algorithm enables us to estimate risk on a population-based scale that implies a high degree of uncertainty when projected on an individual scale; hence, the poor sensitivity of the algorithms. A diagnostic test with 50% sensitivity would not be acceptable to determine whether a patient had myocardial infarction on arriving with chest pain at the Emergency room. For a screening test intended to provide an instrument that helps structure primary cardiovascular prevention, the importance of this characteristic is only relative. Prevention thru advising on lifestyle is universal and practically independent of degree of cardiovascular risk; and the result of estimating risk almost exclusively affects the decision on drug treatment for dyslipidemia. However, no intervention study has selected patients on the basis of coronary or cardiovascular risk. To achieve 100% sensitivity to detect those patients who will experience a cardiovascular event at 10 years we should treat the whole population, and not even then would we manage to avoid all events as preventative efficacy is limited. Table 4 presents data indicating that a 3% improvement in sensitivity in the REGICOR equation would lead to an approximately 1% deterioration in specificity in patients aged 35-74. With the incidence of events observed this implies that each unit of sensitivity is equivalent to 1.8 individuals who will experience a coronary event, and each 1% of specificity is equivalent to some 55 patients who will not. A simple calculation indicates that for each individual aged 35-74 with a future coronary or cardiovascular event in general identified by a risk algorithm, the reduction in specificity implies that approximately 10 patients, who will not experience an event, will receive preventative treatment. In the population aged 35-64, the ratio would be approximately 1:5. This data should enable us to determine how far we can prudently go in our efforts at prevention. Sensitivity and specificity only depend on the risk limit of an algorithm after which, for example, it is decided to administer drugs.
CONCLUSIONS
Our results reveal that the adapted REGICOR algorithm is the option applicable up to age 74 that gives the best balance in its capacity to classify risk of cardiovascular events in terms of sensitivity, specificity and positive predictive value of the risk algorithms available. Its application enables us to classify as high-risk, patients with a profile making them the best candidates for lipid-lowering drug treatment.
See editorial on pages 690-2
ABBREVIATIONS
IHD: ischemic heart disease
CVD: cardiovascular disease
HDL-C: high-density lipoprotein cholesterol
REGICOR: Girona heart register (REgistre GIroní del COR)
SCORE: Systematic COronary Risk Evaluation
VERIFICA: validation of the adapted Framingham individual coronary incident risk algorithm (Validez de la Ecuación del Riesgo Individual de Framingham de Incidentes Coronarios Adaptada)
The VERIFICA study scientific committee (a full list of researchers is available at www.regicor.org/verifica_inv): Jaume Marrugat, Carles Brotons, Carmen Cabezas, Joan Josep Cabré, Arantxa Catalán, Ramon Ciurana, Eva Comín, Ferran Cordón, Joan Gené-Badia, Rafael Masiá, Ramón Morera, Clara Pujol, Rafel Ramos, Joan Sala, Pascual Solanas, and María-Concepció Violan.
This study received unconditional financial support from Grupo Esteve. It was also part-financed by the Fondo de Investigación Sanitaria research network: Redes Temáticas RECAVA (FIS C03/01), RCESP (FIS C03/09), REDIAPP-Catalunya (FISG03/170), and HERACLES (FIS G03/045), and by the Spanish Ministry of Health and Consumer Affairs, the Instituto de Salud Carlos III, (Red HERACLES RD06/0009).
Correspondence: Dr. J. Marrugat.
Unitat de Lípids i Epidemiologia Cardiovascular. Institut Municipal d'Investigació Mèdica.
Dr. Aiguader, 80. 08003 Barcelona. España.
E-mail: jmarrugat@imim.es
Received August 10, 2006.
Accepted for publication April 3, 2007.