Curriculum in Cardiology
Prediction of first coronary events with the
Framingham score: A systematic review
Klaus Eichler, MD, MPH,a Milo A. Puhan, MD, PhD,a Johann Steurer, MD, MME,a and
Lucas M. Bachmann, MD, PhDa,b Zurich and Berne, Switzerland
Background
Uncertainty exists about the performance of the Framingham risk score when applied in different
populations.
Objective We assessed calibration of the Framingham risk score (ie, relationship between predicted and observed
coronary event rates) in US and non-US populations free of cardiovascular disease.
Methods We reviewed studies that evaluated the performance of the Framingham risk score to predict first coronary
events in a validation cohort, as identified by Medline, EMBASE, BIOSIS, and Cochrane library searches (through August
2005). Two reviewers independently assessed 1496 studies for eligibility, extracted data, and performed quality assessment
using predefined forms.
Results We included 25 validation cohorts of different population groups (n = 128 000) in our main analysis.
Calibration varied over a wide range from under- to overprediction of absolute risk by factors of 0.57 to 2.7. Risk prediction
for 7 cohorts (n = 18 658) from the United States, Australia, and New Zealand was well calibrated (corresponding figures:
0.87-1.08; for the 5 biggest cohorts). The estimated population risks for first coronary events were strongly associated
(goodness of fit: R 2 = 0.84) and in good agreement with observed risks (coefficient for predicted risk: h = 0.84; 95% CI
0.41-1.26). In 18 European cohorts (n = 109 499), the corresponding figures indicated close association (R 2 = 0.72) but
substantial overprediction (h = 0.58, 95% CI 0.39-0.77). The risk score was well calibrated on the intercept for both
population clusters.
Conclusion
The Framingham score is well calibrated to predict first coronary events in populations from the United
States, Australia, and New Zealand. Overestimation of absolute risk in European cohorts requires recalibration procedures.
(Am Heart J 2007;153:722231.e8.)
Clinical guidelines for primary prevention of coronary heart disease (CHD) recommend a risk management based on the Framingham score.1,2 The
Framingham score as applied in these guidelines is a
tool to predict the absolute risk of coronary events in
populations free of cardiovascular disease (CVD). It is
available in different application formats (eg, point
scoring systems, risk charts, or Web-based calculators)
From the aHorten Centre for Patient Oriented Research, University hospital of Zurich,
Postfach Nord, Zurich, Switzerland, and bDepartment of Social and Preventive Medicine,
University of Berne, Berne, Switzerland.
The study was supported by the Helmut Horten Foundation. The funding source had no
influence on study design; in the collection, analysis, and interpretation of the data; in the
writing of the manuscript; and in the decision to submit the manuscript for publication.
Submitted October 26, 2006; accepted February 13, 2007.
Reprint requests: Klaus Eichler, MD, MPH, Horten Centre, University Hospital of Zurich,
Postfach Nord, CH-8091 Zurich, Switzerland.
E-mail: [email protected]
0002-8703/$ - see front matter
D 2007, Mosby, Inc. All rights reserved.
doi:10.1016/j.ahj.2007.02.027
and requires only information from patient history and
easily available tests.3
However, there is uncertainty about the performance
of the Framingham score to predict absolute risk for first
coronary events when applied in different populations.
Validation studies showed inconsistent results across
different populations4,5 leading to a debate about the
appropriateness of this tool for risk prediction in
Europe. Finally, it is important to ensure that CHD risk
prediction is as accurate as possible for the group of
patients to which it is applied.6
We aimed to systematically review all cohort studies
that validated calibration (ie, relationship between
predicted and observed event rates) of the Framingham
risk score in US- and non-US populations free of CVD.
Methods
Data sources
We systematically searched for studies of prediction rules for
cardiovascular risk using electronic databases (Medline,
EMBASE, BIOSIS, Cochrane library; from 1966 to August 2005;
American Heart Journal
Volume 153, Number 5
Eichler et al 723
Figure 1
Study flow of the systematic review.
no language restriction). Validation studies of prediction rules
are difficult to retrieve in electronic data bases because such
studies are not indexed in a standardized manner. Thus, we
used a sensitive search strategy not specific for the Framingham model but for cardiovascular prediction rules in general
(see Appendix Table II for search strategy).
We screened reference lists of included papers, contacted
experts, and performed searches for related articles and
citation searches of key references. We hand-searched relevant
journals (Circulation, European Heart Journal, and Medical
Decision Making) for supplements and conference proceedings. Where we did not find sufficient data for our analysis, we
contacted authors. Of 21 contacted authors, 8 provided
additional data. All references were stored in an EndNote 6.0
database (Thomson/ISI ResearchSoft, Berkeley, CA).
Study selection
We undertook the systematic review in accordance with
current recommendations.7 We included studies with a
prospective cohort design and cohorts derived from randomized trials. We defined a validation study as prospective if the
data set of the validation population had been collected
prospectively. Thus, a prospective validation study could be set
up on a historical data set if it had been collected prospectively.
American Heart Journal
May 2007
724 Eichler et al
Table I. Details of 24 validation cohorts of the Framingham risk score for main analysis
Validation cohort4,
location
Author, y
Brindle et al, 20035
Brindle, 200412
Cooper et al, 200513
DPCG, 20029
DPCG, 2002
DPCG, 2002
DPCG, 2002
DPCG, 2002
DPCG, 2002
DPCG, 2002
Empana et al, 200314
Empana et al, 2003
Ferrario et al, 200515
Greenland, 200416
Hense et al, 20034
Hense et al, 2003
Milne et al, 2003
17
Mora et al, 200518
Orford et al, 200219
Ramachandran
et al, 200020
Simons, 200321
Stephens et al, 200422
British regional heart study
cohort; United Kingdom
Caerphilly and
Speedwell cohort;
United Kingdom
NPHS-II; United Kingdom
NHANES I, men; United States
NHANES II, men;
United States
Scottish Collaborative Study,
men; Scotland
Renfrew and Paisley Study,
men; United Kingdom
Norwegian Counties Study,
men; Norway
Iceland Reykjavik Study,
men; Iceland
Yugoslavia Cardiovascular
Disease Study, men;
former Yugoslavia
PRIME Study; Ireland
PRIME Study; France
CUORE cohorts, men; Italy
Sample size (n)
(men, %)
Clinical outcomeyy
(follow-up)
6643 (100%)
NA (40-59), British, GP
Total CHD (10 y)
3213 (100%)
NA (45-63), British,
general population
Hard CHD (10 y)
2732 (100%)
3856 (100%)
56 (50-64), British, GP
NA (35-74), US American,
general population
NA (35-74), US American,
general population
NA (35-72), Scottish, OCC
Hard CHD (10 y)
Death from CHD (8 y)
NA (45-64), British,
general population
NA (35-49), Norwegian,
general population
NA (36-74), Icelandic,
general population
NA (35-64), Yugoslavian,
general population
Death from CHD (8 y)
3354 (100%)
5734 (100%)
6999 (100%)
24204 (100%)
8151 (100%)
6340 (100%)
2399 (100%)
7359 (100%)
5794 (100%)
South Bay Heart Watch cohort,
United States
PROCAM cohort;
Germany (Münster)
MONICA cohort, men;
Germany (Augsburg)
Fletcher Challenge Heart and
Health Study; New Zealand
John Hopkins Sibling Study;
United States
Normative Aging Study;
United States (Boston, MA)
Wickham Study; England
1700 (44.2%)
Dubbo Study; Australia
1800 (41.9%)
University College
London Hospitals cohort;
United Kingdom
Glostrup cohort; Denmark
SHHEC; United Kingdom
Mean age (range) (y),
ethnicity, recruitment
1029 (90%)
5527 (100%)
2861 (100%)
6354 (73%)
872 (51%)
1393 (100%)
798 (63%)
4757 (50.1%)
Thomsen et al, 200223
Tunstall-Pedoe and
13037 (49%)
Woodward, 200524
WOSCOPS Group, 199825 WOSCOPS Group (placebo
1251 (100%)
control group of RCT); Scotland
NA (50-59), Irish, OCC, GP
NA (50-59), French, OCC, GP
NA (35-69), Italian,
general population
67.5 (NA-NA), US American,
general population
46.5 (35-64), German, OCC
49.5 (35-64), German,
general population
NA (35-74), ethnicity: see comment,
OCC, general population
45.8 (NA-60), US American,
family members
58.2 (30-74), Northern American,
general population
NA (30-75), English,
general population
NA (60-79), Australian,
general population
NA (35-74), ethnicity: see comment,
University diabetes clinic
Death from CHD (8 y)
Death from CHD (8 y)
Death from CHD (8 y)
Death from CHD (8 y)
Death from CHD (8 y)
Total CHD (5 y)
Total CHD (5 y)
Total CHD (10 y)
Hard CHD (7 y)
Hard CHD
(variable follow-up y)
Hard CHD
(variable follow-up y)
CVD (71% of events
were total CHD) (5 y)
Total CHD (10 y)
Total CHD (10 y)
Total CHD (20 y)
Hard CHD (10 y)
Total CHD (10 y)
58.9 (49-70), Danish, general population Death from CHD (10 y)
NA (30-74), Scottish, general population Total CHD (10 y)
NA (45-64), Scottish, moderately
hypercholesterinemic men
Hard CHD (4 y)
NA, Not available; OCC, occupational cohort; GP, cohort from general practitioner; ATP III, Adult Treatment Panel III; DPCG, Diverse Populations Collaborative Group; SHHEC,
Scottish Heart Health Extended Cohort; RCT, randomized controlled trial.
4Design: prospective cohort study (if not stated otherwise).
yDeath from CHD (includes fatal myocardial infarction; sudden or nonsudden coronary death); hard CHD (includes coronary revascularization; nonfatal myocardial infarction;
death from CHD); total CHD (includes angina; coronary insufficiency; hard CHD); and CVD (includes total CHD, stroke [including transient ischemia], peripheral vascular disease,
congestive heart failure).
zBaseline risk: observed risk for an event in the validation cohort (for the reported follow-up period).
§Predicted risk for an event when the Framingham model was applied to the validation cohort (for the given follow-up period).
tValues b1.0 indicate underprediction; values N1.0 indicate overprediction.
bThe number of quality items grouped to the categories byes,Q bno,Q and bnot clearQ.
American Heart Journal
Volume 153, Number 5
Eichler et al 725
Table I (continued)
Baseline
riskzz
Predicted
risk§§
Predictive
capacity: P/Ot
t
Q-item (yes/no/
not clear)bb
Comment
677
10.2%
16.0%
P/O: 1.57
7/0/2
Data extracted for combined outcome (all coronary events)
276
8.6%
10.7%
P/O: 1.24
7/0/2
Data extracted only for the first 10-y follow-up period
219
205
8.0%
5.3%
17.0%
5.0%
P/O: 2.12
P/O: 0.94
6/2/1
6/0/3
Applied Framingham model: ATP III guidelines
NHANES I
142
4.2%
4.6%
P/O: 1.08
6/0/3
NHANES II
190
3.3%
2.1%
P/O: 0.63
6/0/3
445
6.4%
3.8%
P/O: 0.60
6/0/3
218
0.9%
1.4%
P/O: 1.54
6/0/3
240
2.9%
3.3%
P/O: 1.13
6/0/3
82
1.3%
1.6%
P/O: 1.27
6/0/3
120
197
304
5.0%
2.7%
5.2%
6.7%
6.3%
14.2%
P/O: 1.34
P/O: 2.35
P/O: 2.70
5/1/3
5/1/3
6/0/3
84
8.2%
11.2%
P/O: 1.37
7/2/0
307
5.6%
9.8%
P/O: 1.77
6/1/2
146
5.1%
10.2%
P/O: 2.00
6/1/2
411
6.5%
5.6%
P/O: 0.87
6/0/3
110
12.6%
8.4%
P/O: 0.67
8/0/1
206
14.8%
15.9%
P/O: 1.08
7/0/2
529
31.1%
23.6%
P/O: 0.76
6/1/2
192
10.7%
11.6%
P/O: 1.08
6/2/1
269
33.7%
19.2%
P/O: 0.57
6/2/1
148
783
3.1%
6.0%
3.4%
8.9%
P/O: 1.09
P/O: 1.49
5/1/3
6/2/1
87
7.0%
7.6%
P/O: 1.09
6/2/1
Events (n)
Participants included from different Italian cohorts
Framingham model: ATP III guidelines; participants:
volunteers at high risk for CHD
Population: 85% European or other ethnicity, 10% Maori,
5% Pacific people; additional data from related article26
Population: siblings of hospitalized patients with premature
CHD; 15% black
Population: community dwelling men of the greater
Boston area (98% white)
Framingham model: ATP III guidelines; population:
noninstitutionalized residents of the town Dubbo
Population: 69% white, 23% Indian, 5% Afro-Caribbean,
3% Chinese
1251 of 3293 participants of the placebo control
group of the WOSCOPS included in this analysis
American Heart Journal
May 2007
726 Eichler et al
We applied the following end point definitions: CHD death
(includes fatal myocardial infarction, sudden or nonsudden
coronary death); hard CHD (includes coronary revascularization, nonfatal myocardial infarction, CHD death); total CHD
(includes angina, coronary insufficiency, hard CHD); CVD
(cardiovascular disease, includes total CHD, stroke [including
transient ischemia], peripheral vascular disease, congestive
heart failure).
Studies had to evaluate the Framingham risk score in a
validation population different from the derivation population8 and free of documented CVD at study entry. A
comparison of predicted outcome events in the validation
population (according to the Framingham risk score) with
observed events in the validation population had to be
provided. We did not focus on a specific version of the
Framingham equation.
We excluded studies investigating highly selected populations not representative for primary care (eg, patients with
end stage renal disease), cohorts of trials with active
intervention, cohorts with b70 events during follow-up,9 and
studies with a considerable shorter follow-up period (ie,
b80%) compared to the follow-up time of the risk prediction
model. If data of a specific cohort were published in several
papers or if follow-up data were presented, we included each
cohort only once.
Data extraction
Two reviewers independently screened titles and abstracts
for relevance and assessed potentially relevant studies for
inclusion by full text. Agreement between reviewers was
assessed using chance-adjusted n statistics. We resolved
disagreements by consensus meetings. Data were extracted by
one reviewer using a predefined form and checked independently by a second reviewer.
We extracted data on general study information (eg, length
of follow up, applied version of the Framingham risk score),
study setting (eg, level of population recruitment), population
details, predicted and observed events, and information about
risk classes.
We performed a quality assessment using selected criteria of
methodological standards for observational studies.7,8,10 Our
quality items covered aspects of internal and external validity
(eg, assessment of outcome blinded for predictive variables;
follow-up completeness; for details of quality assessment, see
Appendix C).
Statistical analysis
First, we evaluated calibration of the Framingham score
relying on data derived from risk classes. We generated a
calibration plot with the observed risk ( y axis) and the
predicted risk (x axis) for each cohort where information about
risk classes was available. We assessed strength of association
(as a prerequisite for adequate calibration) between predicted
and observed risk estimates for different risk classes. We
studied how closely the regression line of our model represented the data points (R 2, goodness of fit; % of variation
explained with the model). A consistently strong association
across several cohorts would allow for use of aggregated
(population level) data for all remaining analyses.
Second, we calculated the ratio of predicted risk to observed
risk (P/O) for each validation cohort (with the number of
predicted events according to the Framingham model in the
numerator and the number of observed events in the
validation cohort in the denominator). With a perfectly
calibrated score, the predicted risk is the same as the observed
risk (P/O ratio of 1.0).
Third, we divided our data set into 2 prespecified subgroups
to compare cohorts from Northern America, Australia, and
New Zealand (where guidelines generally recommend the
Framingham score) with cohorts from Europe (where additional risk scores are used). We constructed forest plots and
calculated pooled estimates of the P/O ratio using a random
effects model.
Fourth, we assessed factors that may influence calibration for
each population subgroup. With a metaregression analysis, we
evaluated the influence of a priori chosen independent factors
(eg, applied version of the Framingham model, start of followup, study size, or completeness of follow-up) on calibration.
Finally, we generated a calibration plot for each population
cluster. With a variance-weighted linear regression model,
weighted for the reciprocal value of the square of the standard
error of the baseline risk,11 we evaluated the relationship
between babsolute observed risk in %Q (as dependent variable)
and babsolute predicted risk in %Q (as independent variable).
With this approach, we focused on two features of calibration,
which are critical for application in practice:
(1)
(2)
Strength of association (R 2: goodness of fit; as applied
for risk classes)
Agreement of predicted with observed absolute risk
values: agreement is determined as the slope and the
intercept of the regression line in the calibration plot;
a slope of 1.0 (ie, h-coefficient for predicted risk: h =
1.0) and an intercept of zero (ie, a = 0.0) are identical
with the 458 concordance line and indicate perfect
agreement. A slope of N1.0 indicates underprediction,
and a value of b1.0, overprediction.
Analyses were performed using the STATA 8.2 software
package (StataCorp 2004, Stata Statistical Software, College
Station, TX).
Results
Description of validation studies
Our searches retrieved 1496 potentially relevant
studies (Figure 1). Seventeen validation studies of the
Framingham risk score4,5,9,12-25 with 25 independent
prospective validation cohorts met the inclusion criteria
(for excluded studies, see Appendix Table IV). Agreement between reviewers was high (chance adjusted n
statistics: 0.89).
Seventeen cohorts provided data for hard CHD or total
CHD. One cohort reported CVD outcomes, of which
71% were total CHD events,17 and in 8 cohorts, the
outcome was CHD death (for study details, see Table I).
Definitions of clinical events and of end point categories
were often similar to Framingham criteria.27,28 For
example, the World Health Organization MONICA
diagnostic categories for myocardial infarction were
American Heart Journal
Volume 153, Number 5
Eichler et al 727
Figure 2
Calibration of 25 Framingham validation studies according to reported end points: death from CHD (A) and total or hard CHD (B). For each end
point, group studies are sorted for increasing baseline risk (calculated as 10-year risk in %). For study acronyms and references, see Table I.
Framingham model: Wilson*, Adult Treatment Panel III version2; DPCG, Framingham model of the Diverse Population Collaborative Group.9
derived from electrocardiogram, enzymes, symptoms,
and autopsy findings.4 To determine end points during
follow-up, mainly medical records as well as study
registers (via International Classification of Diseases
codes) were analyzed. Information about CHD death
was also derived from death certificates (validated by
chart review and autopsy findings, where available).
A review panel for ascertainment of diagnoses was
established in 8 cohorts.
The 25 prospective validation cohorts included
128 000 persons. Population sizes varied from 798 to
24 204 participants (median 3856, interquartile range
1800-6354). Participants were recruited from general
populations or occupational settings (n = 23 cohorts)
and from clinics (n = 2 cohorts). Cohorts were set up in
the United States (n = 5), in Europe (n = 18), and in
Australia/New Zealand (n = 2). Age of participants in the
validation cohorts ranged from 30 to 79 years at
728 Eichler et al
American Heart Journal
May 2007
Figure 3
Calibration of Framingham validation studies according to population groups: cohorts from the United States, Australia, and New Zealand (A)
and from Europe (B). For each population group, studies are sorted for increasing baseline risk (calculated as 10-year risk). White rectangles
represent cohorts reporting total or hard CHD; black triangles, cohorts reporting death from CHD.
inclusion; the mean follow-up was 8.8 years (median
8 years, interquartile range 8-10) and started between
1961 and 1992. Observed event rates ranged from 4.8%
to 33.7% for hard/total CHD and from 0.5% to 4% for
CHD death end points. The Framingham equations
published by Wilson27 and Anderson29 were used in
18 cohorts; in 7 cohorts, a proportional hazards
Framingham function for CHD death was applied.9
cohort such as Framingham risk ranges or age ranges of
participants). We found a consistently close linear
association between predicted and observed estimates
for risk classes within cohorts (range of the linear model
fit for seven cohorts, R 2 = 0.903-0.998; for details, see
Appendix A). Thus, we relied on aggregated (population
level) data for our further analyses.
Methodological quality
Overall quality of studies for our main analysis (covering
25 cohorts) was moderate to good. Description of
population recruitment and participant characteristics
was given for all 25 cohorts. Information in detail for the
clinical end points was provided in 24 of 25 cohorts;
specific definitions of applied predictive variables were
less often available (details given for 19 cohorts).
For only 2 of 25 cohorts, information was available that
outcome assessors were blinded for predictive variables.16,18 Information about completeness of inclusion
(defined as inclusion of z75% of the screened population) and completeness of follow-up (defined as follow-up
of z90% of the included population) was scarce (criteria
fulfilled in 1 and in 7 of 25 cohorts, respectively).
Calibration for different end points
For cohorts reporting CHD death (Figure 2, A),
calibration varied from underprediction (P/O 0.60 in a
study from the United Kingdom9) to overprediction of
absolute risk (P/O 1.54 in a cohort from Norway9).
Variation was more pronounced for cohorts reporting
total or hard CHD (from P/O 0.57 in a cohort from the
United Kingdom22 to P/O 2.70 in a study from Italy,15
Figure 2, B). Pooled estimates of the P/O ratio were 0.99
(95% CI 0.85-1.16, I 2 = 98.4%) for cohorts reporting
death from CHD and 1.30 (95% CI 1.10-1.52, I 2 = 99.2%)
for total and hard CHD. Because the distribution pattern
of P/O was similar for both outcome categories and fatal
coronary events are known to be highly correlated with
total coronary events,28 we relied on both outcome
categories for our further analysis.
Association of predicted and observed estimates for
risk classes
Six studies4,5,15,17,20,21 with 7 cohorts (2 cohorts from
Australia/New Zealand and 5 from Europe) reported risk
estimates for single risk classes (3-14 risk classes per
Calibration for population clusters
The Framingham score was well calibrated to
predict absolute risk for the included seven cohorts
from the United States, Australia, and New Zealand
(n = 18 658) (Figure 3, A). The 5 biggest studies
American Heart Journal
Volume 153, Number 5
Eichler et al 729
Figure 4
Association between predicted and observed coronary heart disease event rates using a weighted linear regression model: cohorts from the
United States, Australia, New Zealand (A) and from Europe (B). Each circle represents a single cohort. The circle area represents the weight given
to each cohort in the linear regression model. The solid fitting lines show the relationship between predicted and observed risk. Data points above
the dashed 458 concordance line represent underprediction; data points below this line represent overprediction.
showed a P/O ratio between 0.87 and 1.08, and the
pooled estimate for 7 studies was 0.99 (95% CI
0.90-1.10, I 2 = 96.6%). For the 18 cohorts of various
European populations (n = 109 499) (Figure 3, B),
calibration varied over a wider range (P/O ratio:
minimum 0.57, maximum 2.70) with 14 of 18 studies
showing overprediction of risk. The pooled P/O ratio
for 18 European studies was 1.28 (95% CI 1.08-1.51,
I 2 = 99.2%). In our metaregression analysis, we did
not find a significant association of our predefined
factors (eg, applied version of the Framingham model)
with the P/O ratio in both population clusters.
The linear regression model showed a strong association between predicted and observed population risks
(R 2 = 0.84) for the 7 cohorts from the United States,
Australia, and New Zealand (Figure 4, A). In addition,
agreement of predicted and observed absolute risk was
high (coefficients for predicted risk: h = 0.84 [95% CI
0.41-1.26]; a = .01 [95% CI 0.02 to 0.04]) with the
regression line near to the 458 concordance line. For the
18 European cohorts, predicted and observed risk
estimates were closely associated too (R 2 = 0.72)
(Figure 4, B). Although the regression line was well
calibrated on the intercept (a = .003 [95% CI 0.01 to
0.01], the overall agreement of predicted and observed
absolute risk was poor because of substantial overprediction (h = .58, 95% CI 0.39-0.77).
A sensitivity analysis showed that the results remained
basically unchanged after exclusion of 2 studies with
high event rates of N30%20,22 or of one study that also
reported CVD events.17
Discussion
The Framingham risk score is well calibrated to
predict absolute risk for first coronary events in
populations from the United States, Australia, and New
Zealand. Mean predicted and observed population risk
estimates are closely associated and show a high
agreement. In European populations, overall agreement
of predicted and observed absolute risk is poor due to
substantial overestimation.
Strengths and limitations
We applied a thorough search strategy with a stepwise
retrieval of studies using electronic databases and
additional sources. We cannot exclude having missed
validation studies, but we believe that we found a near
complete sample of relevant papers. For study selection,
we insisted on accepted methodological standards for
validation studies and were interested in transferability
of results to primary care.
The analysis on the population level was based on the
assumption that the consistency in the relationship
American Heart Journal
May 2007
730 Eichler et al
between predicted and observed estimates for risk
classes in 7 cohorts would also apply for the remaining
studies. We believe that this approach is reasonable. The
strong linear relationship was detectable in cohorts of
both population subgroups. While the ideal way for this
research question is by using individual patient data, we
had to rely on aggregated study data. Thus, the observed
relationships are relevant to the populations that
comprise the 25 cohorts. Inferring these populationbased results to risk prediction in individuals may lead to
bias known as ecological fallacy.30 In addition, the
relatively low number of studies in the United States/
Australia/New Zealand group reduced precision of the
regression coefficients, and the 95% CIs for the slope of
the regression lines of both population clusters overlap.
We interpreted pooled estimates cautiously because
statistical heterogeneity between studies was considerable, and metaregression did not reveal significant
associations of study descriptors with study results.
However, we judge the analyzed cohorts as fairly similar
from a clinical and methodological point of view
(cohorts mostly population-based, participants free of
CVD, and of similar age ranges; comparable length of
follow-up; definitions of outcome categories similar to
Framingham criteria).
We did not evaluate discriminative ability of the
Framingham risk score (ie, how well high- and low-risk
groups are discriminated on the basis of the score) and
focused on calibration. Although discrimination is an
important feature of prediction rules, the results are
difficult to interpret for clinicians,31 and calibration, that
is, agreement of predicted with observed absolute risk,
shows a higher variability between populations.32
Existing systematic reviews and implications
for practice
Because validation studies of the Framingham score
show inconsistent results, we believe that a systematic
review of this prognostic prediction rule is an appropriate approach to weight the overall evidence.
Much work has been done in a systematic review by
D’Agostino et al32 who addressed calibration in US
cohorts of multiple ethnic groups. We were not able to
extract sufficient information for predicted and observed events to include data of this study in our
systematic review. A recent review by Brindle et al33 has
also assessed the accuracy of risk assessment with the
Framingham score for primary prevention of cardiovascular disease. This review differs from ours: we excluded
7 studies that were included by Brindle et al but did not
meet our inclusion criteria (cohorts with up to 20% of
persons with established CVD, active intervention
groups of clinical trials, or cohorts with b70 events).
Conversely, we included 8 studies that reported CHD
death end points, retrieved 4 additional more recent
studies with a search up to August 2005, and performed
a formal assessment of heterogeneity and study quality.34 In addition, our approach to assess different
population subgroups separately had substantial influence on the overall interpretation of results.
Guidelines from the United States,2 Australia,35 and
New Zealand36 have incorporated the Framingham risk
score in their recommendations for risk assessment in
primary prevention of CHD. This systematic review adds
further empiric foundation for this decision. The
Framingham risk score is an adequate instrument to
estimate absolute coronary risk in populations free of
CVD from the United States, Australia, and New Zealand.
The Framingham score, in its present form, leads to
substantial overprediction of coronary risk in European
populations, which may be due to different risk factor
distributions and event rates.32 Accordingly, one might
consider a recalibration procedure of the Framingham
score with European data to account for this overestimation.5,37 Similar activities have been undertaken for
populations from Asia,38 and additional epidemiological
research may be needed in other regions to evaluate
transferability of the Framingham score worldwide.
To intend a more precise risk prediction in European
populations, the third Joint European Guidelines39
recommend the use of the EU-SCORE, a risk score based
on data from European cohort studies. Whether this
instrument, however, shows a stronger association and
higher agreement between predicted and observed risk
estimates compared to a recalibrated Framingham score
merits further research in a direct comparison in
European populations.
Conclusion
The Framingham risk score is an adequate instrument
to estimate risk of first coronary events in populations
from the United States, Australia, and New Zealand.
Estimation of absolute risk for first coronary events in
European cohorts is limited by overprediction and
requires adjustment.
We thank Dr Alfons Kessels (MD, MSc; statistician,
Department of Clinical Epidemiology and Medical
Technology Assessment, University Hospital Maastricht, The Netherlands) for methodological and statistical advice. We would like to especially
acknowledge the helpful assistance of Dr. Ewout W.
Steyerberg, (PhD, Center for Medical Decision Making,
Department of Public Health, Erasmus MC, University
Medical Center, Rotterdam, The Netherlands) for
methodological support and critical review during
manuscript writing.
References
1. Wood D. Joint British recommendations on prevention of coronary
heart disease in clinical practice. Heart 1998;80:S1-S29.
American Heart Journal
Volume 153, Number 5
2. Anonymous. Executive Summary of The Third Report of The National
Cholesterol Education Program (NCEP) Expert Panel on Detection,
Evaluation, and Treatment of High Blood Cholesterol in Adults
(Adult Treatment Panel III). JAMA 2001;285:2486 - 97.
3. Sheridan S, Pignone M, Mulrow C. Framingham-based tools to
calculate the global risk of coronary heart disease: a systematic
review of tools for clinicians. J Gen Intern Med 2003;18:1039 - 52.
4. Hense HW, Schulte H, Lowel H, et al. Framingham risk function
overestimates risk of coronary heart disease in men and women
from Germany—results from the MONICA Augsburg and the
PROCAM cohorts. Eur Heart J 2003;24:937 - 45.
5. Brindle P, Emberson J, Lampe F, et al. Predictive accuracy of the
Framingham coronary risk score in British men: prospective cohort
study. BMJ 2003;327:1267.
6. Madhok V, Fahey T. Cardiovascular risk estimation: important but
may be inaccurate. BMJ 2006;332:1422.
7. NHS Centre for Reviews and Dissemination. CRD’s guidance for
those carrying out or commissioning reviews. New York7 University
of York; 2001.
8. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review
and suggested modifications of methodological standards. JAMA
1997;277:488 - 94.
9. Diverse Populations Collaborative Group. Prediction of mortality
from coronary heart disease among diverse populations: is there a
common predictive function? Heart 2002;88:222 - 8.
10. Altman DG. Systematic reviews of evaluations of prognostic
variables. BMJ 2001;323:224 - 8.
11. Altman DG, Machin D, Bryant TN, et al. Statistics with confidence:
confidence intervals and statistical guidelines. 2nd ed. London7 BMJ;
2000.
12. Brindle P, Montaner D, May M, et al. The effect of timescale on the
validity of the Framingham risk score: a prospective study. The 3rd
Conference of Epidemiological Longitudinal Studies in Europe
(CELSE). 2004; Bristol. Poster Session A: p. 57.
13. Cooper JA, Miller GJ, Humphries SE. A comparison of the PROCAM
and Framingham point-scoring systems for estimation of individual
risk of coronary heart disease in the Second Northwick Park Heart
Study. Atherosclerosis 2005;181:93 - 100.
14. Empana JP, Ducimetiere P, Arveiler D, et al. Are the Framingham
and PROCAM coronary heart disease risk functions applicable to
different European populations? The PRIME Study. Eur Heart J
2003;24:1903 - 11.
15. Ferrario M, Chiodini P, Chambless LE, et al. Prediction of coronary
events in a low incidence population. Assessing accuracy of the CUORE
Cohort Study prediction equation. Int J Epidemiol 2005;34:413 - 21.
16. Greenland P, LaBree L, Azen SP, et al. Coronary artery calcium
score combined with Framingham score for risk prediction in
asymptomatic individuals. JAMA 2004;291:210 - 5.
17. Milne R, Gamble G, Whitlock G, et al. Framingham Heart Study risk
equation predicts first cardiovascular event rates in New Zealanders
at the population level. N Z J Med 2003;116:U662.
18. Mora S, Yanek LR, Moy TF, et al. Interaction of body mass index and
Framingham risk score in predicting incident coronary disease in
families. Circulation 2005;111:1871 - 6.
19. Orford JL, Sesso HD, Stedman M, et al. A comparison of the
Framingham and European Society of Cardiology coronary heart
disease risk prediction models in the normative aging study. Am
Heart J 2002;144:95 - 100.
20. Ramachandran S, French JM, Vanderpump MPJ, et al. Using the
Framingham model to predict heart disease in the United Kingdom:
retrospective study. BMJ 2000;320:676 - 7.
Eichler et al 731
21. Simons LA, Simons J, Friedlander Y, et al. Risk functions for
prediction of cardiovascular disease in elderly Australians: the
Dubbo Study. Med J Aust 2003;178:113 - 6.
22. Stephens JW, Ambler G, Vallance P, et al. Cardiovascular risk and
diabetes. Are the methods of risk prediction satisfactory? Eur J
Cardiovasc Prev Rehabil 2004;11:521 - 8.
23. Thomsen TF, McGee D, Davidsen M, et al. A cross-validation of risk
scores for coronary heart disease mortality based on data from the
Glostrup Population Studies and Framingham Heart Study. Int J
Epidemiol 2002;31:817 - 22.
24. Tunstall-Pedoe H, Woodward M. By neglecting deprivation cardiovascular risk scoring will exacerbate social gradients in disease.
Heart 2006;92:307 - 10.
25. West of Scotland Coronary Prevention Study Group. Influence of
pravastatin and plasma lipids on clinical events in the West of
Scotland Coronary Prevention Study (WOSCOPS). Circulation
1998;97:1440 - 5.
26. Milne R, Gamble G, Whitlock G, et al. Discriminative ability of a
risk-prediction tool derived from the Framingham Heart Study
compared with single risk factors. N Z J Med 2003;116:U663.
27. Wilson PW, D’Agostino RB, Levy D, et al. Prediction of coronary heart
disease using risk factor categories. Circulation 1998;97:1837 - 47.
28. Anderson KM, Odell PM, Wilson PW, et al. Cardiovascular disease
risk profiles. Am J Heart 1990;121:293 - 8.
29. Anderson KM, Wilson PW, Odell PM, et al. An updated coronary
risk profile. A statement for health professionals. Circulation
1991;83:356 - 62.
30. Thompson SG, Higgins JP. How should meta-regression analyses be
undertaken and interpreted? Stat Med 2002;21:1559 - 73.
31. Hense HW. Observations, predictions and decisions-assessing
cardiovascular risk assessment. Int J Epidemiol 2004;33:235 - 9.
32. D’Agostino Sr RB, Grundy S, Sullivan LM, et al. Validation of the
Framingham coronary heart disease prediction scores: results of a
multiple ethnic groups investigation. JAMA 2001;286:180 - 7.
33. Brindle PM, Beswick AD, Fahey TF, et al. The accuracy and impact
of risk assessment in the primary prevention of cardiovascular
disease: a systematic review. Heart 2006;92:1752 - 9.
34. Eichler K, Puhan M, Steurer J, et al. Accuracy of risk assessment in
primary prevention of cardiovascular disease (letter). Heart 2006
Accessed at http://heart.bmjjournals.com/cgi/eletters/
hrt.2006.087932v3 [on 5 July 2006].
35. The Cardiac Society of Australia and New Zealand. Lipid
management guidelines–2001. National Heart Foundation of
Australia, The Cardiac Society of Australia and New Zealand. Med
J Aust 2001;175(Suppl):S57.
36. Jackson R. Updated New Zealand cardiovascular disease riskbenefit prediction guide. BMJ 2000;320:709 - 10.
37. Marrugat J, R DA J, Sullivan L, et al. An adaptation of the
Framingham coronary heart disease risk function to European
Mediterranean areas. J Epidemiol Community Health
2003;57:634 - 8.
38. Liu J, Hong Y, Agostino DRBS, et al. Predictive value for the Chinese
population of the Framingham CHD risk assessment tool compared
with the Chinese Multi-Provincial Cohort Study. JAMA
2004;291:2591 - 9.
39. De Backer G, Ambrosioni E, Borch-Johnsen K, et al. European
guidelines on cardiovascular disease prevention in clinical practice:
third joint task force of European and other societies on cardiovascular disease prevention in clinical practice (constituted by
representatives of eight societies and by invited experts). Eur J
Cardiovasc Prev Rehabil 2003;10:S1-S10.
American Heart Journal
May 2007
Eichler et al 731.e1
Appendix A
Figure 5
Association between predicted and observed coronary heart disease event rates using estimates from risk classes: cohorts with data for risk
classes from Australia and New Zealand (Figure 5, A) and European cohorts (Figure 5, B). Each data point represents a single risk class. For each
risk class, bpredicted riskQ is plotted against bobserved risk.Q R 2, linear model fit. References of validation cohorts: Australia, Dubbo cohort,
Simons et al21; New Zealand, Fletcher CHH Study, Milne et al17; United Kingdom, Caerphilly and Speedwell, Brindle et al5; Germany, MONICA,
Hense et al4; Germany, PROCAM, Hense et al4; United Kingdom, Wickham Study, Ramachandran et al20; Italy, CUORE, Ferrario et al.15
American Heart Journal
Volume 153, Number 5
731.e2 Eichler et al
Appendix B. Detailed search strategy
Query Number
Database
Search terms
Search terms in MEDLINE (MEZZ):
1
MEZZ
2
MEZZ
3
MEZZ
4
MEZZ
5
MEZZ
6
MEZZ
7
MEZZ
8
MEZZ
9
MEZZ
10
MEZZ
11
MEZZ
12
MEZZ
13
MEZZ
14
MEZZ
15
MEZZ
16
MEZZ
17
MEZZ
18
MEZZ
19
MEZZ
20
MEZZ
21
MEZZ
22
MEZZ
23
MEZZ
24
MEZZ
25
MEZZ
26
MEZZ
27
MEZZ
28
MEZZ
29
MEZZ
30
MEZZ
31
MEZZ
32
MEZZ
CARDIOVASCULAR-DISEASES#
RISK# RISK.TI.
PREDICT$ DECISION$ DECISION-SUPPORT-TECHNIQUES#
1 AND 2 AND 3
FRAMINGHAM
NCEP
NATIONAL WITH CHOLESTEROL WITH EDUCATION
SYSTEMATIC WITH RISK WITH EVALUATION
EUROP$5 WITH SCORE
PROCAM
MONICA
SHEFFIELD
BRHS
BRITISH WITH REGIONAL WITH HEART
DUNDEE
BHF
(BRITISH NATIONAL) WITH HEART WITH FOUNDATION
GCRS
GLOBAL WITH RISK
PRIME WITH STUDY
FINRISK
FINNMARK
SCOTTISH WITH HEART
GLOSTRUP
BIRNH
RIFLE
PARIS WITH PROSPECTIVE
CATALONIA WITH HEART
FACTORY WITH HEART
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
20 21 22 23 24 25 26 27 28 29 30
4 AND 31
Search terms in BIOSIS (BIZZ):
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
CARDIOVASCULAR-RISK
145$3.CC.
RISK.TI.
RISK.DE.
35 36
34 AND 37
33 38
39 AND (PREDICT$5 DECISI$10)
FRAMINGHAM
NCEP
NATIONAL WITH CHOLESTEROL WITH EDUCATION
SYSTEMATIC WITH RISK WITH EVALUATION
EUROP$5 WITH SCORE
PROCAM
MONICA
SHEFFIELD
BRHS
BRITISH WITH REGIONAL WITH HEART
DUNDEE
BHF
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
American Heart Journal
May 2007
Eichler et al 731.e3
Appendix B (continued)
Query Number
Database
Search terms in BIOSIS (BIZZ):
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
BIZZ
Search terms in EMBASE (EMZZ):
69
EMZZ
70
EMZZ
71
EMZZ
72
EMZZ
73
EMZZ
74
EMZZ
75
EMZZ
76
EMZZ
77
EMZZ
78
EMZZ
79
EMZZ
80
EMZZ
81
EMZZ
82
EMZZ
83
EMZZ
84
EMZZ
85
EMZZ
86
EMZZ
87
EMZZ
88
EMZZ
89
EMZZ
90
EMZZ
91
EMZZ
92
EMZZ
93
EMZZ
94
EMZZ
95
EMZZ
96
EMZZ
97
EMZZ
98
EMZZ
99
EMZZ
100
EMZZ
101
EMZZ
102
EMZZ
Search terms
(BRITISH NATIONAL) WITH HEART WITH FOUNDATION
GCRS
GLOBAL WITH RISK
PRIME WITH STUDY
FINRISK
FINNMARK
SCOTTISH WITH HEART
GLOSTRUP
BIRNH
RIFLE
PARIS WITH PROSPECTIVE
CATALONIA WITH HEART
FACTORY WITH HEART
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55
56 57 58 59 60 61 62 63 64 65 66
40 AND 67
CARDIOVASCULAR-DISEASE#
RISK.TI.
CARDIOVASCULAR-RISK
69 AND 70
71 72
73 AND (PREDICT$ DECISI$10)
FRAMINGHAM
NCEP
NATIONAL WITH CHOLESTEROL WITH EDUCATION
SYSTEMATIC WITH RISK WITH EVALUATION
EUROP$5 WITH SCORE
PROCAM
MONICA
SHEFFIELD
BRHS
BRITISH WITH REGIONAL WITH HEART
DUNDEE
BHF
(BRITISH NATIONAL) WITH HEART WITH FOUNDATION
GCRS
GLOBAL WITH RISK
PRIME WITH STUDY
FINRISK
FINNMARK
SCOTTISH WITH HEART
GLOSTRUP
BIRNH
RIFLE
PARIS WITH PROSPECTIVE
CATALONIA WITH HEART
FACTORY WITH HEART
75 76 77 78 79 80 81 82 83 84 85 86 87 88 89
90 91 92 93 94 95 96 97 98 99 100
74 AND 101
American Heart Journal
May 2007
731.e4 Eichler et al
Appendix C. Details of quality assessment of 25 validation cohorts of the Framingham risk score:
Author, y
Brindle et al, 20035
Brindle, 200412
Cooper et al, 200513
DPCG, 20029
DPCG, 20029
DPCG, 2002
DPCG, 2002
DPCG, 2002
DPCG, 2002
DPCG, 2002
Empana et al, 200314
Empana et al, 2003
Ferrario et al, 200515
Greenland et al, 200416
Hense et al, 20034
Hense et al, 2003
Milne et al, 200317
Mora et al, 200518
Orford et al, 200219
Ramachandran
et al, 200020
Simons et al, 200321
Stephens et al, 200422
Q-item 1
Q-item 2
Q-item 3
Q-item 4
Outcome
assessment:
blinded*
Follow-up:
lengthyy
Follow-up:
completenesszz
Population:
characteristics§§
British regional heart study cohort
Caerphilly and Speedwell cohort
NPHS-II
NHANES I, men
NHANES II, men
Scottish Collaborative Study, men
Renfrew and Paisley Study, men
Norwegian Counties Study, men
Iceland Reykjavik Study, men
Yugoslavia Cardiovascular
Disease Study, men
PRIME Study, Ireland
PRIME Study, France
CUORE cohorts, men
South Bay Heart Watch cohort
MONICA cohort, men
PROCAM cohort
Fletcher Challenge Heart and
Health Study
John Hopkins Sibling Study
Normative Aging Study
Wickham Study
?
?
?
?
?
?
?
?
?
+
+
+
+
+
+
+
+
+
+
+
?
?
?
?
?
?
?
?
+
+
+
+
+
+
+
+
+
+
?
?
?
+
?
?
?
+
+
+
+
+
+
+
?
?
?
+
+
?
+
+
+
+
+
+
+
+
?
?
+
+
+
+
+
+
+
+
+
Dubbo Study
University College London
Hospitals cohort
Glostrup cohort
SHHEC
?
+
+
+
+
+
?
+
+
?
?
+
+
?
+
+
Validation cohort
Thomsen et al, 200223
Tunstall-Pedoe and
Woodward, 200524
WOSCOPS Group, 199825 WOSCOPS (placebo control
group of RCT)
The symbols refer to the result of the quality assessment: b+,Q criterion fulfilled; b,Q criterion not fulfilled; b?,Q no information available;
DPCG, Diverse Populations Collaborative Group; SHHEC, Scottish Heart Health Extended Cohort.
*Assessment of outcome blinded for predictive variables: fulfilled if stated in the text or all cause mortality is the outcome.
yFollow-up length in the validation cohort comparable to the follow-up time of the prediction rule: fulfilled if longer than 80% (or accounted for shorter follow-up in analysis).
zCompleteness of follow up: fulfilled if completeness z90%.
§Population characteristics described in sufficient detail: fulfilled if at least data for sex, age, and morbidities are provided.
tRecruitment of population described in sufficient detail: fulfilled if qualitative description is provided.
bCompleteness of inclusion of eligible patients (participation rate): fulfilled if completeness z75%.
#Q-items indicates the number of quality items grouped to the categories yes (+), no (), and not clear (?).
American Heart Journal
May 2007
Eichler et al 731.e5
Q-item 5
Q-item 6
Q-item 7
Q-item 8
Q-item 9
Q-items
Population:
recruitmentt
Inclusion:
completenessb
Predictive variables:
definition provided
Outcome:
definition
provided
Math. techniques:
description
provided
(Yes/ no/
not clear)#
?
+
?
?
?
?
?
?
?
?
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
7/0/2
7/0/2
6/2/1
6/0/3
6/0/3
6/0/3
6/0/3
6/0/3
6/0/3
6/0/3
?
?
?
?
?
?
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
5/1/3
5/1/3
6/0/3
7/2/0
6/1/2
6/1/2
6/0/3
?
?
?
+
+
+
+
+
+
+
+
+
+
+
8/0/1
7/0/2
6/1/2
?
+
+
+
+
+
+
+
6/2/1
6/2/1
?
+
+
+
+
+
+
+
5/1/3
6/2/1
+
+
+
+
6/2/1
American Heart Journal
Volume 153, Number 5
731.e6 Eichler et al
Appendix D. Details of excluded studies (sorted for author)
Author, y
Validation cohort*,
location
Sample
size, n
Recruitment
Clinical outcomey
follow-up)
Assmann et al,
200240
PROCAM cohort;
Germany
5389
OCC
Hard CHD (10 y)
Bastuji et al,
200241
INSIGHT cohort
(from RCT);
Europe
4147
Hospital hypertension
centers in Italy and
GP in the remaining
countries
Total CHD (4 y)
Boutouyrie et al,
200242
Hopital Broussais
cohort; France
1045
Total CHD (NA)
Brand et al,
197643
Western Collaborative
Group Study cohort;
United States
3154
Outpatient
hypertension
clinic of a
Paris hospital
OCC
Total CHD (12 y)
D’Agostino
et al, 200132
6 Ethnically diverse
US cohort studies
23424
NA
Hard CHD (5 y)
Ducimetiere
et al, 198044
Paris Prospective
Study cohort;
France
7434
OCC
Hard CHD (9 y)
Dunder, 200445
ULSAM cohort;
Sweden
LRC Study cohort;
United States
1108
50-year-old men
from Uppsala
Random sample of
North American clinics
Hard CHD (10 y)
428
GP
Total CHD (10 y)
Kornitzer and
Koyunco,
200048
Poole Diabetes
study cohort;
United Kingdom
Belgian Physical
Fitness Study
cohort; Belgium
2186
OCC
Total CHD (10 y)
Kornitzer and
Koyunco, 2000
BIRNH cohort;
Belgium
6068
Randomly chosen
subjects from
4 communes
Total CHD (10 y)
Liu et al, 200438
Chinese CMCS
cohort; China
30121
Centers of
MONICA-sino
project
Hard CHD (10 y)
Grover et al,
199546
Guzder et al,
200547
3678
Total CHD (12 y)
Reason for exclusion
Not sufficient data to
assess the association
between predicted
and observed risk;
PROCAM cohort
covered with data
from Hense et al4
Cohort of trial with
active intervention in
both groups (RCT of
hypertension
treatment: nifedipine
vs co-amilorid)
b70 Events during
follow-up
Different follow up in
the validation cohort
compared to the
applied risk model
Not sufficient data to
assess the association
between predicted
and observed risk
Not sufficient data to
assess the association
between predicted
and observed risk
b70 Events during
follow-up
Not sufficient data to
assess the association
between predicted
and observed risk
b70 Events during
follow-up.
Not sufficient data to
assess the association
between predicted
and observed risk;
b70 events during
follow-up.
Not sufficient data to
assess the association
between predicted
and observed risk
Not sufficient data to
assess the association
between predicted
and observed risk
American Heart Journal
May 2007
Eichler et al 731.e7
Appendix D (continued)
Author, y
Validation cohort*,
location
Sample
size, n
Menotti et al,
200049
IRA cohort
(Italian Rural
Area); Italy
1656
Jimeno Mollet
et al, 200550
Protopsaltis
et al, 200451
Raval Sud Study
cohort; Spain
Piraeus Diabetes
Centre Database;
Greece
Schulz et al,
200352
Recruitment
Clinical outcomey
follow-up)
Total CHD (10 y)
112
Sample of 2 rural
communities in
northern and
central Italy
NA
Total CHD (10 y)
339
NA
Total CHD (10 y)
Bad Bevensen
cohort; Germany
42
Hard CHD (10 y)
Shang et al,
200453
Stern et al,
200454
Japanese Collaborative
Cohort Study; Japan
SAHS cohort;
United States
359
Patients referred
to hospital for
cardiologic evaluation
OCC
Suka et al,
200255
Japanese cohort;
Japan
von Eckardstein
et al, 200456
Wang and Hoy,
200557
2570
Total CHD (10 y)
Total CHD (10 y)
5611
Randomly selected
from 3 types of
neighborhoods
in San Antonio, TX
OCC
Total CHD (10 y)
PROCAM cohort;
Germany
4818
OCC
Hard CHD (10 y)
Australian Aboriginal
cohort; Australia
687
Community wide
screening program
Total CHD (10 y)
Reason for exclusion
Not sufficient data to
assess the association
between predicted
and observed risk
b70 Events during
follow-up
Not sufficient data to
assess the association
between predicted
and observed risk
b70 Events during
follow-up
b70 Events during
follow-up
Not sufficient data to
assess the association
between predicted
and observed risk
Different follow up in
the validation cohort
compared to the
applied risk model
Not sufficient data to
assess the association
between predicted
and observed risk.
PROCAM cohort
covered with data
from Hense et al 4
b70 Events during
follow-up
RCT, randomized controlled trial.
*Design: prospective cohort study (if not stated otherwise).
yDeath from CHD (includes fatal myocardial infarction; sudden or nonsudden coronary death); hard CHD (includes coronary revascularization; nonfatal myocardial infarction;
death from CHD); total CHD (includes angina; coronary insufficiency; hard CHD); and CVD (includes total CHD, stroke [including transient ischemia], peripheral vascular disease,
congestive heart failure).
731.e8 Eichler et al
References of Appendix
40. Assmann G, Cullen P, Schulte H. Simple scoring scheme for
calculating the risk of acute coronary events based on the 10-year
follow-up of the Prospective Cardiovascular Munster (PROCAM)
study. Circulation 2002;105:310 - 5.
41. Bastuji S, Garin A, Deverly D, et al. The Framingham prediction rule
is not valid in a European population of treated hypertensive
patients. J Hypertens 2002;20:1973 - 80.
42. Boutouyrie P, Tropeano AI, Asmar R, et al. Aortic stiffness is an
independent predictor of primary coronary events in hypertensive
patients: a longitudinal study. Hypertension 2002;39:10 - 5.
43. Brand RJ, Rosenman RH, Sholtz RI, et al. Multivariate prediction of
coronary heart disease in the Western Collaborative Group Study
compared to the findings of the Framingham study. Circulation
1976;53:348 - 55.
44. Ducimetiere P, Richard JL, Cambien F, et al. Coronary heart disease
in middle-aged Frenchmen. Comparisons between Paris Prospective
Study. Seven Countries Study, and Pooling Project. Lancet
1980;1:1346 - 50.
45. Dunder K, Lind L, Zethelius B, et al. Evaluation of a scoring scheme,
including proinsulin and the apolipoprotein B/apolipoprotein A1
ratio, for the risk of acute coronary events in middle-aged men:
Uppsala Longitudinal Study of Adult Men (ULSAM). Am Heart J
2004;148:596 - 601.
46. Grover SA, Coupal L, Hu XP. Identifying adults at increased risk of
coronary disease. How well do the current cholesterol guidelines
work? JAMA 1995;274:801 - 6.
47. Guzder RN, Gatling W, Mullee MA, et al. Prognostic value of the
Framingham cardiovascular risk equation and the UKPDS risk
engine for coronary heart disease in newly diagnosed type 2
diabetes: results from a United Kingdom study. Diabet Med
2005;22:554 - 62.
American Heart Journal
Volume 153, Number 5
48. Kornitzer M, Koyunco R. Multifactorial approach to the prevention
of coronary heart disease: from computer to paper and pencil?
J Cardiovasc Risk 2000;7:201 - 7.
49. Menotti A, Puddu PE, Lanti M. Comparison of the Framingham risk
function-based coronary chart with risk function from an Italian
population study. Eur J Heart 2000;21:365 - 70.
50. Jimeno Mollet J, Molist Brunet N, Franch Nadal J, et al. Variabilidad
en la estimacion del riesgo coronario en la diabetes mellitus tipo 2.
Aten Primaria 2005;35:30 - 6.
51. Protopsaltis J, Kostadinopulos P, Kamaratos A, et al. Comparative
study of prognostic value for coronary disease risk between the U.K.
Prospective Diabetes Study and Framingham Models. Diabetes Care
2004;27:277 - 8.
52. Schulz H, Sinn R, Wolf R. Kardiales Risiko bei Männern mit
angiographisch normalen oder minimal arteriosklerotisch
veränderten Koronararterien. Z Kardiol 2003;92:245 - 53.
53. Shang E, Sakauchi F, Washio M, et al. Predictive index of coronary
heart disease (CHD) risk based on the results of health check up.
Sapporo Med J 2004;73:13 - 7.
54. Stern MP, Williams K, Gonzalez VC, et al. Does the metabolicsyndrome improve identification of individuals at risk of type 2
diabetes and/or cardiovascular disease? Diabetes Care
2004;27:2676 - 81.
55. Suka M, Sugimori H, Yoshida K. Validity of the Framingham risk
model applied to Japanese men. Methods Inf Med 2002;41:
213 - 5.
56. von Eckhardstein A, Schulte H, Assmann G. Vergleich internationaler Konsensus-Empfehlungen zur Erkennung des
präsymptomatischen Hochrisikopatienten fqr den Herzinfarkt in
Deutschland. Z Kardiol 2005;94:52 - 60.
57. Wang Z, Hoy WE. Is the Framingham coronary heart disease
absolute risk function applicable to Aboriginal people? Med J Aust
2005;182:66 - 9.