The Cambridge Risk Score (CRS) was developed to screen for type 2 diabetes mellitus risk. We assessed the ability of the CRS to predict glycosylated hemoglobin (HbA1c) levels and determined whether the CRS was better than body mass index (BMI) at predicting HbA1c levels in midlife.
We included 7452 participants without known diabetes in a biomedical survey of the 1958 British Birth Cohort at 45 years of age. Receiver operator characteristic curves were used to compare the ability of the CRS and BMI to identify individuals with elevated HbA1c levels using thresholds of 7.0% or more, 6.0% or more, and 5.5% or more.
Of the total sample, 0.9% (95% confidence interval [CI], 0.7%-1.1%) had HbA1c levels of 7.0% or more; 3.8% (95% CI, 3.2%-4.5%), 6.0% or more; and 24.4% (95% CI, 23.1%-25.9%), 5.5% or more. The CRS detected individuals with elevated HbA1c levels with reasonable accuracy (area under the curve, 0.84 for HbA1c level ≥7.0%; 0.76 for HbA1c level ≥6.0%). Similar area under the curve values were obtained using BMI alone (0.84 for HbA1c level ≥7.0%; 0.79 for HbA1c level ≥6.0%). When tested using the lower HbA1c threshold of 5.5% or more, the CRS and BMI did not perform well (areas under the curve, 0.65 and 0.63 for CRS and BMI, respectively). Both measures indicated that approximately 20% of the cohort were at increased risk of diabetes. Owing to the low prevalence of diabetes at 45 years of age, only 2% to 3% of those considered at risk had elevated HbA1c levels.
For a population in mid-adult life, the CRS identified individuals with elevated HbA1c levels reasonably well. However, the CRS had no advantage compared with BMI alone in identifying diabetes risk.
Type 2 diabetes mellitus is increasingly common and is predicted to continue to increase nationally and internationally.1 Average life expectancy is reduced by as much as 10 years in people with type 2 diabetes, and there is a considerable health burden related to microvascular and macrovascular complications.2 Approximately 15% to 20% of individuals with newly diagnosed diabetes present with retinopathy, and 5% to 10% have proteinuria.3 Disease onset can occur as early as 12 years before diabetes is diagnosed, because hyperglycemia develops gradually, with symptoms appearing later, and thus approximately half of all cases are undiagnosed.4 Although it is argued that there is no justification for general population screening,5 early detection of individuals at risk of diabetes could be beneficial because early intervention has the potential to prevent the development of diabetes and its complications.
Several risk scores for type 2 diabetes have recently been developed that use information collected from primary care. For example, the Cambridge Risk Score (CRS) was developed in a British population and uses information on age, smoking status, body mass index (BMI) (calculated as weight in kilograms divided by the square of height in meters), family history of diabetes, and prescribed antihypertensive medication and corticosteroids.6 The CRS has been found to perform well when compared against objective measures of glucose metabolism such as glycosylated hemoglobin (HbA1c) level.7 The HbA1c level is increasingly used in screening studies for type 2 diabetes after recent international standardization of its measurement.8 In addition, evidence suggests that individuals with HbA1c levels of greater than 5.0% (below thresholds used to detect impaired glucose tolerance) have a higher risk of cardiovascular disease and mortality, regardless of their diabetes status.9,10 A similar finding for mortality has also been reported for those with a high CRS.11 The identification of individuals at risk of diabetes and treatment of risk factors is therefore relevant to prevent cardiovascular disease and mortality in addition to diabetes.12
Diabetes risk scores such as the CRS are potentially informative in terms of later risk related to complications and mortality. However, no nationwide evaluation studies against objective measures of glucose metabolism exist. Previous studies have been limited to selected populations13 or conducted in geographically restricted populations.6,7 Hence, the CRS has not been analyzed in a sample nationally representative of Great Britain at midlife, when there may be disturbances in glucose metabolism but no apparent disease. This study uses data from the 1958 British Birth Cohort at 45 years of age to evaluate the CRS in a nationwide sample. Specifically we examine (1) the ability of the CRS to identify individuals with elevated HbA1c levels, using different HbA1c thresholds; and (2) whether the CRS is better than the BMI alone in predicting HbA1c levels in this midlife age group in whom obesity may be more important than other components of the CRS.
The 1958 Birth Cohort consists of data on approximately 17 000 individuals born during a single week in March 1958 in England, Scotland, and Wales, originally collected as part of the Perinatal Mortality Study.14 Survivors have been reinterviewed regularly at 7, 11, 23, 33, and 42 years of age.15,16 At 45 years of age, a survey of biomedical risk factors and disease outcomes was undertaken that included physical assessments and blood collection.17 Ethical approval for the biomedical survey was given by the South East Multi-Centre Research Ethics Committee, National Health Service (NHS), Great Britain. Of a target sample of 12 070, data were available from 9377 people who participated in the biomedical survey, and of these, 84.2% gave a blood sample from which an HbA1c measure was obtained (n = 7899). A further 329 (4.1%) were missing data required to calculate the CRS. For evaluation of the CRS in this population, individuals known to have type 1 or type 2 diabetes at the 42-year survey were excluded because any diagnostic test of glucose homeostasis would be altered by diabetes treatment (n = 118). The analysis sample used herein (n = 7452) is broadly similar to the original birth population. For example, for social class at birth, 18.8% of the analysis sample compared with 17.0% of the original population was in classes I and II; 21.7% compared with 24.3% of the original population was in classes IV and V. Thus, those from unskilled manual backgrounds were slightly underrepresented in the analysis sample.
Nonfasting venous blood samples were collected in citrate collection tubes. The HbA1c analyses were conducted by the Department of Clinical Biochemistry, Newcastle upon Tyne Hospitals NHS Trust, Newcastle upon Tyne, England, using high-performance liquid chromatography (HLC-723GHb V A1c2.2; Tosoh Corp, Tokyo, Japan)18 and standardized to Diabetes Control and Complications Trial values.8,19 The HbA1c level was dichotomized to allow analyses according to 3 different thresholds: 5.5% or more, 6.0% or more, and 7.0% or more. The 7.0% HbA1c threshold is associated with the diagnosis of diabetes by World Health Organization criteria20; 6.0%, with elevated HbA1c levels consistent with hyperglycemia.21-23 In addition, the lower value of 5.5% was used because values greater than 5% have been found to be associated with cardiovascular disease and mortality.9 Approximately 70% of our population had HbA1c values ranging from 5.0% to 6.9%; however, most of these (84%) were below 5.5%. Therefore, 5.5% was chosen as a lower HbA1c threshold against which to evaluate the performance of the CRS.
Measures to construct the CRS were obtained at the 45-year survey unless otherwise stated.
The variables used to construct the CRS consisted of age, sex, BMI, family history of diabetes, smoking, and prescribed antihypertensive medication and corticosteroids. Weight and standing height were measured without shoes and in light clothing by the nurse using scales and a stadiometer. The BMI was calculated as the weight in kilograms divided by the square of the height in meters. Smoking status was derived from information relating to historical and current behavior at 42 years of age. Information on currently prescribed medication was collected and antihypertensive medication and corticosteroids were identified for inclusion in the CRS. Information on first-degree relatives (parents or siblings) who had diabetes was obtained from 2 data sources. First, during the follow-up at 7 years of age, parents of the cohort members were asked “Is there a history of any diabetes in parents, brothers, or sisters?” This information was supplemented by data on parental mortality due to diabetes-related causes (original or underlying) by the end of December 2003 using codes E10 to E14 from the International Statistical Classification of Diseases, 10th Revision.17
The CRS was calculated as the probability (ranging from 0 to 1) of having type 2 diabetes using the coefficients derived by Griffin et al.6 The coefficients were derived from a logistic regression model that used data available from a diabetes prevalence survey of patients registered with a Cambridgeshire general practice (the Ely Study) and validated in a diabetes screening study of 41 practices in southern England (the Wessex Study). The CRS is the sum of the coefficients of the variables included in the model; the higher the score, the greater the chance of having diabetes. The probability of having type 2 diabetes is calculated using the following equation:
P = 1/1 + e −(−6.322 − 0.879 [Female] + 0.063 [Age] + 0.699 [BMI, 25.00-27.49] + 1.97 [BMI, 27.50-29.99] + 2.518 [BMI, ≥30.00] + 1.22 [Antihypertensive Medication] + 2.194 [Corticosteroids] + 0.728 [Parent or Sibling Diabetes] + 0.753 [Parent and Sibling Diabetes] − 0.218 [Ex-smoker] + 0.855 [Current Smoker]).
The ability of the CRS and BMI to identify people with elevated HbA1c levels was tested using receiver operating characteristic (ROC) analysis. The ROC curves were constructed to assess how well the CRS and BMI could detect individuals with HbA1c levels greater than or equal to 5.5%, 6.0%, and 7.0%. The area under the ROC curve (AUC)24 was used to compare the ability of the CRS and BMI to differentiate individuals at these HbA1c thresholds. The larger the AUC, the better the performance of the CRS or the BMI. We compared the sensitivity, specificity, and positive predictive value of cutoff points of the CRS (those published previously and those identified in our analysis) and BMI (those identified in our analysis). Associations between the individual CRS variables and HbA1c level were evaluated using multivariate logistic regression. All analyses were undertaken using Stata software, version 8.2 (StataCorp, College Station, Tex).
Sixty-five study participants (0.9%; 95% confidence interval [CI], 0.7%-1.1%) were identified as having HbA1c levels of 7.0% or more. The percentage of people who had HbA1c levels of 6.0% or more was 3.1% (95% CI, 2.7%-3.5%); 19.9% (95% CI, 19.0%-20.8%) of participants had HbA1c levels of 5.5% or more. Table 1 presents descriptive data for the 1958 Birth Cohort according to HbA1c threshold.
Figure 1 shows the ROC curves for the performance of the CRS and BMI against the 3 different HbA1c thresholds. Comparison of the AUC values for the 3 upper curves shows that the CRS was best able to differentiate individuals at the 7.0% threshold (0.84; 95% CI, 0.79-0.89). For detection of individuals with HbA1c levels of 6.0% or more or 5.5% or more, the AUCs were 0.76 (95% CI, 0.73-0.79) or 0.65 (95% CI, 0.63-0.66), respectively. The 3 lower curves show that the BMI performs similarly to the CRS, with better prediction at the higher HbA1c thresholds, with nearly identical AUC values to the corresponding CRS curve for each threshold. The CRS cutoff of 0.199 or more obtained from the original CRS study6 had optimal sensitivity (76.9%; 95% CI, 64.8%-86.5%) and specificity (77.8%; 95% CI, 76.9%-78.8%) for identification of individuals with HbA1c levels of 7.0% or more in our population (Table 2). The optimal CRS cutoff for detection of HbA1c levels of 6.0% or more was identified in our analysis as 0.128 or more, with a sensitivity of 78.2% (95% CI, 72.2%-83.3%) and specificity of 63.9% (95% CI, 62.7%-65.0%). For BMI, a cutoff of 30.00 or more was estimated to have the best sensitivity and specificity to differentiate individuals at the HbA1c thresholds of 6.0% and 7.0%. No cutoff with adequate sensitivity or specificity could be identified to differentiate above and below the HbA1c threshold of 5.5%. Figure 2 shows that people with HbA1c levels ranging from 5.5% to 6.0% have values of CRS and BMI across a wider range than is seen for those with HbA1c levels of more than 6.0%, thereby illustrating the poorer differentiation for individuals at the HbA1c threshold of 5.5%.
Using the optimal CRS cutoffs ascertained from ROC analyses, 37% of individuals would be considered to be at increased risk of hyperglycemia (CRS, ≥0.128) and 22.3% at increased risk of diabetes (CRS, ≥0.199). Similar to the results obtained by the CRS cutoff of 0.199 or more, 23.7% of the cohort was considered to be at risk of diabetes using the BMI of 30.0 or more. Although the CRS and BMI had good sensitivity and captured most individuals with elevated HbA1c levels at this age, only 3.0% of all individuals with a CRS of 0.199 or more and 2.6% of those with a BMI of 30.0 or more would have an HbA1c level of 7.0% or more (positive predictive value). When the HbA1c threshold is lowered to 6.0% or more, the proportion of individuals captured increases, thus the positive predictive value rises to 6.4% for the CRS (≥0.128) and 8.2% for the BMI (Table 2).
The populations detected by the CRS or the BMI with respect to the different HbA1c thresholds are shown in Figure 2. Forty-five of 65 people with HbA1c levels of 7.0% or more (69%) were detected by both measures, 5 (7.7%) by the CRS alone, and 6 (9.2%) by the BMI alone. For HbA1c levels of 6.0% or more, 66.8% were identified by the CRS and BMI, 11.0% by the CRS only, and 4.4% by the BMI only. Nonetheless, there may be some important differences between individuals identified by the CRS and BMI. For example, 16.4% of those at risk on the CRS were currently using antihypertensive medication compared with 10.0% of those at risk on the BMI.
Logistic regression models examine the associations between the individual CRS variables and the different HbA1c thresholds (Table 3). The odds ratio for BMI of 30.0 or more was 2.97 (95% CI, 2.54-3.48) for an HbA1c level of 5.5% or more (model 1) and increased to 10.37 (95% CI, 6.51-16.52) for an HbA1c level of 6.0% or more (model 2) and 17.09 (95% CI, 5.26-55.50) for an HbA1c level of 7.0% or more. As the HbA1c threshold increases, the prevalence of obesity increases, ie, 38.4%, 68.1%, and 76.9% for the 5.5%, 6.0%, and 7.0% HbA1c thresholds, respectively. The associations for antihypertensive medication, corticosteroids, and family history of diabetes also strengthen as the HbA1c threshold is raised, but the trend for smoking status is in the opposite direction.
Little is known about the prevalence of risk for type 2 diabetes in Great Britain. The CRS was developed to identify individuals at risk of type 2 diabetes; however, it has not been tested on a large, nationally representative population. To our knowledge, this is the first study to compare the CRS against an objective measure of glucose metabolism using a national sample. Until recently, data on type 2 diabetes in the 1958 British Birth Cohort have been limited to self-reported information with 0.8% reported cases of type 2 diabetes by 42 years of age.16 In this report, we found that a further 1% had HbA1c levels of 7.0% or more, which could reflect the prevalence of undiagnosed diabetes at 45 years of age, and 3% were hyperglycemic (HbA1c level, ≥6.0%). Results from this study confirm that the CRS identified individuals with elevated HbA1c levels in our population with good accuracy; however, BMI alone performed just as well as the CRS. The CRS and BMI indicated that 22.6% and 23.7% of the study sample, respectively, could be at increased risk of developing diabetes. Although the prevalence of elevated HbA1c levels was low, hence the low positive predictive values of the CRS or BMI, both measures indicated that approximately 20% of the population were at increased risk of development of diabetes. Thus, at this relatively young age, before most diabetes outcomes have occurred, the CRS or the BMI is a potentially useful measure to estimate population prevalence levels of risk for type 2 diabetes. Thereby, information on the prevalence of risk in adults who are of middle age indicates the potential future burden of type 2 diabetes.
Some limitations of our study are acknowledged. Attrition of sample size has occurred and, most relevantly for the purposes of our study, it has disproportionately affected those from unskilled manual backgrounds. Given that the unskilled manual classes are at greater diabetes risk than others, our study provides a conservative estimate of the true population at risk. As our main outcome, we used HbA1c level rather than an oral glucose tolerance test result, which was not available from the survey. The HbA1c level is used to monitor long-term (2-3 months) glucose control in patients with diabetes and is considered to be as good at predicting microvascular complications associated with type 2 diabetes as the oral glucose tolerance test.25 The risk of complications rises with increasing levels of HbA1c,26 and reductions in HbA1c levels lessen the occurrence of complications as shown in 2 randomized controlled trials.27,28 However, the use of HbA1c level as a screening test has not been recommended, mainly due to lack of standardization.29 Recent international standardization of HbA1c assays has facilitated the use of HbA1c levels in several diabetes screening studies.20,22,23,30-32 Our study is also limited by the lack of complete information on first-degree relatives with diabetes in the 1958 Birth Cohort. We have relied on information collected once when the study participants were 7 years of age, supplemented by whether either of their parents had a diabetes-related cause of death.17 Consequently, we may have underestimated the CRS, although it has been reported elsewhere that omitting family history does not have an appreciable impact on the score.6,7
In general, the CRS performs comparably in the 1958 Birth Cohort to other British studies. The AUC value of 84% for the CRS against the 7.0% HbA1c threshold is higher than that reported by the European Prospective Investigation of Cancer and Nutrition (74%),7 but similar to the findings from the original Cambridgeshire Study (80%).6 The CRS identified a slightly smaller proportion of individuals who may be at risk of development of diabetes in the 1958 Birth Cohort (22%) than in the Cambridgeshire Sample (30%). Outside Great Britain, the CRS identified 50% of the Horn population in the Netherlands.11 However, the specificity of the cutoff point (0.199) was lower (52%) in that population than in the 1958 Birth Cohort. A study from Germany also found poor sensitivity and specificity of the 0.199 CRS cutoff.32 Thus, risk scores for diabetes derived from one population may not be suitable for use in different populations because the extent to which risk is estimated will vary considerably. This is in common with other scores such as the Framingham Coronary Risk Score.33,34
The ability of the BMI to predict HbA1c level as well as the CRS suggests that the BMI is a particularly important component of the CRS in British adults in the mid-40s. Compared with results for the CRS in other populations, the odds ratio for the association between obesity (BMI, >30.0) and HbA1c level of 7.0% or more was 17.0, stronger than the associations reported from the European Prospective Investigation of Cancer and Nutrition (odds ratio, approximately 7.0)7 and from the Cambridgeshire Study (odds ratio, 12.4).6 In common with the European Prospective Investigation of Cancer and Nutrition but in contrast with the Cambridgeshire Study, we found no association between corticosteroid use and diabetes. The association for antihypertensive medication varied greatly for all 3 studies, although it was highest in our population, and smoking status had slightly weaker associations in the 1958 Birth Cohort using this simple classification of smoking status. These results suggest that the CRS may perform differently depending on the population characteristics and the distribution of risk factors in that population. The 3 populations described herein differ in a number of ways, including geographical location, the outcome measure used to define diabetes, and the average age of the study participants. The 1958 Birth Cohort is the youngest, with participants aged 45 years, approximately 10 to 15 years younger than the mean age of the other 2 British studies. Although our ROC and regression analyses suggest that the BMI is as good a predictor of diabetes risk as the CRS in midlife, the CRS may have additional benefits in detecting risk of cardiovascular disease.
Results for the 1958 Birth Cohort at 45 years of age indicate that the proportion of individuals with markedly elevated HbA1c levels who could be considered to be in the range of diabetes was low. The CRS identified those with HbA1c levels outside the reference range reasonably well and indicated that approximately 20% of the sample was at increased risk of development of type 2 diabetes. However, in this relatively young population, the CRS did not provide any additional advantage to the identification of diabetes risk than the BMI on its own.
Correspondence: Claudia Thomas, PhD, Centre for Paediatric Epidemiology and Biostatistics, Institute of Child Health, 30 Guilford St, London WC1N 1EH, England (email@example.com).
Accepted for Publication: August 15, 2005.
Author Contributions: The authors all had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Financial Disclosure: None.
Funding/Support: This study was supported by the Secretary of State for Health, Department of Health(NHS Research and Development Programme), London. Dr Hyppönen is a Department of Health Public Health Career Scientist. Data collection at 45 years of age was supported by grant G0000934 from the UK Medical Research Council, London. Research at the Institute of Child Health and Great Ormond Street Hospital for Children NHS Trust benefits from research and development funding received from the NHS Executive.
Disclaimer: The views and opinions expressed in the article represent those of the authors and do not necessarily reflect those of the Department of Health.
Previous Presentation: This study was presented as a poster at the First International Congress on Prediabetes and the Metabolic Syndrome; April 16, 2005; Berlin, Germany.
Acknowledgment: We thank the following data providers: Centre for Longitudinal Studies, London, England; Institute of Education and National Birthday Trust Fund, London; National Children's Bureau, London; and City University Social Statistics Research Unit, London (original data producers).
H Global prevalence of diabetes: estimates for the year 2000 and projections for 2030. Diabetes Care
2004;271047- 1053PubMedGoogle ScholarCrossref
MW Onset of NIDDM occurs at least 4-7 yr before clinical diagnosis. Diabetes Care
1992;15815- 819PubMedGoogle ScholarCrossref
SJ Should we screen for type 2 diabetes? evaluation against National Screening Committee criteria. BMJ
2001;322986- 988PubMedGoogle ScholarCrossref
NJ Diabetes risk score: towards earlier detection of type 2 diabetes in general practice. Diabetes Metab Res Rev
2000;16164- 171PubMedGoogle ScholarCrossref
NJ The performance of a risk score in predicting undiagnosed hyperglycemia. Diabetes Care
2002;25984- 988PubMedGoogle ScholarCrossref
et al. Glycated haemoglobin, diabetes, and mortality in men in Norfolk cohort of European Prospective Investigation of Cancer and Nutrition (EPIC-Norfolk). BMJ
2001;32215- 18PubMedGoogle ScholarCrossref
N Association of hemoglobin A1c with cardiovascular disease and mortality in adults: the European Prospective Investigation into Cancer in Norfolk. Ann Intern Med
2004;141413- 420PubMedGoogle ScholarCrossref
NJ What is the risk of mortality for people who are screen positive in a diabetes screening programme but who do not have diabetes on biochemical testing? diabetes programmes from a public health perspective. J Med Screen
2002;9187- 190PubMedGoogle ScholarCrossref
NJ The performance of a risk score as a screening test for undiagnosed hyperglycemia in ethnic minority groups. Diabetes Care
2004;27116- 122PubMedGoogle ScholarCrossref
DG Perinatal Mortality. Edinburgh, Scotland E&S Livingstone1963;
E Life at 33: The Fifth Follow-up of the National Child Development Study. London, England National Children's Bureau1993;
G Prenatal growth, BMI, and risk of type 2 diabetes by early midlife. Diabetes Care
2003;262512- 2517PubMedGoogle ScholarCrossref
C An intergenerational and lifecourse study of health and mortality risk in parents of the 1958 Birth Cohort, I: methods and tracing. Public Health
2005;119599- 607PubMedGoogle ScholarCrossref
F Multicenter evaluation of Tosoh glycohemoglobin analyzer. Clin Chem
1999;451833- 1841PubMedGoogle Scholar
V A clinical approach for the diagnosis of diabetes mellitus: an analysis using glycosylated hemoglobin levels: Meta-analysis Research Group on the Diagnosis of Diabetes Using Glycated Hemoglobin Levels. JAMA
1996;2761246- 1252PubMedGoogle ScholarCrossref
BL Relationship between fasting plasma glucose and glycosylated hemoglobin: potential for false-positive diagnoses of type 2 diabetes using new diagnostic criteria. JAMA
1999;2811203- 1210PubMedGoogle ScholarCrossref
measurement improves the detection of type 2 diabetes in high-risk individuals with nondiagnostic levels of fasting plasma glucose: the Early Diabetes Intervention Program (EDIP). Diabetes Care
2001;24465- 471PubMedGoogle ScholarCrossref
et al. Use of GHb (HbA1c
) in screening for undiagnosed diabetes in the US population. Diabetes Care
2000;23187- 191PubMedGoogle ScholarCrossref
JAC Medical Statistics. 2nd ed. Malden, Mass Blackwell Publishers2003;
et al. Comparison of tests for glycated haemoglobin and fasting and two hour plasma glucose concentrations as diagnostic methods for diabetes. BMJ
1994;3081323- 1328PubMedGoogle ScholarCrossref
GA Macrovascular risk and diagnostic criteria for type 2 diabetes: implications for the use of FPG and HbA1c
for cost-effective screening. Diabetes Care
2003;26485- 490PubMedGoogle ScholarCrossref
The Diabetes Control and Complications Trial Research Group, The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med
1993;329977- 986PubMedGoogle ScholarCrossref
UK Prospective Diabetes Study (UKPDS) Group, Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes. Lancet
1998;352837- 853PubMedGoogle ScholarCrossref
DE Tests of glycemia for the diagnosis of type 2 diabetes mellitus. Ann Intern Med
2002;137263- 272PubMedGoogle ScholarCrossref
et al. Screening for diabetes in an outpatient clinic population. J Gen Intern Med
2002;1723- 28PubMedGoogle ScholarCrossref
EZ Utility of hemoglobin A1c
in predicting diabetes risk. J Gen Intern Med
2004;191175- 1180PubMedGoogle ScholarCrossref
et al. Performance of screening questionnaires and risk scores for undiagnosed diabetes. Arch Intern Med
2005;165436- 441PubMedGoogle ScholarCrossref
et al. Predictive accuracy of the Framingham Coronary Risk Score in British men: prospective cohort study. BMJ
2003;3271267- 1270PubMedGoogle ScholarCrossref
T 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;31817- 822PubMedGoogle ScholarCrossref