ORIGINAL ARTICLE

The contribution of educational class in improvingaccuracy of cardiovascular risk prediction acrossEuropean regions: The MORGAM Project CohortComponentMarco M Ferrario,1 Giovanni Veronesi,1 Lloyd E Chambless,2 Hugh Tunstall-Pedoe,3

Kari Kuulasmaa,4 Veikko Salomaa,4 Anders Borglykke,5 Nigel Hart,6

Stefan Söderberg,7 Giancarlo Cesana,8 for the MORGAM Project

▸ Additional material ispublished online only. To viewplease visit the journal online(http://dx.doi.org/10.1136/heartjnl-2013-304664).

For numbered affiliations seeend of article.

Correspondence toProfessor Marco M Ferrario,Centro Ricerche EPIMED,Dipartimento di MedicinaClinica e Sperimentale,Università degli studidell’Insubria, Via O. Rossi,Varese 9-21100, Italy;marco.ferrario@uninsubria.it

Received 9 September 2013Revised 10 March 2014Accepted 7 April 2014

To cite: Ferrario MM,Veronesi G, Chambless LE,et al. Heart Published OnlineFirst: [please include DayMonth Year] doi:10.1136/heartjnl-2013-304664

ABSTRACTObjective To assess whether educational class, anindex of socioeconomic position, improves the accuracyof the SCORE cardiovascular disease (CVD) riskprediction equation.Methods In a pooled analysis of 68 455 40–64-year-old men and women, free from coronary heart disease atbaseline, from 47 prospective population-based cohortsfrom Nordic countries (Finland, Denmark, Sweden), theUK (Northern Ireland, Scotland), Central Europe (France,Germany, Italy) and Eastern Europe (Lithuania, Poland)and Russia, we assessed improvements in discriminationand in risk classification (net reclassification improvement(NRI)) when education was added to models includingthe SCORE risk equation.Results The lowest educational class was associatedwith higher CVD mortality in men (pooled age-adjustedHR=1.64, 95% CI 1.42 to 1.90) and women (HR=1.31,1.02 to 1.68). In men, the HRs ranged from 1.3 (CentralEurope) to 2.1 (Eastern Europe and Russia). Afteradjustment for the SCORE risk, the association remainedstatistically significant overall, in the UK and EasternEurope and Russia. Education significantly improveddiscrimination in all European regions and classificationin Nordic countries (clinical NRI=5.3%) and in EasternEurope and Russia (NRI=24.7%). In women, afterSCORE risk adjustment, the association was notstatistically significant, but the reduced number of deathsplays a major role, and the addition of education led toimprovements in discrimination and classification in theNordic countries only.Conclusions We recommend the inclusion ofeducation in SCORE CVD risk equation in men,particularly in Nordic and East European countries, toimprove social equity in primary prevention. Weakerevidence for women warrants the need for furtherinvestigations.

INTRODUCTIONThe existence of a social gradient in cardiovasculardisease (CVD) mortality in recent years has beenconfirmed by comparative studies conducted inEurope in the 1990s and early 2000s,1–3 althoughthe magnitude of the gradient varied widely acrosspopulations.

Inequalities in risk factors account for only 25%–

50% of the CVD mortality excess risk in the lowersocioeconomic strata.4 In European populationswhere CVD mortality rates were declining in thelast decades, the social gradient in traditional riskfactors did not decrease much.5–9 This is not sur-prising since health policies adopted in many coun-tries, such as smoking bans or salt restrictions, aregenerally not targeted at social groups. In addition,whether low socioeconomic group should be specif-ically targeted in primary prevention is not fullyestablished, as no CVD risk prediction equation10–13 in primary prevention takes socioeconomic statusinto account. Indeed, in the UK, the Framinghamscore was found to underestimate the risk amongsubjects in the lower socioeconomic groups.14 15 Toovercome these inadequacies in risk prediction, itwas suggested either to consider socioeconomicstatus-specific risk score thresholds for treatmentallocation16 or to add a measure of socioeconomicstatus as an independent predictor in risk equa-tions.17–19 The latter topic is still controversial,20 ascomparative studies on risk prediction accuracy bysocial class in different geographical areas arelimited.The MORGAM Project (http://www.thl.fi/

morgam),21 pooling a number of population-basedcohorts from several European countries investi-gated with standardised methods for baseline datacollection and follow-up, represents a uniqueopportunity to explore educational class differencesin CVD mortality risk across Europe.The aim of the present paper is to assess whether

educational class as an index of socioeconomic pos-ition in addition to the SCORE equation improvesthe prediction of CVD mortality risk across severalEuropean regions.

METHODSIncluded cohortsThe present analysis is based on 47 cohorts from18 populations adhering to the MORGAMProject21 recruited from 10 European countries.The baseline examinations were done between themid-1980s and mid-1990s. Participation rates were70% or more in almost all the populations, andonly one cohort had a rate below 60% (Belfast).

Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664 1

Epidemiology

Full details for the included cohorts have been published,22 andmajor characteristics are summarised in table 1.

Educational statusThe number of years of schooling was investigated as a singlequestionnaire item (‘How many years have you spent at schoolor in full-time study?’) included in standardised questionnaires.The three-class study variable (high, intermediate and low edu-cation) is derived from cohort-specific, gender-specific and age-specific tertiles of the distribution of years of schooling. Thisclassification is less affected than educational attainment bymodifications in the schooling system, which may occur at dif-ferent times in different populations.23

Baseline CVD risk factors assessmentThe baseline assessments of CVD risk factors either followedthe standardised WHO- MONICA Project protocol or adoptedMONICA-like procedures; detailed descriptions are available.24

In brief, blood pressure (BP) was measured on sitting subjectsafter 5 min rest using either a standard or random-zero sphyg-momanometer, with the exception of the PRIME study, inwhich an automated device was used. Only systolic BP (mean oftwo subsequent measurements) was the study variable in thepresent analysis. Total cholesterol and HDL-cholesterol (inmmol/L) were determined on sera, with the exceptions of thePRIME study and the Poland cohorts (on plasma). Participatinglaboratories underwent well-defined external and/or internalquality control.24

Weight and height were measured on subjects wearing lightgarments and without shoes. Body mass index (BMI) was calcu-lated as weight (kg) divided by the square of height (m2).Information on cigarette smoking habits, previous history ofcoronary events and diabetes mellitus, including specific treat-ments, were collected through questionnaires or interviews. Forthe present analysis, we combined former and never-smokers asnon-smokers.

Individual SCORE riskWe adopted the gender-specific SCORE risk prediction equationto estimate the individual’s absolute risk of CVD death in10 years based on age, systolic BP, total cholesterol and smokingstatus. Subjects less than 40 or more than 64 years old, or witha positive history of a previous coronary event, or with eithertotal cholesterol or systolic BP out of the SCORE computationalrange (3–8 mmol/L and 100–180 mm Hg, respectively) wereexcluded.11The baseline survival term was determined by thespecific risk categories (low or high) of each population, asrecommended.25

Follow-up proceduresThe cohorts were followed up for death using different proce-dures, including tracking of the vital status for each subject andcollecting death certificates or through record linkages tonational or regional death registers. The underlying causes ofdeath were based on the final official codes or death certificates,except in the PRIME study, where they were determined by thestudy Medical Committee. Full details on follow-up proceduresand quality are publicly available.24

Statistical analysisA total number of 79 790 men and women 40–64 years old atbaseline and with no history of previous coronary heart disease(CHD) were originally selected. Of these, 4210 (5.3%) wereexcluded due to missing data, of them 958 were missing yearsof schooling and a further 3252 were missing at least one of therisk factors. Finally, 68 455 had total cholesterol and systolic BPvalues within the SCORE range and were included in thepresent analysis. The cohorts were aggregated according to geo-graphical proximity into four regions: Nordic countries(Finland, Sweden and Denmark), the UK (Scotland andNorthern Ireland), Central European cohorts (France, SouthernGermany and Northern Italy, all sharing low CVD rates25) andEastern Europe and Russia (Poland, Lithuania and Russia). The

Table 1 Characteristics of the surveyed populations

Region Country (population) Cohorts # Setting Obs period

Analysiscohort size* Education

M W

M W YS Δ YS YS Δ YS

Nordic countries Sweden (Northern Sweden) 3 U/R 1986–1994 1203 1222 9.6 6.6 9.8 6.3Finland(East Finland) 4 R 1982–1997 4126 4471 8.9 6.3 9.4 6.2Finland (West Finland) 4 U 1982–1997 2751 2912 10.3 6.7 10.3 6.2Denmark (Glostrup)† 3 U 1982–1992 2244 2086 10.7 5.8 9.9 5.1

The UK Northern Ireland (Belfast) 1 U 1991–1994 2390 – 11.2 5.1 – –

Scotland‡ 6 U/R 1984–1995 5214 4969 11.1 4.9 11.0 4.3Central Europe France§ 3 U 1991–1993 6888 – 11.4 7.4 – –

Germany (Augsburg) 3 U/R 1984–1995 3192 3217 12.4 5.4 10.7 4.5Italy (Brianza¶) 4 U 1986–1994 1823 1899 8.2 7.8 6.9 5.7Italy (Friuli) 4 R 1986–1994 1645 1739 8.8 7.9 7.2 6.4

Eastern Europe and Russia Lithuania (Kaunas) 2 U 1986–1993 1008 956 11.2 9.2 11.2 8.2Poland (Warsaw) 3 U 1983–1993 1759 1733 11.9 8.6 11.1 7.0Poland (Tarnobrzeg Voivodship) 3 R 1983–1993 1696 1865 8.3 5.8 7.6 4.5Russia (Novosibirsk) 4** U 1983–1995 3276 2171 10.0 8.6 10.1 6.9

Total – 47 – – 39 215 29 240 10.5 6.6 9.8 5.5

*40–64 years old, with no previous CVD event at baseline.†Birth cohorts of subjects 40, 50 and 60 years old at baseline.‡MONICA Glasgow, MONICA Edinburgh and Scottish Heart Health Study).§MONICA Lille, Strasbourg and Toulouse.¶MONICA-Brianza and PAMELA Study.**The first survey (1983–85) is on men only.Setting: U, urban; R, rural; CVD, cardiovascular disease; YS: mean number of years of schooling. Δ YS: mean difference in YS between the highest and the lowest educational classes.

2 Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664

Epidemiology

age-adjusted mean (prevalence) of risk factors at age 50 in eacheducational class was estimated fitting gender-specific linear orlogistic models, with dummy variables for each of the two lowereducational classes.

The SCORE risk was calibrated according to the methoddescribed by Chambless et al,26 based on the observed survivalin each of our four regions. HRs and 95% CIs of CVD risk, bysex, education and geographic region, were calculated from Coxsurvival model analysis, considering the highest educationalclass as reference, and adjusting for (a) age and MORGAMcentre alone and (b) calibrated SCORE risk and MORGAMcentre. Model calibration was assessed through the Grønnesby–Bogan goodness-of-fit test for survival models27; findings arepresented as online supplementary table S1. A pooled analysisincluding all cohorts has been also carried out, and HR are pre-sented by sex and education, adjusting for MORGAM centres.We formally tested the null hypothesis of no difference in CVDdeath hazard among educational categories (2 degrees offreedom). We also assessed the improvement given by educationwhen added to a basic model with the calibrated SCORE riskalone, in terms of discrimination (difference in the area underthe receiver–operator characteristic curve, Δ-AUC) and reclassifi-cation (continuous net reclassification improvement (NRI)).Both measures were estimated accounting for censoring.28 Inaddition, clinical NRI was estimated among those at intermedi-ate 10-year risk, that is, 1–5% according to the European guide-lines based on the SCORE risk.20 The improvement metricswere corrected for overfitting through tenfold cross-validation(mean of 49 runs29), and corresponding 95% CIs were derivedfrom bootstrap (n=1000, bias corrected percentile method;

standard SAS macro available at http://support.sas.com/kb/24/addl/fusion_24982_1_jackboot.sas.txt).

The analysis was conducted using SAS software V.9.2.

RESULTSThe analysis included 39 215 men and 29 240 women, aged40–64 years. The average number of years of schooling rangedfrom 8.2 to 12.4 and from 6.9 to 11.2 in men and women,respectively (table 1). The mean difference in years of schoolingbetween the highest and the lowest educational groups was 6.6in men and 5.6 in women.

Education and CVD risk factorsTable 2 reports, for men and women respectively, theage-adjusted mean (prevalence) of major CVD risk factors byeducational class and geographical region. Figures 1 and 2 plotthe age-adjusted mean difference or OR between the two lowerand the highest educational class, in each population, ranked byregions for men and women, respectively.

Less educated men were more likely to smoke and to havehigher systolic BP values (table 2) in all the regions, and beingoverweight in Nordic countries, the UK and Central Europe.Contrasting results were observed for blood lipids: less educatedmen had a more favourable lipid profile (either lower total orhigher HDL-cholesterol) in the UK, Central Europe and EasternEurope and Russia, but not in the Nordic countries. A higherprevalence of diabetes among less educated men was observedin Central Europe only.

As reported in figure 1, the association between educationand risk factors was quite homogeneous across populations

Table 2 Age-adjusted mean value† or prevalence† of major CVD risk factors by educational class and geographic region

Region Education N AgeTotalcholesterol HDL-cholesterol BMI

SystolicBP

Cigarettesmokers‡ Diabetes‡

Men, 40–64 years old, with no previous CVD event at baseline screeningNordic countries Low 3725 51.1 6.02 1.27 27.2 137.7 36.5 4.8

Intermediate 2798 51.2 5.96 1.26 27.0 137.4 32.9 4.0High 3801 50.4* 5.90* 1.27 26.5* 135.5* 25.6* 4.5

The UK Low 3111 51.6 6.01 1.30 26.2 134.1 41.1 1.8Intermediate 2065 51.9 6.07 1.30 26.4 133.2 34.0 1.7High 2428 51.4*** 6.07*** 1.31 25.8* 131.9* 23.4* 1.8

Central Europe Low 4513 53.3 5.80 1.30 27.0 134.1 28.2 5.7Intermediate 3957 53.1 5.84 1.29 26.8 133.7 26.3 4.4High 5078 52.8* 5.83 1.29 26.3* 132.4* 21.7* 4.1**

Eastern Europe and Russia Low 2529 51.0 5.50 1.38 26.1 135.9 59.2 3.6Intermediate 2482 51.2 5.61 1.36 26.3 135.3 55.8 3.1High 2728 51.1 5.68* 1.31* 26.7* 134.0** 41.8* 3.3

Women, 40–64 years old, with no previous CVD event at baseline screeningNordic countries Low 3473 51.2 5.93 1.51 27.1 135.5 25.7 4.7

Intermediate 3121 50.7 5.91 1.54 26.4 134.3 21.3 4.0High 4097 50.9*** 5.80* 1.57* 25.7* 133.2* 17.0* 3.9

The UK Low 2654 49.6 6.28 1.62 26.1 132.4 44.4 1.2Intermediate 933 50.8 6.28 1.67 25.4 131.7 33.3 1.3High 1382 50.0* 6.21 1.75* 24.9* 129.7* 23.7* 0.9

Central Europe Low 2635 51.0 5.91 1.60 26.9 133.9 16.3 3.7Intermediate 1791 51.8 5.92 1.60 26.4 133.9 16.6 3.8High 2429 51.5** 5.94 1.65* 25.2* 131.4* 19.7** 2.1**

Eastern Europe and Russia Low 2385 51.0 5.65 1.49 29.5 136.5 11.2 3.8Intermediate 1725 51.0 5.70 1.48 28.9 135.7 11.0 4.7High 2615 51.6** 5.77* 1.52** 28.0* 134.6** 13.9** 3.7

p Values for testing the overall association of each risk factor with education, by geographic area (2 df test): * <.0001; ** <0.01; *** <0.05. Unit of measurement: total- andHDL-cholesterol: mmol/L; body mass index (BMI): kg/m2; systolic blood pressure (BP): mm Hg.†At age of 50.‡Prevalence.CVD, cardiovascular disease.

Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664 3

Epidemiology

within each region, with the exception of the Polish rural popu-lation (Tarnobrzeg-Voivodship) showing a more favourable riskprofile in the low educational class.

Less educated women in Nordic countries and the UK hadhigher mean values of total cholesterol, BMI, systolic BP;lower mean value of HDL-cholesterol; and were more likelyto smoke (table 2). In contrast, Central European more edu-cated women were more likely to smoke but had a lowerprevalence of diabetes, lower mean values of BMI and systolicBP, and higher mean of HDL-cholesterol. Similar findingswere detected in Eastern Europe and Russia, but with highervalues of BMI in the lowest educational class and increasingmean levels of total cholesterol for increasing education.Considering the specific populations (figure 2), a homogeneitywithin geographical areas in social inequalities was observedfor most risk factors, with the exception of the Polish ruralpopulation.

Education and 10-year risk of CVD mortalityAmong men (table 3), the 10-year predicted risk of CVD deathwas associated with education in all regions, the risk being

higher for less educated subjects. With reference to the higheducation group, the pooled age-adjusted HRs for the low andthe intermediate education classes were 1.64 (95% CI 1.42 to1.90) and 1.48 (1.27 to 1.73), respectively (χ2 test, p<0.0001).The social gradient was more evident in Eastern Europe andRussia with HRs of 2.05 (1.57 to 2.68) and 1.71 (1.29 to 2.25),for the lower and the intermediate educational classes, respect-ively, followed by the UK (HRs 1.73, 1.23 to 2.43; and 1.49,1.02 to 2.17) and the Nordic countries (HRs 1.42, 1.09 to1.85; and 1.35, 1.03 to 1.78). A weaker and non-statistically sig-nificant association was found in Central Europe. Adjustmentsfor the SCORE risk reduced the HRs, but the associationsremained statistically significant in Eastern Europe and Russiaand in the UK. In the Nordic countries, in the UK and inCentral Europe, the original SCORE equation overestimated therecalibrated SCORE risk in all the educational classes (sixth andseventh columns in Table 3), with higher values in more edu-cated men. In Eastern Europe and Russia, however, the originalSCORE equation underestimated the risk in all educationclasses, with higher differences in the low and the intermediateclasses.

Figure 1 Age-adjusted mean difference^ or odds ratio§ in cardiovascular disease (CVD) risk factors, by population, between each of the two lowerand the highest educational groups. Men. Black squares are for low education, and grey dots for intermediate education. ^Mean difference for thefollowing RFs: total- and HDL-cholesterol, body mass index, systolic blood pressure. §Odds ratios reported for cigarette smoking and diabetesprevalence (reference=high education). *Russia-NOV and Italy-FRI: last two surveys only (no data in the remaining cohorts).

4 Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664

Epidemiology

In women, the pooled age-adjusted HRs for the low and theintermediate education classes were 1.31 (95% CI 1.02 to 1.68)and 1.03 (0.78 to 1.39), respectively (χ2 test, p=0.09), and asignificant association between 10-year CVD mortality and edu-cation was found in the Nordic countries region only, with aHR for low vs. high education of 1.67 (1.07 to 2.61). Both theassociations were no longer significant when adjusting for theSCORE risk (table 3).

Although the associations between education level and CVDmortality varied between European geographical areas, theinteraction terms education level and geographic area were notstatistically significant in men (p=0.33) and women (p=0.79).

Improvement in discrimination and risk reclassificationTable 3 reports the improvement in discrimination (Δ-AUC) andclassification (NRI and clinical NRI) when education was addedto the Cox models with the calibrated SCORE risk. A significantalthough modest improvement in discrimination was observedin men (Δ-AUC=0.003), consistently in all regions. However,we observed a significant 20.7% improvement in classificationin the pooled analysis, as well as in Eastern Europe and Russia

(24.7%); the NRI was 15% in Central Europe and Nordic coun-tries (not significant). The positive NRI was mainly attributableto improvements in classification of events (ranging from 25%to 54% in the different areas) in the lower educational classes,only partially attenuated by a smaller worsening in non-events(ranging from −28% to −14%). Increments of clinical NRI of8.9%, 5.3% and 10.3% were observed in the entire sample, inthe Nordic countries and in Eastern Europe and Russia,respectively.

In women, the addition of education to the SCORE riskmodel showed small but significant improvements in discrimin-ation and classification in the Nordic countries (NRI=26.0%),mostly driven by an improvement in classification of non-events.In the entire sample, we observed an improvement in classifica-tion (NRI=9.4%), but not statistically significant.

DISCUSSIONThe lowest educational class was associated with higher CVDmortality in men (pooled age-adjusted HR 1.64, 95% CI 1.42to 1.90) and women (HR 1.31, 95% CI 1.02 to 1.68).

Figure 2 Age-adjusted mean difference^ or odds ratio§ in cardiovascular disease (CVD) risk factors, by population, between each of the twolower and the highest educational groups. Women. Black squares are for low education, and grey dots for intermediate education. ^Mean differencefor the following RFs: total- and HDL-cholesterol, body mass index, systolic blood pressure. §Odds ratios reported for cigarette smoking and diabetesprevalence (reference=high education). *Russia-NOV and Italy-FRI: last two surveys only (no data in the remaining cohorts).

Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664 5

Epidemiology

Table 3 Association between education and CVD death, and improvements in discrimination and reclassification due to the addition of education in predictive models with the calibrated SCOREequation, in different European regions

Men, 40–64 years old, with no previous CVD event at baseline screening

Region Education D†/N

Association (HR 95% CI)‡ 10-year risk (%)§

Δ-AUC¶ (95% CI)

Reclassification††

Age-Adj SCORE-Adj Original SCORE Calibrated SCORE NRI % (95% CI) Clin-NRI, % (95%CI)

Nordic countries Low 139/3725 1.42 (1.09 to 1.85) 1.21 (0.93 to 1.57) 4.8 3.9 0.001 (0 to 0.004) 14.5 (−2.0 to 27.8) 5.3 (0.1 to 14.8)Intermediate 106/2798 1.35 (1.03 to 1.78) 1.23 (0.93 to 1.62) 4.8 3.8High 99/3801 REF*** REF 4.1 3.2

The UK Low 116/3111 1.73 (1.23 to 2.43) 1.54 (1.10 to 2.17) 4.7 3.3 0.004 (0.0004 to 0.013) 10.1 (−27.2 to 20.4) 6.3 (−0.9 to 15.9)Intermediate 61/2065 1.49 (1.02 to 2.17) 1.32 (0.90 to 1.92) 4.2 3.2High 48/2428 REF** REF*** 3.8 2.8

Central Europe Low 78/4513 1.29 (0.91 to 1.81) 1.25 (0.89 to 1.76) 2.5 1.8 0.002 (0 to 0.009) 15.0 (−4.3 to 33.0) 4.9 (−1.7 to 13.9)Intermediate 65/3957 1.35 (0.95 to 1.92) 1.32 (0.93 to 1.88) 2.4 1.7High 60/5078 REF REF 2.2 1.6

Eastern Europe and Russia Low 153/2529 2.05 (1.57 to 2.68) 1.81 (1.38 to 2.36) 4.6 6.4 0.008 (0.003 to 0.02) 24.7 (7.7 to 35.8) 10.3 (0.6 to 21.3)Intermediate 129/2482 1.71 (1.29 to 2.25) 1.53 (1.16 to 2.02) 4.7 6.4High 83/2728 REF* REF* 4.2 5.8

All regions Low 486/13 878 1.64 (1.42 to 1.90) 1.46 (1.26 to 1.69) 3.9 3.6 0.003 (0.001 to 0.005) 20.7 (10.9 to 28.0) 8.9 (5.0 to 14.4)Intermediate 361/11 302 1.48 (1.27 to1.73) 1.35 (1.16 to 1.58) 3.8 3.5High 290/14 035 REF* REF* 3.4 3.1

Women, 40–64 years old, with no previous CVD event at baseline screening

Region Education D†/N

Association (HR 95% CI)‡ 10-year risk (%)§

Δ-AUC‡‡ (95% CI)

Reclassification††

Age-Adj SCORE-Adj Original SCORE Calibrated SCORE NRI, % (95% CI) Clin-NRI, % (95% CI)

Nordic Countries Low 48/3473 1.67 (1.07 to 2.61) 1.52 (0.97 to 2.38) 1.5 1.1 0.005 (.0004 to 0.014) 26.0 (2.1 to 53.7) 5.8 (−10.7 to 18.7)Intermediate 25/3121 1.00 (0.60 to 1.68) 0.93 (0.56 to 1.56) 1.3 1.0High 34/4097 REF*** REF 1.3 1.0

The UK Low 32/2654 1.14 (0.62 to 2.11) 0.99 (0.54 to 1.85) 1.3 1.2 0.0005 (−0.0001 to 0.001) −12.7 (−55.2 to 9.2) −1.7 (−23.9 to 4.6)Intermediate 11/933 1.02 (0.47 to 2.21) 0.87 (0.40 to 1.89) 1.3 1.4High 15/1382 REF REF 1.1 1.1

Central Europe Low 18/2635 1.38 (0.67 to 2.82) 1.40 (0.68 to 2.87) 0.8 0.7 0.004 (0.0001 to 0.016) −1.7 (−52.5 to 30.5) 3.3 (−40.3 to 16.3)Intermediate 11/1791 1.05 (0.47 to 2.38) 1.11 (0.49 to 2.49) 1.0 0.8High 13/2429 REF REF 0.9 0.8

Eastern Europe and Russia Low 33/2385 0.98 (0.62 to 1.51) 0.98 (0.61 to 1.57) 1.2 1.8 0.001 (−0.001 to 0.005) −13.7 (−47.8 to 1.8) 0.2 (−17.5 to 4.4)Intermediate 24/1725 1.06 (0.63 to 1.77) 1.13 (0.67 to 1.88) 1.2 1.8High 38/2615 REF REF 1.3 1.9

All regions Low 131/11 147 1.31 (1.02 to 1.68) 1.21 (0.93 to1.58) 1.2 1.2 0.0005 (−0.0001 to 0.002) 9.4 (−8.1 to 22.7) 2.3 (−2.5 to 7.4)Intermediate 71/7570 1.03 (0.76 to 1.39) 0.99 (0.73 to 1.35) 1.2 1.2High 100/10 523 REF REF 1.2 1.2

p Value testing the overall association of CVD mortality with education (2 df test): *<0.0001; **<0.01; *** <0.05.CVD mortality: ICD-9 codes 401–414 and 426–443, excluding 426.7, 429.0, 430.0, 432.1, 437.3, 437.4, 437.; and 798.1–798.2 (SCORE endpoint).†Observed number of events during the first 10 years of follow-up.‡Cox model, adjusting for age (left column) and the calibrated SCORE (right).§10-year risk of CVD mortality, as the average original SCORE risk and calibrated SCORE risk.¶Change in AUC from the basic model with calibrated SCORE only. AUC for the basic model: 0.7888 (Nordic countries); 0.7820 (UK); 0.7282 [Central Europe]; 0.7700 (Eastern Europe and Russia), 0.7840 (all regions). 95% confidence intervals fromn=1000 bootstrapped samples.††Reclassification: NRI (continuous), clinical NRI (NRI among those with 10-year predicted risk of CVD mortality between 1% and 5%—intermediate risk29)‡‡Change in AUC from the basic model with calibrated SCORE only. AUC for the basic model: 0.7491 [Nordic countries]; 0.7250 [UK]; 0.7399 [Central Europe]; 0.7275 [Eastern Europe and Russia]; 0.7669 [all regions]. 95% confidence intervals fromn=1000 bootstrapped samples.AUC, area under the receiver–operator characteristic curve; CVD, cardiovascular disease; N, number of subjects.

6Ferrario

MM,etal.Heart2014;0:1

–9.doi:10.1136/heartjnl-2013-304664

Epidemiology

The association was quite consistent across regions, withHRs ranging from 1.3 (95% CI 0.9 to 1.8) in Central Europeto 2.1 (1.6 to 2.7) in Eastern Europe and Russia, in men andfrom 0.98 (95% CI 0.62 to 1.51) in Eastern Europe andRussia to 1.67 (95% CI 1.07 to 2.61) in the Nordic countries.In men, the association was attenuated but still significantwhen adjusting for the SCORE risk, while in women it was nolonger significant, suggesting that a variable amount of theexcess risk in low educational classes is explained by the differ-ences in the major risk factors. Furthermore, when educationwas added, we observed an improvement of the calibratedSCORE risk in discrimination and classification in men in allEuropean regions combined and in Scandinavian and EasternEuropean Countries and Russia, as well as in women in theNordic countries.

We found in women a smaller socioeconomic gradient thanwhat has been published previously in the Scottish cohorts,14

but methodological differences may explain the discrepancy:first, the study endpoint—CVD deaths vs fatal and non-fatalCHD, which may be more sensitive to social deprivation, andsecond the differences in the indicator of socioeconomic pos-ition—a three class years of education rather than the fiveclasses of the Scottish Index of Multiple Deprivation. Theapparent absence of association in the Eastern European womencan be attributed to a better CVD risk factor profile in thelower class, particularly the lower prevalence of smoker. Thisadvantage has been lost in more recent years.30

The observed improvements in discrimination and classifica-tion when education was added to a model with the calibratedSCORE risk suggest the inclusion of education in the risk equa-tion to improve risk prediction reached through an increase ofsocial equity in CVD prevention. Ramsey et al15 reported amodest improvement in classification when occupation wasadded to the Framingham equation in a sample of UK men; anddiscrimination was not affected by education in theAtherosclerosis Risk in Communities population.18 However, inboth studies, the inclusion of socioeconomic indexes reducedthe discrepancy between expected and observed events amongsocial classes. In our populations, the analysis of the expected(based on calibrated SCORE) to observed CVD deaths ratio sug-gested that the SCORE overestimates the risk in the highest edu-cational class in all regions, with a maximum in Eastern Europeand Russia (50%), and underestimates the risk in the lowestclass in all regions, up to 23% again in Eastern Europe andRussia (data not shown).

Moreover, the improvement in classification in men wasdriven by an improvement in risk stratification of events in thelower educational classes (ranging from 25% to 54% in the UKto East Europe and Russia), only in part attenuated by a worsen-ing in stratification among non-events. In women, an incrementin reclassification (NRI=9.4%) was detected in the entiresample (not statistically significant) and in the Nordic countries(NRI=26.0%), mostly driven by an improvement in classifica-tion of non-events. Therefore, the inclusion of a socioeconomicposition indicator in the risk equation is recommended toprevent CVD cases in the lower social classes and to reduceinappropriate treatments in the higher social classes. TheASSIGN score is currently adopted nationally as the default car-diovascular risk score in Scotland for that reason as it ensuressocial equity.17 When the predictive abilities of the Framinghamand ASSIGN risk equations are compared (table 2),17 the

Framingham score does quite well in allocating treatment tosocially deprived people because relatively more of them are inthe top 20% of risk than the socially privileged. However, itdoes not do enough to compensate for the risk disadvantage inthe deprived, which the ASSIGN score does do.

Study strengths include the high degree of standardisation ofdata collection on CVD death and CVD risk factors, as well asthe relatively high number of available CVD risk factors, unusualfor comparative studies with a similar number of included popu-lations. Educational attainment, an alternative measure of educa-tion, may not be completely comparable in internationalstudies.31 In our baseline examinations, the data distribution andthe average number of years of schooling by educational attain-ment varied greatly across populations, especially among women(data not shown). We derived a classification23 in which the dis-tribution of subjects was similar across most of the populationsand was not correlated with the magnitude of the difference inmean values (or proportion) between the lowest and the highesteducational classes for any of the considered risk factors (datanot shown). Finally, we recalibrated the SCORE equation basedon the cohort-specific survival. Although the SCORE equationdifferentiates high-risk countries from low-risk countries, ourresults indicate that further recalibration is needed when SCOREis applied in European populations.

We recognise some limitations. The study endpoint is officialCVD death as in SCORE. In some Eastern European countries,the quality of the routine diagnosis of CHD deaths is question-able due to a significant number of alcohol-attributable deathsmisclassified as coronary deaths.32 This diagnostic drift bias mayaffect our estimates of the overall and the educational class-specific CVD risk estimates in Eastern Europe. First, CHDdeaths represent about 50% of all CVD deaths in our EasternEuropean populations. Then, we found consistent educationalclass gradients in CHD risk either using the official CHD deathcodes or the MONICA validated CHD deaths in men of allMORGAM Eastern European populations (data not shown).About 5% of subjects was excluded from the analysis due tomissing data on covariates. Excluded subjects did not differ interms of average years of schooling (10.0±3.7 vs 10.1±3.6);moreover, the age-adjusted association of education with CVDmortality was consistent with HRs reported in table 3.

Although data have been collected according to a standardisedprotocol, some differences across populations have to beacknowledged, particularly for blood lipids,24 but it is unlikelythat these differences have played a role in social differenceswithin populations. Self-reported data such as cigarette smokingare potentially exposed to differential reporting across socialclasses.33

In our study, education was classified based on population-specific, gender-specific and birth cohort-specific tertiles of yearsof schooling, taking into account variations across Europe inschooling in different gender groups and time periods. Thisstudy then warrants further research to understand how educa-tion should be incorporated in risk scores.

In conclusion, based on our findings, the inclusion of educa-tion improves the SCORE CVD prediction equation inEuropean male populations, particularly in Nordic countriesand East Europe. The improvement in risk stratificationincreases social equity and reduces inappropriate treatment inprimary prevention. Further investigations are needed inwomen.

Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664 7

Epidemiology

Key messages

What is already known on this subject?▸ There are evidences of a poor accuracy of the SCORE

predictive equation in UK men of the lower social classes.This may worsen social inequalities in CVD mortality.

What does this study adds?▸ The addition of education to the SCORE risk in CVD

prediction models improved discrimination and riskclassification in European men and in Scandinavian women.The improvement was mostly driven by a correctre-classification of events in less educated men; and by are-classification of non-events among women.

How might this impact on clinical practice?▸ We recommend the inclusion of education in SCORE CVD

risk equation in men, in particular in Nordic and EastEuropean countries to improve social equity in primaryprevention.

Author affiliations1Dipartimento di Medicina Clinica e Sperimentale, Centro Ricerche EPIMED—Epidemiologia e Medicina Preventiva, Università degli Studi dell’Insubria, Varese,Italy2Department of Biostatistics, University of North Carolina at Chapel Hill, Chapel Hill,North Carolina, USA3Cardiovascular Epidemiology Unit, Institute of Cardiovascular Research, University ofDundee, Dundee, UK4Department of Chronic Disease Prevention, THL-National Institute for Health andWelfare, Helsinki, Finland5Research Centre for Prevention and Health, Glostrup University Hospital, Glostrup,Denmark6Centre for Public Health, School of Medicine, Dentistry and Biomedical Sciences,Queen’s University Belfast, Belfast, UK7Department of Public Health and Clinical Medicine, Cardiology and Heart Centre,Umeå University, Umeå, Sweden8Centro Studi e Ricerche in Sanità Pubblica (CESP), Università degli Studi di Milano-Bicocca, Monza, Italy

Collaborators Sites and key personnel of contributing MORGAM Centres.Denmark: Glostrup, Capital Region of Denmark, Research Centre for Prevention andHealth, Glostrup University Hospital: T. Jørgensen (principal investigator), C. Agger,A. Borglykke; Finland: FINRISK, National Institute for Health and Welfare (THL),Helsinki: V. Salomaa (principal investigator), A. Juolevi, E. Vartiainen, P. Jousilahti;MORGAM Data Centre, National Institute for Health and Welfare (THL), Helsinki:K. Kuulasmaa (head), Z. Cepaitis, A. Haukijärvi, B.Joseph, S. Kulathinal, M. Niemelä,O. Saarela; France: National Coordinating Centre, National Institute of Health andMedical Research (U258), Paris: P. Ducimetière (national coordinator), A. Bingham;PRIME/Strasbourg, Department of Epidemiology and Public Health, EA. 3430,University of Strasbourg, Faculty of Medicine, Strasbourg: D. Arveiler (principalinvestigator), B.Haas, A. Wagner; PRIME/Toulouse, Department of Epidemiology,Faculty of Medicine, Toulouse-Purpan, Toulouse: J. Ferrières (Principal Investigator),J-B. Ruidavets and V. Bongard; PRIME/Lille, Department of Epidemiology and PublicHealth, Pasteur Institute of Lille: P. Amouyel (principal investigator), M. Montaye,J. Dallongeville; Germany: Augsburg, Helmholtz Zentrum München, GermanResearch Centre for Environmental Health, Institute of Epidemiology, Neuherberg,Germany: A. Döring (principal investigator), E. Wichmann; Italy: NationalCoordinating Centre MORGAM, Centro Ricerche EPIMED -Dipartimento di MedicinaClinica e Sperimentale, Università degli Studi dell’Insubria, Varese: M. Ferrario(national coordinator). Brianza, Dipartimento di Medicina Clinca e Sperimentale,Università degli Studi dell’Insubria, Varese: M. Ferrario (principle investigator),G. Veronesi; Research Centre on Public Health, Università Bicocca, Monza:G. Cesana (principal investigator), C. Fornari; Studio PAMELA, Dipartimento dimedicina Clinica e Preventiva, Università Bicocca, Monza: R. Sega, G. Mancia,R. Facchetti; Friuli: Cardiovascular Prevention Centre, ASS4 Medio Friuli andDirezione Regionale della Salute, Udine. D. Vanuzzo, L. Pilotto, F. Picco, F. Mattiussi,S. Gigante. Lithuania: Kaunas, Kaunas University of Medicine, Institute ofCardiology, Kaunas: A. Tamosiunas (principal investigator), S. Domarkiene (formerprincipal investigator), D. Rastenyte, G. Bernotiene, R. Radisauskas; M. Baceviciene,

D. Luksiene, V. Malinauskiene and L. Bernotaite. Poland: Krakow, Unit of ClinicalEpidemiology and Population Studies, School of Public Health, JagiellonianUniversity, Krakow: A. Pająk (principal investigator), E. Kawalec; Warsaw,Department of Cardiovascular Disease Epidemiology and Prevention and HealthPromotion. The Cardinal Stefan Wyszynski Institute of Cardiology. Warsaw. G. Broda(principal investigator), P. Kurjata, S.L. Rywik, M. Polakowska, A. Pytlak; RussianFederation: Novosibirsk, Institute of Internal Medicine, Siberian Branch of theRussianAcademy of Medical Science, Novosibirsk: Yu. Nikitin (principal investigator),S. Malyutina, V. Gafarov, V. Feigin, G. Siminova, M. Voevoda, T. Vinogradova,N. Nasonova, E. Veriovkin; Sweden: Northern Sweden, Umeå University Hospital,Medicine, Umeå: M. Eliasson and S. Söderberg (principal investigators),B. Stegmayr, M.Eriksson and K. Asplund (former principal investigators). The UK:PRIME/Belfast, Queen’s University Belfast, Belfast, Northern Ireland: F Kee (principalinvestigator), A. Evans (former principal investigator) J. Yarnell, E. Gardner; Scotland,University of Dundee, Dundee, Scotland: H. Tunstall-Pedoe (principal investigator), MWoodward; MORGAM Coordinating Centre, Queen’s University Belfast, Belfast,Northern. Ireland: A. Evans (MORGAM coordinator), S. Cashman; MORGAMManagement Group: A. Evans (chair), S. Blankenberg (Hamburg, Germany),M. Ferrario, F. Kee, K. Kuulasmaa, A. Palotie (Hinxton, UK), M. Perola (Helsinki,Finland) A. Peters (Munich, Germany), V. Salomaa, D. Tregouet (Paris, France),H. Tunstall-Pedoe; Previous members: K. Asplund (Umeå, Sweden), F. Cambien/ LTiret (Paris, France), L. Peltonen (Helsinki, Finland), D. Shields (Dublin, Ireland),B. Stegmayr, P.-G. Wiklund (Umeå, Sweden).

Contributors All authors had full access to results of the data analyses, assistedwith interpretation of results and reviewed and approved the final version of themanuscript. MMF and GC are the principal investigators for the MONICA-Brianzastudy cohorts; MMF planned the data analysis and wrote the paper; he is theguarantor. GV and LEC prepared the data quality assessment and were responsible forstatistical analysis. KK is head of the MORGAM Project and is the overall guarantor ofthe MORGAM data; HTP, VS, CG and SS are principal investigators for the ScotlandHeart Study, the FINRISK study and Northern Sweden cohorts, respectively, andactively contributed in commenting on the results and writing the paper.

Funding This work has been sustained by the MORGAM Project’s current funding:European Community FP 7 projects ENGAGE (HEALTH-F4-2007-201413), CHANCES(HEALTH-F3-2010-242244) and BiomarCaRE (278913). This has supported centralcoordination, workshops and part of the activities of the MORGAM Data Centre, atTHL in Helsinki, Finland. MORGAM Participating Centres are funded by regional andnational governments, research councils, charities and other local sources. Thestatistical analysis was supported by Regione Lombardia within the CAMUNI IIIProject (grant n. 010804/2009).

Competing interests None.

Ethical approval Each MORGAM participating centre is responsible for ethicalapproval and patient consent, according to local rules at the time of studyenrolment.

Provenance and peer review Not commissioned; externally peer reviewed.

REFERENCES1 Avendano M, Kunst AE, Huisman M, et al. Socioeconomic status and ischaemic

heart disease mortality in 10 western European populations during the 1990s.Heart 2006;92:461–7.

2 Kunst AE, del Rios M, Groenhof F, et al. Socioeconomic inequalities in strokemortality among middle-aged men: an international overview. European UnionWorking Group on Socioeconomic Inequalities in Health. Stroke 1998;29:2285–91.

3 Mackenbach JP, Stirbu I, Roskam AJR, et al. Socioeconomic Inequalities in Health in22 European Countries. N Engl J Med 2008;358:2468–81.

4 Harald K, Koskinen S, Jousilahti P, et al. Changes in traditional risk factors nolonger explain time trends in cardiovascular mortality and its socioeconomicdifferences. J Epidemiol Community Health 2008;62:251–7.

5 Jansson JH, Boman K, Messner T. Trends in blood pressure, lipids, lipoproteins andglucose metabolism in the Northern Sweden MONICA project 1986–99. Scand JPublic Health Suppl 2003;61:43–50.

6 Osler M, Gerdes LU, Davidsen M, et al. Socioeconomic status and trends in riskfactors for cardiovascular diseases in the Danish MONICA population, 1982–1992.J Epidemiol Community Health 2000;54:108–13.

7 Ferrario M, Sega R, Chatenoud L, et al. Time trends of major coronary risk factorsin a northern Italian population (1986–1994). How remarkable are socioeconomicdifferences in an industrialized low CHD incidence country? Int J Epidemiol2001;30:285–97.

8 Molarius A, Seidell JC, Sans S, et al. Educational level, relative body weight, andchanges in their association over 10 years: an international perspective from theWHO MONICA Project. Am J Public Health 2000;90:1260–8.

9 Giskes K, Kunst AE, Benach J, et al. Trends in smoking behaviour between 1985and 2000 in nine European countries by education. J Epidemiol Community Health2005;59:395–401.

8 Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664

Epidemiology

10 D’Agostino RB, Vasan RS, Pencina MJ, et al. General cardiovascular risk profile foruse in primary care. The Framingham Heart Study. Circulation 2008;117:743–53.

11 Conroy RM, Pyorala K, Fitzgerald AP, et al. Estimation of ten-year risk of fatalcardiovascular disease in Europe: the SCORE project. Eur Heart J 2003;24:987–1003.

12 Chambless LE, Folsom AR, Sharrett AR, et al. Coronary heart disease risk predictionin the Atherosclerosis Risk in Communities (ARIC) study. J Clin Epidemiol2003;56:880–90.

13 Ferrario MM, Chiodini P, Chambless LE,, et al Prediction of coronary events in a lowincidence population. Assessing accuracy of the CUORE Cohort Study predictionequation. Int J Epidemiol 2005;34:413–21.

14 Tunstall-Pedoe H, Woodward M, for the SIGN group on risk estimation. Byneglecting deprivation, cardiovascular risk scoring will exacerbate social gradients indisease. Heart 2006;92:307–10.

15 Ramsay SE, Morris RW, Whincup PH, et al. Prediction of coronary heart disease riskby Framingham and SCORE risk assessments varies by socioeconomic position:results from a study in British men. Eur J Cardiovasc Prev Rehabil 2011;18:186–93.

16 Franks P, Tancredi DJ, Winters P, et al. Including socioeconomic status in coronaryheart disease risk estimation. Ann Fam Med 2010;8:447–53.

17 Woodward M, Brindle P, Tunstall-Pedoe H, for the SIGN group on risk estimation.Adding social deprivation and family history to cardiovascular risk assessment: theASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart2007;93:172–6.

18 Fiscella K, Tancredi D, Franks P. Adding socioeconomic status to Framingham scoringto reduce disparities in coronary risk assessment. Am Heart J 2009;157:988–94.

19 Collins GS, Altman DG. Predicting the 10 year risk of cardiovascular disease in theUnited Kingdom: independent and external validation of an updated version ofQRISK2. BMJ 2012;344:e4181.

20 Perk J, De Backer G, Gohlke H,, et al European Guidelines on cardiovascular diseaseprevention in clinical practice (version 2012): The Fifth Joint Task Force of theEuropean Society of Cardiology and Other Societies on Cardiovascular DiseasePrevention in Clinical Practice (constituted by representatives of nine societies andby invited experts). Eur Heart J 2012;33:1635–701.

21 Evans A, Salomaa V, Kulathinal S, et al. MORGAM (an international pooling ofcardiovascular cohorts). Int J Epidemiol 2005;34:21–7.

22 Kulathinal S, Niemelä M, Kuulasmaa K; contributors from Participating Centres, forthe MORGAM Project. Description of MORGAM Cohorts. MORGAM Projecte-publications [Internet]. 2005; (2). URN:NBN:fi-fe20051214. http://www.thl.fi/publications/morgam/cohorts/index.html

23 Karvanen J, Veronesi G, Kuulasmaa K. Defining thirds of schooling years inpopulation studies. Eur J Epidemiol 2007;22:487–92.

24 Niemelä M, Kulathinal S, Kuulasmaa K. eds; for the MORGAM Project. Descriptionand quality assessment of MORGAM data. MORGAM Project e-publications[Internet]. 2007;(3). URN:NBN:fi-fe20071495. http://www.thl.fi/publications/morgam/qa/contents.htm

25 European Society of Cardiology. HeartScore: cardiovascular disease (CVD) riskassessment and management. http://www.heartscore.org/Pages/welcome.aspx

26 Chambless LE, Heiss G, Shahar E, et al. Prediction of ischemic stroke risk in theAtherosclerosis Risk in Communities Study. Am J Epidemiol 2004;160:259–69.

27 May S, Hosmer DW. A simplified method of calculating an overall Goodness-of-Fittest for the Cox proportional hazards model. Lifetime Data Anal 1998;4:109–20.

28 Chambless L, Cummiskey C, Cui G. Several methods to assess improvement in riskprediction models: extension to survival analysis. Stat Med 2010;30:22–38.

29 Steyerberg EW. Clinical prediction models. New York, USA: Springer Science +Business Media, LLC, 2009.

30 Schaap MM, Kunst AE, Leinsalu M, et al. Female ever-smoking, education,emancipation and economic development in 19 European countries. Soc Sci Med2009;68:1271–8.

31 Galobardes B, Shaw M, Lawlor DA, et al. Indicators of socioeconomic position (part1). J Epidemiol Community Health 2006;60:7–12.

32 Radisauskas R, Prochorskas R, Grabauskas V, et al. Recent heavy alcoholconsumption at death certified as ischaemic heart disease: correcting mortality datafrom Kaunas (Lithuania). Alcohol Alcohol. 2011;46:614–9.

33 Suadicani P, Hein HO, Gyntelberg F. Serum validated tobacco use and socialinequalities in risk of ischaemic heart disease. Int J Epidemiol 1994;23:293–300.

Ferrario MM, et al. Heart 2014;0:1–9. doi:10.1136/heartjnl-2013-304664 9

Epidemiology