Coronary risk prediction for those with and without diabetes

Abstract

Background

Coronary risk prediction ‘engines’ are now in common use, and their worth is well proven. There remains the question of how to deal with a prior diagnosis of diabetes.

Design

An individual participant meta-analysis of 33 cohort studies involving 364 566 subjects.

Methods

Fatal coronary hazard ratios for age, smoking, systolic blood pressure and cholesterol, were computed from Cox models, comparing those with and without diabetes. Three risk prediction equations were compared: a ‘stepped model', which included the risk factors and diabetes status; an ‘interaction model', which included interactions between diabetes and the risk factors; and a ‘fixed model', which fixed the 10-year rate of coronary death amongst those with diabetes to be 7%. These were compared through the area under the receiver operating characteristic curve (AUC) and Hosmer-Lemeshow statistics.

Results

The hazard ratio for age was greater for those without diabetes than those with, for men (P=0.005) and women (P=0.02); for men only, systolic blood pressure showed a similar differential (P=0.011). Nevertheless, AUCs were only 0.001 different for the stepped and interaction models for each sex. The AUC for the fixed model was lower and, unlike the other two, showed significant lack of fit for both sexes (P< 0.001).

Conclusions

There is no justification for developing separate risk prediction models for those with and without diabetes, nor for assuming that everyone with diabetes should be considered as being at a common high level of risk. Diabetes status might, instead, be used as a risk variable in an overall population equation.

Keywords

coronary heart disease diabetes risk blood pressure cholesterol smoking

Introduction

Risk prediction, often called ‘absolute risk’ prediction [1], is now a well-established tool for decision making in preventive cardiology. The Framingham risk prediction equations for coronary heart disease (CHD) (and sometimes other cardiovascular endpoints) have been formulated [2], reformulated [3], simplified through point scoring [4] and charts [5], recalibrated to other populations [6] and replaced by similar prediction engines developed in local populations [7,8].

Most of the coronary risk prediction tools in current use take account of a prior diagnosis of diabetes. One approach, here called the ‘stepped’ method, is to include a term for diabetes in the risk prediction algorithm [2], so that those with diabetes have a different (always higher) score than those without, when all other risk factors are kept at a constant level. Where risk points are given for each risk factor, and later summed, the total for someone with diabetes is always greater by the same amount as someone without diabetes but with identical values of other risk factors, regardless of the levels of these other risk factors (e.g. blood pressure) [3]. Another approach, the ‘interaction’ method, computes separate risk predictions for those with and without diabetes [9,10]. This method treats those with and without diabetes as totally independent, such that the information on how other risk factors relate to coronary risk in one group is not used to inform the risk prediction for the other group. This approach has limited statistical power for the diabetes group, as it will typically be small in general populations. Some studies have addressed this shortfall by using data from patients with diabetes [11–13]. This approach is only able to derive, and test, algorithms for those with diabetes, however, and cannot directly compare risk predictions for those with and without the condition. Finally, a third approach, the ‘fixed’ method, is that taken, for example, in two national treatment guidelines in the USA [14,15]: pre-existing diabetes is taken as risk-equivalent to pre-existing heart disease, thus implying that treatment to prevent CHD should be considered for all those with diabetes.

This paper uses the data from one of the largest overviews (meta-analyses) of cohort studies to compare the three approaches to coronary risk prediction, taking account of diabetes status and the other major cardiovascular risk factors. The best of these approaches is used to formulate a single coronary risk prediction model applicable to those with and without diabetes.

Methods

The Asia Pacific Cohort Studies Collaboration (APCSC) is an overview of cohort studies carried out in the region which satisfy certain inclusion criteria (age and blood pressure recorded at baseline; vital status recorded at the end of follow-up; total follow-up of at least 5000 person-years) and exclusion criteria (entry not dependent upon having a particular condition or risk factor) and whose investigators were willing to provide individual participant data [16]. Coronary deaths were all deaths with the underlying cause coded according to the ninth revision of the International Classification of Diseases as 410-414. Analyses were restricted to individuals aged 20 years or more at the time of the baseline survey for whom diabetes status was recorded, based on an individual's self-report of a previous diagnosis of diabetes. Additionally, for studies that measured glucose, diabetes status was positive if fasting glucose was greater than 6.1 mmol/l for serum samples, and greater than 7 mmol/l for plasma samples, or if non-fasting glucose was greater than 10 mmol/l for serum samples, and greater than 11.1 mmol/l for plasma samples [17].

Survival analyses

Separately for men and women, Cox survival models, stratified by study [18], were used to estimate hazard ratios for coronary death according to age, systolic blood pressure (SBP), total cholesterol (TC) and smoking (current smoker/not), for those with and without diabetes. These variables were chosen since they appear in all the current cardiovascular or coronary risk prediction algorithms. Age, SBP and TC were taken as continuous variables, so as to increase power and since previous analyses have shown these variables to have a log-linear relationship with coronary outcomes in APCSC [19–21]. Results for SBP and TC are reported after adjustment for regression dilution [18], whilst all hazard ratios were adjusted for age (except where age is the variable of interest). The homogeneity of the effect of each risk factor across diabetes status was tested by adding an interaction term to the Cox model.

Risk prediction

Most risk prediction models look at 10-year risk. Few studies in APCSC, however, have follow-ups of this length. Thus, in order to maximize study inclusion, but yet provide a reasonable length of follow-up over which to estimate risk, 8-year predictions were made. Thus, individuals were right-censored to 8 years and, to avoid bias due to a large number of participants having dropped out, analyses were restricted to studies with a 75th percentile of follow-up equal to at least 8 years. Participants with missing information on any of the risk factors of interest were excluded.

Three logistic regression models were fitted, for each sex. The ‘stepped model’ included age, age squared, SBP, TC, smoking, diabetes and study. Age squared was included for completeness, since it features in the Framingham risk prediction models [2–4]. For both men and women, a stepped risk prediction model for coronary death was formed using the beta coefficients (log odds ratios, not corrected for regression dilution) from the stepped logistic model [4]. The discrimination of this prediction model was evaluated using receiver operating characteristic (ROC) curves and calibration with observed outcomes was evaluated using Hosmer-Lemeshow tests [22].

The ‘interaction model’ added interactions between diabetes and age, SBP, TC and smoking for each sex, unless lack of numbers suggested that estimation of any of the interactions was unreliable. An interaction risk prediction model was formed, for each sex, from the beta coefficients derived from this model. Finally, a ‘fixed model’ was formed, for each sex. For those with diabetes, this model assumes a fixed rate of death from coronary disease at a level that would normally lead to treatment. US guidelines for lipid-lowering therapy [14] define those at highest risk as having an estimated 10-year probability of fatal or non-fatal coronary heart disease of greater than 20%. Assuming a ratio of fatal-to-nonfatal coronary events of approximately 1 : 2, this suggests that those with a 10-year rate of fatal CHD of about 7% should be treated. Thus, the fixed model assumes a probability of 0.7% per annum for coronary death for those with diabetes. In sensitivity analyses, this percentage was varied from 0.1 to 2.1%. For those without diabetes the probabilities found from the stepped model were used. The interaction and fixed models were evaluated as for the stepped model, with which each was compared. The percentage of overall bias in predicted CHD rates with the fixed model was calculated as 100 times (observed rate minus predicted rate) divided by the observed rate, for men and women separately.

Results

From a total study size of just under 600 000 in APCSC, 364 566 (61%) individuals had age, SBP and diabetes status or fasting blood glucose recorded at baseline (Table 1). Overall 6% of participants had diabetes and 41% were female. Over an average of 4 years follow-up, 2265 (0.6%) died from coronary heart disease. Amongst men, there were 198 coronary deaths amongst those with diabetes (1.1%) and 1379 (0.7%) amongst those without. Corresponding results for women were 67 (1.2%) and 621 (0.4%).

In both sexes, there were significant differences (P≤0.001) between those with and without diabetes for all the other risk factors considered (Table 2). In every case, except smoking amongst women (where prevalence was low), those with diabetes had higher values.

The age-adjusted hazard ratio for coronary death, comparing those with diabetes and those without was 1.83 (1.57-2.14) in men and 2.25 (1.73-2.93) in women. Table 3 shows the hazard ratios for other risk factors, both for those with and without diabetes. Results are shown separately by sex, except that there were too few deaths amongst women who both smoked and had diabetes to allow reliable estimation. For men, the effects of age and SBP were significantly (P < 0.05) stronger amongst those without, compared with those with, diabetes. For women, a similarly significant difference by age was observed, but SBP was not significantly different by diabetes status. The effects of TC were not significantly different between those with and without diabetes for either sex; the same was true for smoking amongst men.

Figure 1 compares risk prediction for the three models considered, for each sex. In these analyses data from 43 079 men (436 CHD deaths overall, of which 50 were for those with diabetes) and 48 729 women (182 CHD deaths; 22 with diabetes) were used. Since no reliable estimation was available, the interaction effect for smoking was left out of the interaction model for women. Despite the finding of some significant interactions in Table 3, the stepped and interaction models gave very similar results for both sexes. AUCs were 0.83, to two decimal places, for men and 0.89 for women. Hosmer-Lemeshow tests showed no significant lack of fit, comparing observed and predicted death rates, there being slight underestimation amongst those in the highest decile of predicted risk. The fixed model discriminated between those who did and did not die from CHD almost as well, with an AUC of 0.82 for men and 0.88 for women. It did not, however, calibrate as well with observed data compared with the other models. There was significant (P < 0.001) lack of fit for both sexes, with considerable overestimation of risk amongst those with the highest 10% of predicted risk. For each sex, when percentages other than 0.7% for 10-year risk of fatal CHD for those with diabetes were used in the fixed model, the AUC never approached that of the stepped model.

Table 1

Baseline summary statistics and details of studies

Study	Country	Baseline	Number of subjects	Mean age (years)	Median follow-up (years)	Female (%)	Diabetes (%)	CHD deaths
Aito Town^a	Japan	1980-83	1717	51	15	57	2.7	16
Akabane^a	Japan	1985-86	1836	54	11	56	2.5	7
Anzhen 02	China	1992-93	4142	47	3	51	11.2	1
Busselton^a	Australia	1966-81	6000	46	19	52	3.5	556
CISCH	China	1992-93	2162	44	3	51	2.4	1
Civil service workers	Japan	1990-92	9319	47	7	33	1.7	1
CVDFACTS	Taiwan	1988-96	5730	47	6	55	2.7	13
East Beijing	China	1977-94	1128	44	17	51	5.6	20
EGAT^a	Thailand	1985	3497	43	11	23	1.5	33
Fletcher challenge	NZ	1992-94	10 366	44	6	28	2.6	114
Guangzhou occupational^a	China	1985-95	5798	44	8	34	10.5	5
Hong Kong	Hong Kong	1985-91	2984	79	3	57	8.5	85
Huashan	China	1990-92	1868	53	3	52	13.7	3
Kinmen	Taiwan	1993-97	2535	63	3	49	8.6	10
KMIC	South Korea	1992	183 583	44	4	37	7. 7	114
Konan	Japan	1987-95	1226	52	6	55	12.6	2
Longitudinal study of aging	Australia	1992-93	1613	78	5	48	8.2	81
Melbourne^a	Australia	1990-94	41 286	55	8	59	5.4	323
Miyama	Japan	1988-90	1073	61	7	56	5.2	2
National Heart Foundation^a	Australia	1989-90	9276	43	8	51	1.9	77
Newcastle	Australia	1988-94	3462	53	5	50	3.5	42
Ohasama	Japan	1992-93	2240	60	4	64	10.9	7
Perth^a	Australia	1978-94	10 222	45	14	48	2.1	195
Saitama^a	Japan	1986-90	3624	55	11	62	1.7	24
Seven cities cohorts	China	1987	10734	54	3	54	1.2	83
Shibata^a	Japan	1977	2350	57	20	58	1.1	67
Shigaraki town	Japan	1991-97	3758	57	4	59	7. 2	3
Shirakawa^a	Japan	1974-79	4643	48	18	54	0.9	45
Singapore Heart^a	Singapore	1982-97	2325	41	15	49	11.4	31
Singapore NHS92	Singapore	1992	3305	39	6	52	9.7	22
Tanno/Soubetsu^a	Japan	1977	1980	51	16	53	7. 2	24
WA AAA Screenees	Australia	1996-99	12 203	72	3	0	11.6	240
Yunnan	China	1992	6581	56	4	3	0.5	18
	Total	1966-99	364 566	48	4	41	6.3	2265

Studies included in the risk prediction models. CHD, Coronary heart disease; CISCH, Capital Iron and Steel Hospital; EGAT, Electricity Generating Authority of Thailand; KMIC, Korean Medical Insurance Corporation; NHS, National Health Survey; WA AAA, Western Australian Abdominal Aortic Aneurysm.

Table 2

Mean and 95% confidence intervals for major risk factors by diabetes status, adjusted for age

	Age (years)		SBP (mmHg)		TC (mmol/l)		Smoking (%)
	n	Mean (CI)	n	Mean (CI)	n	Mean (CI)	n	Mean (CI)
Men
Without diabetes	196787	48.6 (48.5-48.6)	196787	129.1 (129.1-129.2)	174504	5.11 (5.10-5.11)	188712	45.6 (45.3-45.9)
With diabetes	17 559	51.8 (51.6-51.9)	17559	133.6 (133.4-133.9)	15633	5.28 (5.26-5.29)	16292	46.4 (45.2-47.6)
Women
Without diabetes	144 803	47.1 (47.1-47.2)	144 803	123.4 (123.3-123.5)	136 336	5.14 (5.14-5.15)	130517	5.86 (5.73-5.98)
With diabetes	5417	53.1 (52.8-53.4)	5417	129.5 (129.0-129.9)	4887	5.39 (5.36-5.41)	4802	4.13 (3.52-4.74)

P values for differences between people with and without diabetes all < 0.001. SBP, systolic blood pressure; TC, total cholesterol; CI, confidence interval.

Fig. 1

Results of three risk prediction models when compared with observed data, by sex. Plots show observed (filled bars) and expected (open bars) coronary heart disease (CHD) death probabilities by decile of expected probability. In the fixed model the rate of CHD death for those with diabetes was assumed to be 0.7% per annum; those without diabetes received probabilities computed from the stepped model. H-L X ², Hosmer-Lemeshow X ² statistic; AUC, area under operating characteristic curve; CI, confidence interval.

Since the stepped model is superior to the fixed model, and is simpler while performing equally well as the interaction model, this is the model of choice. Table 4 shows the beta parameters for this model; these parameters form the basis of the risk scoring system for this approach to coronary risk prediction [4].

Discussion

Primarily using data from the Framingham study [2], current treatment guidelines worldwide mostly approach coronary or cardiovascular disease risk prediction using the interaction method (e.g. New Zealand Guidelines for Assessment and Management of Cardiovascular Risk [23]), or the fixed method (e.g. British Hypertension Society [24], US National Cholesterol Education Program - Adult Treatment Panel III [14]). Others, such as the Joint European Societies [25], use a cardiovascular risk prediction algorithm that omits diabetes status as a predictor variable.

The current analyses, based on several thousands of people and several hundreds of events, suggest that coronary heart disease prediction tools should account for diabetes simply by including a term for diabetes in the risk score, as assumed in the stepped model. That is, someone with diabetes gets a constant added to his or her risk score compared with someone without diabetes but with the same levels of all the other risk factors. The more complex interaction model gives no additional benefit, suggesting that it is not necessary to develop separate prediction tools for those with, and without, diabetes. Treatment algorithms that assume all those with diabetes should be treated to prevent coronary events predict coronary death rather less well, as shown by the results for the fixed model.

Two restrictions apply to the results. First, the definition of diabetes used is inaccurate. It may be that a more precise definition, such as one based on a glucose tolerance test, would produce different results, but our data do not allow this to be tested. Sensitivity analyses restricted to self-reported diabetes (not using information on blood glucose) produced essentially the same results as shown in Table 3 here (results not shown). Second, the basic finding could alter if further risk factors, such as blood triglycerides and high-density lipoprotein cholesterol levels, physical exercise, glycosylated haemoglobin and albuminuria, were introduced into the model [26]. Some of the studies in APCSC measured triglycerides, which some have suggested should appear in risk prediction models for diabetes [27]. Although there were too few data to carry out extensive analyses, triglycerides was a significant risk factor after adjustment for all the variables in the stepped model for men, but not women (results not shown). Furthermore, the diabetes-by-triglyceride interaction was only marginally non-significant for men (P = 0.076), but, again, there was no evidence of an interaction for women. There were only 54 deaths, however, from CHD in the subgroup with data on triglycerides for women compared with 165 for men. The AUC went up slightly, for men, when triglycerides was added to both the stepped (0.878 to 0.881) and interaction (0.877 to 0.883) models. No change appeared for women. Thus triglycerides may, indeed, be useful in risk prediction models, as previous APCSC analyses have suggested [28]. The similarities in the AUCs in the stepped and interaction models with triglycerides suggests, however, that even when triglycerides are included, there is no reason to suggest that a separate algorithm is required for those with and without diabetes.

Table 3

Hazard ratio and 95% confidence intervals for coronary heart disease death for major risk factors by diabetes status, adjusted for age

	Age + 10 years		SBP+10mmHg		TC + 1 mmol/l		Current smoking vs not
	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Men
Without diabetes	2.94	2.76-3.11	1.17	1.13-1.23	1.40	1.28-1.53	1.50	1.33-1.69
With diabetes	2.41	2.10-2.76	1.02	0.90-1.13	1.38	1.10-1.74	1.88	1.39-2.56
P homog		0.005		0.011		0.57		0.16
Women
Without diabetes	3.61	3.33-3.94	1.27	1.20-1.33	1.34	1.20-1.50	1.81	1.48-2.23
With diabetes	2.79	2.24-3.49	1.33	1.15-1.52	1.11	0.76-1.60		NA
P homog		0.02		0.50		0.35		NA

SBP, systolic blood pressure; TC, total cholesterol; HR, hazard ratio; CI confidence interval; NA, not available, only two events were recorded for women currently smoking; P homog, P value for interaction between the risk factor and diabetes status. Results for SBP and TC are adjusted for regression dilution bias.

Table 4

Beta coefficients and their standard errors for the stepped model

Variable	Men		Women
	Beta	SE	Beta	SE
Age	0.2145	0.0492	0.2779	0.0876
Age²	-0.00103	0.00041	-0.00127	0.000676
SBP (mmHg)	0.0125	0.00225	0.0178	0.00291
TC (mmol/l)	0.2111	0.0439	0.1306	0.0620
Current smoking	0.5415	0.1112	0.9126	0.1947
Diabetes status	0.4617	0.1578	0.8206	0.2418

SE, standard error; SBP, systolic blood pressure; TC, total cholesterol.

It has been previously suggested that the risk of myocardial infarction among those with diabetes and no known coronary disease is equivalent to that of non-diabetic individuals with a previous history of myocardial infarction [29]. Other data indicate a range of coronary risk associated with diabetes, not necessarily always similar to that of established coronary heart disease [30–35]. The current analyses are consistent with this conclusion. Some have argued that a prior diagnosis of diabetes should be regarded as equivalent to a prior diagnosis of CHD in coronary risk prediction, because of the high case fatality associated with diabetes [36]. The current data, however, which evaluate prediction of fatal coronary events, do not support this contention. Instead, it seems that the most appropriate way to treat diabetes status in coronary risk prediction is to include it as an independent variable in the risk equation.

Footnotes

Appendix

Writing committee: M. Woodward, F. Barzi, T.H. Lam, K. Jamrozik, H. Ueshima, A. Patel.

Executive committee: D.F. Gu, T.H. Lam, C.M.M. Lawes, S.W. MacMahon, W.H. Pan, A. Rodgers, I. Suh, H. Ueshima, M. Woodward.

Participating studies and principal collaborators: Aito Town: A. Okayama, H. Ueshima; H. Maegawa; Akabane ^∗: N. Aoki, M. Nakamura, N. Kubo, T. Yamada; Anzhen02 ^∗: Z.S. Wu; Anzhen: C.H. Yao, Z.S. Wu; Beijing Aging: Z. Tang; Beijing Steelworkers: L.S. Liu, J.X. Xie; Blood Donors’ Health: R. Norton, S. Ameratunga, S. MacMahon, G. Whitlock; Busselton ^∗: M.W. Knuiman; Canberra-Queanbeyan: H. Christensen; Capital Iron and Steel Company (CISC) ^∗: X.G. Wu; CISCH: J. Zhou, X.H. Yu; Civil Service Workers: A. Tamakoshi; CVDFACTS ^∗: W.H. Pan; East Beijing ^∗: Z.L. Wu, L.Q. Chen, G.L. Shan; EGAT ^∗: P. Sritara; Fangshan ^∗: D.F. Gu, X.F. Duan; Fletcher Challenge ^∗: S. MacMahon, R. Norton, G. Whitlock, R. Jackson; Guangzhou: Y.H. Li; Guangzhou Occupational ^∗: T.H. Lam, C.Q. Jiang; Hisayama: M. Fujishima, Y. Kiyohara, H. Iwamoto; Hong Kong ^∗: J. Woo, S.C. Ho; Huashan ^∗: Z. Hong, M.S. Huang, B. Zhou; Kinmen: J.L. Fuh; KMIC ^∗: I. Suh, S.H. Jee, I.S. Kim; Konan ^∗: H. Ueshima, Y. Kita, S.R. Choudhury; Longitudinal Study of Aging ^∗: G Andrews; Melbourne ^∗: G. Giles; Miyama ^∗: T Hashimoto, K Sakata; National Heart Foundation ^∗: T.A. Welborn; Newcastle ^∗: A. Dobson; Ohasama ^∗: Y. Imai, T. Ohkubo, A. Hozawa; Perth ^∗: K. Jamrozik, M. Hobbs; Saitama ^∗: K. Nakachi; Seven Cities ^∗: X.H. Fang, S.C. Li, Q.D. Yang; Shanghai Factory Workers: Z.M. Chen; Shibata ^∗: H. Tanaka; Shigaraki Town ^∗: Y. Kita, A. Nozaki, H. Ueshima; Shirakawa ^∗: H. Horibe, Y. Matsutani, M. Kagaya; Singapore Heart: K. Hughes, J. Lee; Singapore NHS92 ^∗: D. Heng, S.K. Chew; Six Cohorts ^∗: B.F. Zhou, H.Y. Zhang; Tanno/Soubetsu ^∗: K. Shimamoto, S. Saitoh; Tianjin: Z.Z. Li, H.Y. Zhang; Western Australia AAA Screenees ^∗: P. Norman, K. Jamrozik; Xi'an: Y. He, T.H. Lam; Yunnan ^∗: S.X. Yao.

^∗Studies providing data used in this paper.

References

1 Tunstall-Pedoe

“Absolute” is inappropriate for quantitative risk estimation. BMJ 2000; 320:723.

Anderson

Wilson

PWF

Odell

Kannel

WB.

An updated coronary risk profile. A statement for health professionals. Circulation 1991; 83:356–362.

Wilson

PWF

D'Agostino

Levy

Belanger

Silbershatz

Kannel

WB.

Prediction of coronary heart disease using risk factor categories. Circulation 1998; 97:1837–1847.

Sullivan

Massaro

D'Agostino

RB.

Presentation of multivariate data for clinical use: The Framingham Study risk score. Statist Med 2004; 23:1631–1660.

Wood

Durrington

Poulter

McInnes

Rees

Wray

On behalf of the British Cardiac Society, British Hyperlipidaemia Association, British Hypertension Society, and British Diabetic Association. Joint British recommendations on prevention of coronary heart disease in clinical practice. Heart 1999; 80(Suppl 2):S1–S29.

6 The Diverse Populations Collaboration. Prediction of mortality from coronary heart disease among diverse populations. Is there a common predictive function? Heart 2002; 88:222–228.

Assmann

Cullen

Schulte

The Munster Heart Study (PROCAM). Results of follow-up at 8 years. Eur Heart J 1998; 19(Suppl A):A2–A11.

8 Conroy

Pyorala

Fitzgerald

Sans

Menotti

De Backer

, et al. Estimation of ten year risk of fatal cardiovascular disease in Europe: the SCORE project. Eur Heart J 2003; 24:987–1003.

Song

Brown

PM.

Coronary heart disease risk assessment in diabetes mellitus: comparison of UKPDS risk engine with Framingham risk assessment function and its clinical implications. Diabet Med 2004; 21:238–245.

10.

Folsom

Chambless

Duncan

Gilbert

Pankrow

JS.

Prediction of coronary heart disease in middle-aged adults with diabetes. Diabetes Care 2003; 26:2777–2784.

11.

Turner

Millns

Neil

HAW

Stratton

Manley

Matthews

Holman

RR.

Risk factors for coronary artery disease in non-insulin dependent diabetes mellitus: United Kingdom prospective diabetes study (UKPDS). BMJ 1998; 316:823–828.

12.

Stephens

Ambler

Vallance

Betteridge

Humphries

Hurel

SJ.

Cardiovascular risk and diabetes. Are the methods of risk prediction satisfactory? Eur J Cardiovasc Prev Rehabil 2004; 11:521–528.

13.

Game

Bartlett

Bayly

Jones

AF.

Comparative accuracy of cardiovascular risk prediction methods in patients with diabetes mellitus. Diabetes Obes Metab 2001; 3:279–286.

14.

14 Expert Panel on Detection, Evaluation and Treatment of High Blood Cholesterol in Adults. 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; 285:2486–2497.

15.

15 Chobanian

Bakris

Black

Cushman

Green

Izzo

Jr , et al. The seventh report of the Joint National Committeee on prevention, detection, evaluation and treatment of high blood pressure: the JNC 7 report. JAMA 2003; 289:2560–2572.

16.

16 Asia Pacific Cohort Studies Collaboration. Determinants of cardiovascular disease in the Asia Pacific region: protocol for a collaborative overview of cohort studies. CVD Prevent 1999; 2:281–289.

17.

17 Asia Pacific Cohort Studies Collaboration. The effects of diabetes on the risks of major cardiovascular diseases and death in the Asia-Pacific region. Diabetes Care 2003; 26:360–366.

18.

Woodward

Epidemiology: study design and data analysis. 2nd ed. Boca Raton: Chapman and Hall/CRC Press; 2005.

19.

19 Asia Pacific Cohort Studies Collaboration. Blood pressure and cardiovascular disease in the Asia Pacific region. J Hypertens 2003; 21:707–716.

20.

20 Asia Pacific Cohort Studies Collaboration. Cholesterol, coronary heart disease and stroke in the Asia Pacific region. Int J Epidemiol 2003; 32:563–572.

21.

21 Asia Pacific Cohort Studies Collaboration. A comparison of the associations between risk factors and cardiovascular disease in Asia and Australasia. Eur J Cardiovasc Prev Rehabil 2005; 12:484–491.

22.

Hosmer

Lemeshow

Applied logistic regression. 2nd ed. New York: John Wiley; 2000.

23.

23 New Zealand Guidelines Group. Assessment and management of cardiovascular risk [online resource]. www.nzgg.org.nz. [Accessed 31 May 2005].

24.

24 Williams

Poulter

Brown

Davis

McInnes

Potter

, et al. Guidelines for management of hypertension: report of the fourth working party of the British Hypertension Society, 2004: BHS IV. J Human Hypertens 2004; 18:139–185.

25.

25 European Society of Cardiology, European Atherosclerosis Society, European Society of Hypertension, International Society of Behavioural Medicine, European Society of General Practice/Family Medicine, European Heart Network. Prevention of coronary heart disease in clinical practice. Recommendations of the Second Joint Task Force of European and other Societies on coronary prevention. Eur Heart J 1998; 19:1434–1503.

26.

Wincour

Fisher

Prediction of cardiovascular risk in people with diabetes. Diabet Med 2003; 20:515–527.

27.

Game

Jones

AF.

Coronary heart disease risk assessment in diabetes mellitus: a comparison of PROCAM and Framingham risk assessment functions. Diabet Med 2001; 18:355–359.

28.

28 Asia Pacific Cohort Studies Collaboration. Serum triglycerides as a risk factor for cardiovascular diseases in the Asia Pacific region. Circulation 2004; 110:2678–2686.

29.

Haffner

Lehto

Ronnemaa

Pyorala

Laakso

Mortality from coronary heart disease in subjects with type 2 diabetes and in nondiabetic subjects with and without prior myocardial infarction. N Engl J Med 1998; 339:229–234.

30.

30 Simons

Simons

Engler

McGurk

Harper

McCance

, et al. Diabetes and coronary heart disease. N Engl J Med 1998; 339:1714–1716.

31.

31 Becker

Bos

de Vegt

Kostense

Dekker

Nijpels

, et al. Cardiovascular events in type 2 diabetes: comparison with nondiabetic individuals without and with prior cardiovascular disease: 10-year follow-up of the Hoorn Study. Eur Heart J 2003; 24:1406–1413.

32.

Wannamethee

Shaper

Lennon

Cardiovascular disease incidence and mortality in older men with diabetes and in men with coronary heart disease. Heart 2004; 90:1398–1403.

33.

Lee

Folsom

Pankow

Brancati

FL.

For the Atherosclerosis Risk in Communities (ARIC) Study Investigators. Cardiovascular events in diabetic and nondiabetic adults with or without history of myocardial infarction. Circulation 2004; 109:855–860.

34.

Evans

JMM

Wang

Morris

AD.

Comparison of cardiovascular risk between patients with type 2 diabetes and those who had had a myocardial infarction: cross sectional and cohort studies. BMJ 2002; 324:939.

35.

Whiteley

Padmanabhan

Hole

Isles

Should diabetes be considered a coronary heart disease risk equivalent? Results from 25 years of follow-up in the Renfrew and Paisley study. Diabetes Care 2005; 28:1588–1593.

36.

Davis

TME

Parsons

Broadhurst

Hobbs

MST

Jamrozik

Arrhythmias and mortality after myocardial infarction in diabetic patients. Diabetes Care 1998; 21:637–640.