Predicted and observed cardiovascular disease in South Asians: application of FINRISK, Framingham and SCORE models to Newcastle Heart Project data

Transcription

1 Journal of Public Health VoI. 27, No. 1, pp doi: /pubmed/fdh202 Predicted and observed cardiovascular disease in South Asians: application of FINRISK, and SCORE models to Newcastle Heart Project data Raj Bhopal, Colin Fischbacher, Erkki Vartiainen, Nigel Unwin, Martin White and George Alberti Abstract Background South Asian populations in the United Kingdom have a high risk of cardiovascular disease (CVD) mortality. Risk prediction models appear to be inaccurate in South Asians. Objective To explore the predictive capacity of the FINRISK, (1991) and SCORE risk prediction models in the Newcastle Heart Project population (n = 1301). Methods Mortality data for England and Wales were used to define the expected ranking of CVD risk by country of birth. CVD mortality in the Newcastle Heart Project sample was examined. Risk factor measures were obtained from the Newcastle Heart Project, where 90 percent of South Asians were born in the Indian Subcontinent. The predicted outcomes for FINRISK were acute myocardial infarction and CHD mortality, for CHD mortality, myocardial infarction, new angina and coronary insufficiency and for SCORE CHD and non-chd CVD mortality. Results The FINRISK model predicted in South Asian men combined, compared with Europeans, a risk ratio of 122 per cent (SMR 142) with substantial subgroup heterogeneity, e.g. 154 per cent in Bangladeshis (SMR 151), 129 per cent in Pakistanis (SMR 148), 99 per cent in Indians (SMR 142). The FINRISK risk ratios for South Asian women combined were 160 per cent (SMR 145), for Bangladeshis 184 per cent (SMR 91), Pakistanis 172 per cent (SMR 111) and for Indians 145 per cent (SMR 158). The model results were very similar to FINRISK, but the SCORE model showed comparatively low 10 year risk in all South Asian groups. Both the stroke model and the SCORE non-chd CVD model predicted comparatively low rates, while national data showed these to be high. Control of the five major risk factors was modelled by FINRISK to reduce risk by about 59 per cent in South Asian men and 67 per cent in South Asian women, with some subgroup heterogeneity, compared to 50 per cent in European men and 48 per cent in European women. The model results were similar. The absolute rates for each ethnic group varied by model. Conclusion The and FINRISK models gave similar results, mostly following expected patterns, but the SCORE model did not, probably reflecting its lack of inclusion of HDL and diabetes as risk factors. National mortality data and modelled predictions agreed reasonably well for South Asians combined, and Bangladeshi and Pakistani men, but not for Indian men and Pakistani and Bangladeshi women. The varying rates show the limits of modelling. The models suggest the potential gains from controlling major established risk factors could be substantial in South Asians and greater than in Europeans. Introduction British South Asians, defined here as people of Indian, Pakistani and Bangladeshi ancestry, are widely acknowledged as having per cent higher risk of both coronary heart disease and stroke mortality, compared to the whole population. 1 3 Morbidity data are sparse and less clear cut in showing excess risks. 3 The excess mortality risk cannot be fully explained by the pattern of classical risk factors, 4,5 though when a wider range of risk factors is included Bangladeshi and Pakistani populations, in Section of Public Health Sciences, College of Medicine and Veterinary Medicine, Edinburgh EH8 9AG Raj Bhopal, Professor of Public Health Section of Public Health Sciences, Medical School, Teviot Place, Edinburgh EH8 9AG Colin Fischbacher, Clinical Research Fellow Department of Epidemiology and Health Promotion, National Public Health Institute, Finland and Medical School, Edinburgh Erkki Vartiainen, Professor and Director School of Clinical Medical Sciences (Diabetes) and School of Population and Health Sciences, Medical School, Framlington Place, Newcastle NE2 4HH Nigel Unwin, Senior Lecturer Public Health Research Group, School of Population and Health Sciences, Medical School, Framlington Place, Newcastle NE2 4HH Martin White, Senior Lecturer School of Clinical Medical Sciences, Medical School, Framlington Place, Newcastle NE2 4HH George Alberti, Emeritus Professor Address correspondence to Raj Bhopal. Raj.Bhopal@ed.ac.uk The Author 2005, Published by Oxford University Press on behalf of Faculty of Public Health. All rights reserved.

2 94 JOURNAL OF PUBLIC HEALTH particular, are worse off than Indians and white Europeans. 6 There are no published cohort studies supplying robust information on risk factor outcome relations in South Asians, or to test the validity of cardiovascular risk predictions. The few published prediction studies in the United Kingdom are summarized in Table 1. Game and Jones 7 reported that in a clinic population of people with diabetes the risk equation predicted no difference between South Asian and white Europeans in risk of mortality. Kooner and Chambers 8 found the, Dundee and Procam cardiovascular risk prediction models did not predict an excess risk in South Asian men in West London. Cappuccio et al. 5 found in South London that the equation predicted absolute risks of CHD and stroke close to the values in Europeans. Quirke et al. 9 applied the equation to data from the Health Survey for England 1998 and 1999 and found major discrepancies between predicted risk scores by ethnic group and SMR by country of birth. In the South Asian groups important discrepancies were seen in Indian and Pakistani men and Indian and Bangladeshi women. Overall, the findings show that either risk prediction models work poorly or the mortality and morbidity data showing an excess are wrong, which seems unlikely. In the absence of accurate models for risk prediction, and assuming that mortality data are accurate, guidelines such as those of the British Cardiac Society simply warn readers that risk prediction methods may not be valid for South Asians. 10 One pragmatic and advocated solution to the need to base clinical decisions on risk prediction is to use equations derived from studies on white Europeans but to multiply the result by 1.5 to account for the mismatch between expected and actual prediction in South Asian populations. This is conceptually equivalent to using a lower threshold of predicted risk. Cappuccio et al. 5 concluded by recommending that general practitioners apply a lower risk threshold in people of South Asian and of African Table 1 Summaries of UK studies on risk prediction in South Asians Author/year of publication Populations studied/year/ location South Asian sample size Source of outcomes data Prediction models Analysis of potential risk of reduction Risk prediction ratio of 10 year risk in South Asians versus comparison population (%) Game and Jones; Kooner and Chambers; Capuccio et al.; Quirke et al.; Bhopal et al. (this paper) Clinic population of people with diabetes years (1998); Birmingham Sample from 56 general practices in West London; years (fieldwork dates not stated) Wandsworth, London, 9 general practices, years ( ) England, 1998, 1999; population health surveys Newcastle, England, Population survey years None, but refer to raised CHD risk of mortality in South Asians in literature 514 Mortality data from ONS * None, but refer to raised CHD in South Asians in literature Gill et al. 3 Wild and McKeigue 2 Gill et al. 3 Wild and McKeigue 2 Study sample mortality (preliminary analysis) for CHD only For CHD Dundee (unpublished) Procam (unpublished) For CHD and stroke For CHD FINRISK SCORE for CHD and stroke (two models) Not done Not done Blood pressure only Not done Done using FINRISK and models SA SA SA (Mortality) SA (events) / ratio SA I P B SA I P B See Tables 2 4 and text *Combined for analysis. SA, South Asian; I, Indian; P, Pakistani; B Bangladeshi.

3 PREDICTED AND OBSERVED CARDIOVASCULAR DISEASE IN SOUTH ASIANS 95 origin. Kooner and Chambers 8 make a similar recommendation, pointing out that this approach needs testing. The European guidelines on cardiovascular disease prevention emphasize that the total cardiovascular risk should be the foundation, 11 not just coronary heart disease, and that the SCORE formula should be used for prediction in Europe. 12 We used a European risk prediction model FINRISK which includes diabetes, 13 an important risk factor in South Asians, the SCORE model 12 and the USA based (1991) model, 14 and data from the Newcastle Heart Project. 6 Our selection of FINRISK was based on its European origin, its inclusion of diabetes and HDL, and the experience of using it (EV). While mortality outcomes in the Newcastle Heart Project are too few to permit robust analyses for some years to come we have made preliminary observations on the CHD and stroke death rates. The strength of the Newcastle Heart Project is its population sample of 684 South Asians and 825 Europeans, attention to heterogeneity within the South Asian populations and the wide range of risk factor measures. In particular, diabetes and insulin resistance were characterized using an oral glucose tolerance test. As diabetes prevalence is three to five times higher in South Asian populations than in the European origin populations this is particularly important. Methods The methods of the Newcastle Heart Project have been published 6 (with more details in 215/dc1). We summarize them here. Study sample and recruitment The samples were of men and women aged years. The Newcastle Health and Lifestyle Survey (NHLS) provided the sampling frame for Europeans (N = 6448). The sampling frame was a sub sample of The Family Health Services Authority Register (FHSA). Europeans were screened between April 1993 and October The response rate was 64.2 per cent (840 of 1308 contacted). A name analysis of the FHSA register produced a sampling frame for the South Asian population. South Asians were screened between May 1995 and March The response rate was 67.5 per cent (709 of 1050, of whom 684 were Indian, Pakistani or Bangladeshi.) Newcastle upon Tyne Joint Ethics Committee approved the study. Informed consent was obtained from participants. Measurements Subjects attended the Royal Victoria Infirmary in Newcastle having fasted (excepting those on insulin) from 22:00 h the previous evening. Cholesterol and HDL-cholesterol measurements were fasting. People with diabetes were either on hypoglycemic drugs or insulin; or reported a diagnosis of diabetes and had a fasting plasma glucose of 7.8 mmol/l; or on glucose tolerance test had a 2 h value of 11.1 mmol/l. The mean of two measures of systolic (Korotkoff phase 1) and diastolic (Korotkoff phase V) blood pressures were used for analysis. Europeans self-completed the questionnaire at screening, with help at the screening clinic as required. The South Asian questionnaires were completed at interview by trained interviewers in the subjects home and in the interviewee s preferred language. People reporting smoking regularly or occasionally were regarded as smokers. Analysis Risk factor prevalence data from the Newcastle Heart Project were age-standardized using linear and logistic regression. Figures in this paper do not match exactly those previously published 6 because these are a subsample i.e. excluding those with myocardial infarction, and those with incomplete data on the range of data needed for the model. Patients with previous myocardial infarction were excluded from the analyses. The predicted risk of myocardial infarction or coronary heart disease death over a 10 year period was calculated using the FINRISK, 11,13 14 and SCORE models. 12 The FINRISK model was derived from the occurrence of acute myocardial infarction (International Classification of Diseases or ICD-10 I20 I21) or coronary heart disease death (ICD or ICD-10 I20 I25) during a 10 year follow-up of populations in eastern and southwestern Finland, based on the 1982 and 1987 cohorts. The data were included in the SCORE prediction project. 12 The SCORE and equations have been published. 12,14 The FINRISK equation is similar and is based on a logistic regression model where the log odds of myocardial infarction or CHD death are in men * age in years * smoking (1 if a current smoker, 0 otherwise) * systolic blood pressure in mmhg * total cholesterol in mmol/l * HDL-cholesterol (mmol/l) * diabetes (1 if diagnosed diabetes and 0 otherwise); and in women * age * smoking * systolic blood pressure * total cholesterol * HDL cholesterol * diabetes. The 10 year risk (per cent) was calculated as 1 / (1 + exp(log odds)) * 100. To take account of age differences between ethnic groups, the predicted risk was adjusted for age using linear regression. Because the distribution of risk scores was positively skewed, it was log transformed for analyses. We also explored an alternative method of taking account of age differences: mean levels of the risk factors used in the FINRISK equation were adjusted for age using ANOVA in each ethnic group and sex and the resulting values substituted in the FINRISK prediction equation. The two methods produced very similar results (comparative data available from authors), but this alternative method did not yield confidence intervals easily. We repeated the analysis using the published (1991) and SCORE prediction models. The model

4 96 JOURNAL OF PUBLIC HEALTH also includes ECG changes, with the other risk factors as per FINRISK. The outcomes are 10 year risk of CHD death, non fatal MI, new angina and coronary insufficiency. We also applied the stroke risk model. Finally, to assess the effect of the currently recommended model (based on smoking, systolic blood pressure, and total cholesterol) we used the European SCORE model to predict fatal CHD and non-chd cardiovascular disease outcomes. Deaths among Newcastle Heart Project participants were flagged with the Office of National Statistics. The median follow up time for Europeans was 9.6 years and for South Asians 7.1 years. Death certificates were obtained and the cause of death classified as due to CHD if they had ICD9 codes A410 A414 or ICD10 codes I20 I25 at any position in the list of causes of death and stroke for ICD or ICD 10 I60 I69. Among the Europeans 22/100 deaths and among South Asians 19/31 deaths were classified as CHD deaths, the corresponding figures for stroke being 9/100 and 3/31. We inspected survival plots and used a Cox regression model to estimate the adjusted hazard of CHD death and stroke death among South Asians compared to Europeans. Results Based on deaths in the Newcastle Heart Project population, 22 in Europeans and 19 in South Asians, the hazard ratio adjusted for age and sex for CHD death in South Asians combined compared to Europeans was 2.23 (95 per cent CI 1.13, 4.38), corroborating national data (Table 2) of an excess of CHD mortality in this population; the corresponding ratio for stroke mortality was 1.35 (95 per cent CI ), again in concordance with national data. The number of deaths was too small to permit accurate subgroup analysis, but the excess was in men and women. Tables 2 and 3 gives the risk factor levels, the national mortality data SMRs and the model predictions. There were variations between the South Asian groups in national SMRs, risk factors, and predicted risks, as shown in Tables 2 and 3, so despite the small numbers, we did subgroup analysis. As shown in Table 2 the FINRISK CHD model predicted in South Asian men combined compared with Europeans, a risk ratio of 122 per cent (SMR 142) with substantial subgroup heterogeneity, e.g. 154 per cent in Bangladeshis (SMR 151), 129 per cent in Pakistanis (SMR 148), 99 per cent in Indians (SMR 142). The CHD model predictions were Table 2 National mortality data*, risk factor levels from the Newcastle Heart Project and predicted 10 year risk in men European South Asian Indian Pakistani Bangladeshi n = 362 n = 262 n = 85 n = 129 n = 48 National mortality data by country of birth* SMR for CHD SMR for stroke Systolic blood pressure (mmhg) Cholesterol (mmol/l) HDL-cholesterol (mmol/l) Smoking (%) Diabetes (%) FINRISK model (CHD) Predicted risk (%) (110, 135) 99 (85, 115) 129 (114, 146) 154 (128, 187) SCORE model (CHD mortality) Predicted risk (%) (76, 91) 79 (69, 91) 87 (77, 97) 79 (67, 95) model (CHD) Predicted risk (%) (110, 136) 103 (88, 119) 128 (113, 146) 150 (123, 182) model (stroke) Predicted risk (%) (57, 87) 68 (50, 92) 81 (62, 105) 52 (35, 77) SCORE model (non-chd CVD mortality) Predicted risk (%) (74, 88) 78 (68, 89) 84 (75, 94) 76 (64, 91) *The SMR was calculated using the population structure and rates of England and Wales. Data on Indian, Pakistani and Bangladeshi populations published at (web site accessed 22 August 2003). Analysis of South Asians combined is from Wild and McKeigue. 2

5 PREDICTED AND OBSERVED CARDIOVASCULAR DISEASE IN SOUTH ASIANS 97 Table 3 National mortality data*, risk factor levels from the Newcastle Heart Project and predicted 10 year risk in women European n 363 South Asian n 314 Indian n 145 Pakistani n 132 Bangladeshi n 37 National mortality data by country of birth* SMR for CHD SMR for stroke Systolic blood pressure mmhg Cholesterol (mmol/l) HDL-cholesterol (mmol/l) Smoking (%) Diabetes (%) FINRISK model (CHD) Predicted risk (%) (139, 184) 145 (122, 173) 172 (144, 207) 184 (135, 250) SCORE model (CHD mortality) Predicted risk (%) (75, 91) 88 (78, 99) 81 (72, 92) 68 (55, 84) model (CHD) Predicted risk (%) (124, 172) 140 (114, 172) 154 (124, 191) 147 (102, 212) model (stroke) Predicted risk (%) (74, 122) 110 (80, 150) 99 (71, 138) 46 (26, 80) SCORE model (non CHD CVD mortality) Predicted risk (%) (81, 96) 96 (86, 107) 85 (76, 95) 70 (57, 85) *The SMR was calculated using the population structure and rates of England and Wales. Data on Indian, Pakistani and Bangladeshi populations published at (web site accessed 22 August 2003). Analysis of South Asians combined is from Wild and McKeigue. 2 very similar to FINRISK. The SCORE CHD model predicted lower mortality in all three South Asians groups. The model predicted less stroke in South Asian men than in Europeans, while the SCORE model predicted less non-coronary atherosclerotic disease in South Asians than Europeans. As shown is Table 3 the FINRISK risk ratios for CHD were in South Asian women combined 160 per cent (SMR 145), and for Bangladeshis 184 per cent (SMR 91), Pakistanis 172 per cent (SMR 111) and Indians 145 per cent (SMR 158). The SCORE model showed lower risk in the South Asian groups. The model also showed excess risks in all three South Asian groups. The model predicted less stroke in South Asian women combined, and in all subgroups, than in European women while the SCORE model predicted less non-coronary atherosclerotic disease. The absolute predicted risks for CHD varied by model for each ethnic group, and the model predicted higher morbidity and mortality. For stroke the absolute rates were similar but predicts morbidity and mortality but SCORE only mortality. Age-adjusted predicted risk for CHD is shown in Table 4, re-calculated after substituting a hypothetical low-risk value in the FINRISK equation for each risk factor in turn, assuming risk-factor outcome relationships in the Finnish population hold in the South Asian groups, and that associations are causal and reversible. Systolic blood pressure was substituted with the lowest level within each sex group. The relative importance of total serum cholesterol was assessed by replacing total cholesterol by 4.5 mmol/l, taken from international populations where the coronary mortality is low. The relative importance of HDL-cholesterol was estimated by replacing the mean of HDL cholesterol by the highest level, which was in the European population. Smoking and diabetes prevalences were reduced to zero whilst recognising this is, in practice, an unachievable ideal. The table shows major differences in the modelled impact of these factors between ethnic groups and by sex. For example, if HDL-cholesterol increased to the level in European men modelled coronary outcomes reduce in Bangladeshi men by 36 per cent, by 19 per cent in Pakistani men and by 6 per cent in Indian men; eliminating smoking reduces incidence by 27 per cent in Bangladeshi, 16 per cent in Pakistani and European and

6 98 JOURNAL OF PUBLIC HEALTH Table 4 Predicted 10 year risk from Finrisk model, adjusted for age, and predicted reduction after change in risk factors Risk and risk factor Risk (%) European... reduction Risk (%) South Asian... reduction Risk (%) Indian... reduction Risk (%) Percentage reduction in age-adjusted predicted risk after hypothetical changes to individual and all risk factors. Pakistani... reduction Risk (%) Bangladeshi... Predicted risk at present Blood pressure to mmhg Cholesterol to 4.5 mmol/l HDL-cholesterol to mmol/l No smoking No diabetes All changes Predicted risk at present Blood pressure to mmhg Cholesterol to 4.5 mmol/l HDL-cholesterol to mmol/ litre No smoking No diabetes All changes reduction 8 per cent in Indian men; and eliminating diabetes from the population reduces modelled outcomes by 9 per cent in South Asians combined, 10 per cent in Bangladeshi, 11 per cent in Pakistani, 6 per cent in Indian and only 2 per cent in European men. The role of diabetes was very similar in women and men, HDL-cholesterol was more important in women than in men, and the role of total serum cholesterol in women was about half that of men mainly because the regression coefficient was higher for cholesterol in men and higher for HDL-cholesterol in women. The total predicted risk associated with measured risk factors varied from 49 to 68 per cent in men and from 48 to 71 per cent in women, the proportions being higher in the South Asian populations than in Europeans. A similar analysis with the equation gave corresponding results (available to readers from authors on request). Discussion The main findings in context This work shares some of the limitations of other recent publications 5,7 9 including a small sample size resulting in imprecision of estimates as indicated by wide confidence intervals. Nonetheless, in the absence of large scale cohort studies in South Asian populations the analyses help develop the field. The insights gained from examining the outputs from three models, including the current European recommendations, the use of preliminary outcomes on mortality from the study sample as well as national mortality data, the inclusion of non CHD outcomes, the attention to subgroup heterogeneity, and an assessment of the potential for prevention through the control of risk factors, add to the small existing knowledge base. The FINRISK and models gave similar results but the SCORE model s outputs were not compatible with the evidence of high CHD and stroke mortality in South Asians. 1 4 In the South Asian populations combined in comparison to the European group, the FINRISK and models predicted 10 year risk was compatible with published SMRs (Table 2) and with the preliminary analysis of mortality in the Newcastle Heart Project sample population. The SCORE model s very different results probably reflect the absence of HDL and diabetes as risk factors in the model. The heterogeneity between the South Asian populations, as in the work of Quirke et al., 9 focused attention on subgroups, even though this has the drawback of reduced numbers of subjects. Such predictions were in accord with prior expectation from cross-sectional risk factor data demonstrating heterogeneity in South Asian populations, the highest prevalence of risk factors being in Bangladeshis. 6 The rankings based on modelled predictions were sometimes inconsistent with those based on SMRs. There was, nonetheless, reasonable consistency for South Asian men and women combined, Pakistani and Bangladeshi men, and Indian women. This level of success in prediction with both the FINRISK and models, though imperfect, is more encouraging than previous studies (Table 1). It may have resulted from our using models based on both European and USA populations, measuring diabetes using both self-reported data and a glucose tolerance test, and predicting for both South Asians together and Indians,

7 PREDICTED AND OBSERVED CARDIOVASCULAR DISEASE IN SOUTH ASIANS 99 Pakistanis and Bangladeshis separately. The difference between our findings and those of Quirke etal., 9 other than methodological ones, might be that the Newcastle Heart Project has more accurate data than the Health Survey for England, e.g. we used the WHO definition of diabetes based on the oral glucose tolerance test, where Quirke etal. used doctor-diagnosed diabetes. There was some discrepancy between the FINRISK and predictions and the expected pattern based on the SMR in South Asian women combined (models showing deficit), Pakistani and Bangladeshi women (models showing an excess) and Indian men (models showing a deficit). Of these discrepancies the greatest was in Bangladeshi women and it is difficult to judge whether this discrepancy is reflecting errors in the modelling or in the mortality data. Self-reported prevalence of cardiovascular disorders, and major ECG changes indicative of coronary heart disease, were comparatively low in Bangladeshi women in the Health Survey for England 1999, 15 supporting the expected pattern shown in mortality data, and contradicting the predictions of the models. While the models gave fairly good results for comparing outcomes between ethnic groups, the differences in the absolute predicted rates point to inherent limitations of modelling. Methodological issues The mortality analyses are imprecise because of problems in accurately recording country of birth in both the death certificate and at census, in diagnosing the cause of death, in accounting for deaths abroad among UK residents and in deaths in the United Kingdom of visitors, and the small populations (of Bangladeshis in particular). 3 The limitations of country of birth as a proxy for ethnicity also need to be noted. In the case of the Newcastle Heart Project participants, 90 per cent of the South Asians were born on the Indian Subcontinent. The mortality analyses are for England and Wales, while the predictions are for Newcastle residents. Mortality data by ethnic group are not published specific to Newcastle residents. The Newcastle Heart Project is mortality flagged but because of small numbers of deaths we expect to wait another 5 10 years before robust analyses are possible. On the preliminary analysis reported here, in South Asian men and women combined, the hazard ratios of death for CHD and stroke were compatible with national mortality data. The FINRISK and predicted risk included both morbidity and mortality and the expected rankings were derived from mortality only. However, the coefficients of risk factors in different cohort studies are very similar for mortality and morbidity. Hence, it is unlikely that this affects the difference in predicted and observed mortality patterns. In risk factors relating to CHD and diabetes, Newcastle Heart Project South Asian subgroups, are very similar to those described in the Health Survey for England Newcastle populations of South Asian origin are also similar to those described nationally in the 1991 census in age structure, levels of employment and other indicators of social and economic position. 6 Our findings are unlikely to be idiosyncratic. Implications for modelling, clinical practice and research The results of modelling are dependent on the variables included, as so clearly demonstrated by the differences between the SCORE and the other two models, and the accuracy of measurement. It seems likely that models used in South Asian populations need to include diabetes and HDL. Current models for cardiovascular risk do not include factors of known importance such as socio-economic position, exercise and dietary habits. If these could be taken into account the predictions might be more accurate, but the burden of data collection would make such models less valuable in clinical practice. On the basis of published data on exercise and socio-economic position 6,16,17 we would predict even higher risk in Bangladeshis and Pakistanis, compared to Indians and white Europeans. Extreme caution is required in assessing the need for preventive interventions in South Asian populations using a cut-off based on North American and European cohort studies of per cent modelled 10 year absolute risk. Such caution, however, makes it problematic to manage patients in accordance with the recommendations of the National Service Framework on CHD. 18 Adding 50 per cent to the predicted risk, as sometimes and increasingly advocated, would not have been consistently helpful. This action would have overestimated the risk in Bangladeshi men and South Asian women in all subgroups. The need for cardiovascular research and practice to take into account the heterogeneity of South Asian populations, 6 despite the consequent reduction in study power in research, is demonstrated here and in the study by Quirke et al. 9 We need multi-ethnic cohort studies to examine this process and provide data on risk-factor-disease outcome relations specific to each of Britain s major ethnic groups so that risk prediction models can be developed and tested. Alternatively, prediction models may be adaptable for different populations using risk factor data and registry based incidence rates as shown by Marrugat et al. 19 and D Agostini et al. 20 These techniques assume risk factor-outcome relationships are quantitatively similar across a range of ethnic groups. This was demonstrated by otti et al. 21 in European populations and by the Diverse Populations Collaborative Group 22 in a range of cohort studies. The assumption of similar risk factors outcome relationships is reasonable but needs testing in South Asians. If such assumptions hold in South Asians the prospects for prevention are excellent as demonstrated by the hypothetical benefits shown by the model (Table 3). Conclusion We conclude that the FINRISK and models predict comparatively CHD outcomes fairly well in South Asians combined, and inconsistently but with some success in the subgroups. Subgroup analysis is essential because of the heterogeneity in risk factor patterns, morbidity and mortality and modelled outcomes. Our conclusion is compatible with work using the equation by Quirke et al, 9 but it

8 100 JOURNAL OF PUBLIC HEALTH contrasts with other studies all reporting on South Asians as a single ethnic group, which showed little or no excess risk. 5,7,8 The models make explicit the vast potential for prevention, surprisingly greater in South Asian groups than in Europeans, through the control of a few risk factors on the reasonable but as yet untested assumption that risk factor-outcome relationships are causal and quantitatively similar to those demonstrated in European origin populations. More work is needed to achieve consistent and reliable estimates of absolute risk. Acknowledgements We are grateful to Louise Hayes for managing the deaths database and for providing data, and to Robin Prescott for invaluable statistical advice. We have previously acknowledged individually the many people who have contributed to the Newcastle Heart Project 6 and thank them once again. We also thank our research collaborators for their important work: Julie Yallop, Sheila Patel, Naseer Ahmad, Catherine Turner, Bill Watson, Dalvir Kaur, Anu Kulkarni, Mike Laker, Anna Tavridou. We thank the following for financial support of the Newcastle Heart Project: The Barclay Trust, British Diabetic Association (now Diabetes UK), the former Newcastle Health Authority, R&D Directorate of the former Northern Regional Health Authority; The Department of Health, and The British Heart Foundation. Conceptually, this work owes a debt to John Chambers and Jaspal Kooner. Contributions and guarantors G.A., R.S.B., N.U. and M.W. developed and supervised the European and South Asian components of the Newcastle Heart Project. R.S.B. and E.V. planned and wrote this paper. C.F. did the analysis of data and helped in writing. N.U. and M.W. provided critique, and GA both support and comment. The guarantors are R.S.B., E.V. and C.F. Competing interests We have no competing interests to declare. References 1 Balarajan R. Ethnicity and variations in mortality from coronary heart disease. Health Trends 1996; 28: Wild S, McKeigue P. Cross sectional analysis of mortality by country of birth in England and Wales, Br Med J 1997; 314: Gill PS, Kai J, Bhopal RS, Wild S. Health Care Needs Assessment: Black and Minority Ethnic Groups. In: Raftery J, ed. Health care needs assessment. The epidemiologically based needs assessment reviews. Third Series. Abingdon: Radcliffe Medical Press Ltd; published online at: bemgframe.htm, McKeigue PM. Coronary heart disease in South Asians overseas: areview. J Clin Epidemiol 1989; 42: Cappuccio FP, Oakeshott P, Strazzullo P, Kerry SM. Application of risk estimates to ethnic minorities in United Kingdom and implications for primary prevention of heart disease in general practice: cross sectional population based study. Br Med J 2002; 325: Bhopal R, Unwin N, White M et al. Heterogeneity of coronary heart disease risk factors in Indian, Pakistani, Bangladeshi and European origin populations: cross sectional study. Br Med J 1999; 319: Game FL, Jones AF. Ethnicity and risk factors for coronary heart disease in diabetes mellitus. Diabetes Obes Metab 2000; 2: Kooner JS. Conceptualising the causes of coronary heart disease in South Asians. In: Patel KC, Bhopal RS, eds. The epidemic of coronary heart disease in South Asian populations: causes and consequences. Birmingham: South Asian Health Foundation; Quirke TP, Gill PS, Mant JW, Allan TF. The applicability of the coronary heart disease prediction function to black and minority ethnic groups in the UK. Heart 2003; 89: British Cardiac Society, British Hyperlipidaemia Association, British Hypertension Society, British Diabetic Association. Joint British recommendations on prevention of coronary heart disease in clinical practice. Heart 1998; 80: S1 S De Backer G, Ambrosioni E, Borch-Johnsen K, Brotons C, Cifkova R, Dallongeville J et al. European guidlines on cardiovascular disease prevention in clinical practice. Eur Heart J 2003; 24: Conroy RM, Pyorala K, Fitzgerald AP, Sans S, otti A, De Backer G et al. Estimation of ten-year risk of fatal cardiovascular disease in Europe: the SCORE project. Eur Heart J 2003; 24: Vartiainen E, Jousilahti P, Alfthan G, et al. Cardiovascular risk factor changes in Finland Int J Epidemiol 2000; 29: Anderson KM, Odell PM, Wilson PWF, Kannel WB. Cardiovascular disease risk profiles. Am Heart J 1991; 121: Joint Health Services Unit. Health Survey for England. The health of minority ethnic groups 99. London: The Stationery Office, Hayes L, White M, Unwin N et al. Patterns of physical activity and relationship with risk markers for cardiovascular disease and diabetes in Indian, Pakistani, Bangladeshi and European adults in a UK population. J Publ Hlth Med 2002; 24: Nazroo J. The health of Britain s ethnic minorities. London: Policy Studies Institute; UK Department of Health. National service framework for coronary heart disease. Modern standards and service models. London: The Department of Health, Marrugat J, D Agostino R, Sullivan L et al. An adaptation of the coronary heart disease risk function to European Mediterranean areas. J Epidemiol Commun Hlth 2003; 57: D Agostino RB Sr, Grundy S, Sullivan LM, Wilson P, CHD Risk Prediction Group. Validation of the coronary heart disease prediction scores: results of a multiple ethnic groups investigation. J Am Med Assoc 2001; 286: otti A, Capocaccia R, Conti S et al. Identifying subsets of major risk factors in multivariate estimation of coronary risk. J Chron Dis 1977; 30: Diverse Populations Collaborative G. Prediction of mortality from coronary heart disease among diverse populations: is there a common predictive function? Heart 2002; 88: