Article Text

Download PDFPDF

Original article
Ethnic variations in heart failure: Scottish Health and Ethnicity Linkage Study (SHELS)
  1. R S Bhopal1,
  2. N Bansal1,
  3. C M Fischbacher1,2,
  4. H Brown1,
  5. S Capewell3 on behalf of the Scottish Health and Ethnicity Linkage Study
  1. 1Edinburgh Ethnicity and Health Research Group, Centre for Population Health Sciences, University of Edinburgh, Edinburgh, UK
  2. 2Information Services Division, NHS Scotland National Services, Gyle Crescent, Edinburgh, UK
  3. 3Division of Public Health, University of Liverpool, Liverpool, UK
  1. Correspondence to Professor R Bhopal, Edinburgh Ethnicity and Health Research Group, Centre for Population Health Studies, University of Edinburgh, Teviot Place, Edinburgh EH8 9AG, UK; raj.bhopal{at}


Objective Ethnic variations in heart failure are, apparently, large (eg, up to threefold in South Asians compared with White populations in Leicestershire, UK) but data are limited and conflicting. The incidence of first occurrence of heart failure hospitalisation or death by ethnic group in Scotland was studied.

Design, setting, populations and outcomes A retrospective cohort study was developed of 4.65 million people using non-disclosive, computerised methods linking the Scottish 2001 census (providing ethnic group) to community death and hospital discharge/deaths data (SMR01). Annual, directly age standardised incidence rates per 100 000, incidence rate ratios (RRs) and risk ratios using Poisson regression were calculated. Ratios were multiplied by 100. Risk ratios were adjusted for age and highest education qualification. Statements of difference imply the 95% CI excludes 100 (reference), otherwise the CI is given.

Results In men, other White British (RR=86.4) and Chinese (RR=54.2) had less heart failure than White Scottish (100) populations while Pakistani men had more (RR=134.9). In women, the pattern was similar to men. Adjustment for highest educational qualification attenuated differences in risk ratios in other White British men (risk ratio=75.8 to 85.4) and women (66.2 to 74.6), made little difference to Pakistani men (146.9 to 142.1) and women (177.4 to 158.1), and augmented them in Indian men (115.4 (95% CI 93.1 to 143.0) to 131.7 (107.4 to 161.5)).

Conclusions Ethnic variations in heart failure were important in this population setting and not abolished by adjusting for highest education, one important indicator of socioeconomic differences. The ethnic variations were substantial but did not support other studies showing 3–20-fold differences between ethnic groups.

Statistics from


Ethnic variations in most manifestations of cardiovascular disease (CVD), including stroke and ischaemic heart disease (IHD), are large—for example, in the UK there is an excess of 50–70% in IHD mortality rates in South Asians with the highest rates in Bangladeshi and Pakistani men.1 Heart failure is a major and increasing cause of hospitalisation (and hospital costs) and death because of population ageing and fewer early deaths from myocardial infarction due to better interventions.2 3 Most heart failure in industrialised countries, including the UK, is due to IHD or idiopathic cardiomyopathy or hypertension.4–6 Socioeconomic status is strongly related to IHD and hypertension, and also to infectious diseases, including those leading to rheumatic fever. In recent studies, IHD, and risk factors such as hypertension and diabetes, were commonly associated with heart failure in South Asian and African origin ethnic minority groups in the UK, while both valvular disease and atrial fibrillation were much rarer, even although a large proportion of the population studied were born abroad, and valvular disease might have been expected to be commoner than actually observed.4 5

There are data indicating that heart failure burden is greater in non-White ethnic minority groups7 and investigations into both the burden and causes of this phenomenon (if true) are warranted. In 2010 the Canadian Cardiovascular Society published guidelines on heart failure in ethnic minority populations6 based on a review of the international literature with emphasis on South Asian, Chinese, Black and Aboriginal populations.3 They concluded that incidence, survival and management vary importantly by ethnic group and that researchers and clinicians need to act upon such data. In an extreme example, the guidelines noted that in the Coronary Artery Risk Development in Young Adults (CARDIA) study, African–American men at an average age of 39 years had 20 times the incidence rate of heart failure seen in White people.3 6 Alexander et al's linkage study showed important differences in hospitalisation incidence rate and outcome in heart failure in California.8 African–American men and women had about twice the hospitalisation rate of the White population, but lower mortality. Hispanic populations were similar to the White population while Asians, men in particular, had lower hospitalisation and mortality rates (the nature of the Asian population was not clarified but East Asians—Chinese and Japanese—are likely to be predominant). Kaul et al found, compared with White populations, comparatively high 1 year mortality in Chinese populations in Canada, but not Indians.9 CVD rates in Chinese populations are highly variable geographically.10 In Leicestershire, UK, South Asians had 3–5-fold higher rates of heart failure admission than White people but better survival.11 12 In Europe, national data are only available from The Netherlands, and these show variations by country of birth in survival after hospitalisation for heart failure.13

To establish whether the burden does indeed vary by ethnic group, we need accurate population based statistics on the incidence of, and survival from, heart failure by ethnic group.7 Moe and Tu's review concluded that most published studies are based on small samples, and given the increasingly multiethnic world and the rising frequency of heart failure, additional studies across different ethnic groups are needed.3 Sosin et al also pointed to the paucity of data, particularly for those of South Asian origin.14 In the absence of prospective cohort studies to provide accurate incidence and/or mortality data by ethnic group in Europe,15 we have created a retrospective cohort study of 4.65 million people in Scotland.16 17 Scotland's major migrant populations, and hence ethnic majority groups, are from England and Wales, Ireland, Pakistan, India and China. Mostly these are long established populations (settling mostly in the 1960s and 1970s), although statistical data on length of residence are not available. More recently established populations include those from Africa and the Caribbean.18 This cohort study permitted us to examine heart failure first hospitalisation and/or mortality rates by self-reported ethnic group.


The detailed methods have been published, including issues relating to ethics approval and our strict controls on data disclosure.16 17 In brief, we developed a retrospective cohort study using computerised matching of names, addresses, sex and dates of birth by linking the census 2001 for Scotland (providing self-reported ethnic group and other exposure and confounding variables) to the National Health Service Community Health Index (a register of National Health Service (NHS) users) and then to death records and hospital discharge/deaths data (the database is called SMR01). Ninety-five per cent of the 4.9 million Scottish population that completed forms at the census 2001 were linked to health records, with at least 85% linked in every ethnic group. At census, one category is selected from a list of 14 categories in response to the question ‘What is your ethnic group?’ A guiding brief note asks people to choose the category that indicates best the person's cultural background. The category labels are self-explanatory. The cohort captured about 90% of the Scottish population in 2001.17 As numbers were small, we combined Bangladeshis with other South Asians; and Caribbean, African and any other Black background as African origin (henceforth African for short).

Of the total, 24.3% of heart failure events were deaths; the remainder were hospitalisations without death. We calculated for first incident of heart failure discharge and/or death between May 2001 and April 2008 separately in men and women aged 30 years or more, directly age standardised incidence rates per 100 000 per year and incidence rate ratios (RRs) from these rates. First heart failure meant there was no other hospital admission for that person for heart failure in the SMR01 database in the previous 10 years. The ICD codes used were I50 (ICD10) and 428 (ICD 9). Also, we calculated risk ratios using Poisson regression with robust variance. For ratios, we multiplied by 100 so the standard reference White Scottish population had a value of 100; 95% CIs were calculated for each summary measure. Poisson regression models were chosen due to their suitability for analysing frequency data. The independent variables were selected in advance in order to test our prespecified hypotheses. We did not test for interaction given the lack of power for most analyses in most ethnic groups.

Results are reported in the text for rate and risk ratios where the 95% CI excludes the reference value of 100. We always adjusted for age group, and sometimes for age and highest education status (see below for reasoning). No other confounders were included.

In independent analyses (manuscript on our methods in preparation), we examined the relationship between eight indicators of socioeconomic position and all CVD rates, the numbers being insufficient for examining heart failure alone. The indicators were: (1) the postcode (zipcode) based Scottish Index of Multiple Deprivation, (2) car ownership, (3) highest qualification of the individual, (4) highest qualification in the household, (5) National Statistics Socioeconomic Classification at individual and (6) household levels, (7) household tenure and (8) economic activity in the previous week. The indicator that showed the most consistent associations across all ethnic groups was highest educational qualification of the individual so we used this (discussion section below gives the rationale) to help assess potential socioeconomic confounding. Henceforth all references in this paper to education and highest qualification refer to the indicator ‘highest qualification of the individual’. This variable was categorised as (1) no qualifications, (2) school level qualifications and (3) higher education level qualifications (eg, professional qualifications, diplomas and degrees). In every ethnic group the RR for all CVD deaths/hospitalisations for those individuals with a higher qualification compared with no qualification was <1 (mostly about 0.8), and the 95% CI excluded 1 in 8/10 ethnic groups for men and in 9/10 groups in women. The other seven indicators were less consistent. Education varied by ethnic group. The population ≥75 years was excluded as they were requested in the census form not to answer the education question, so the number of events is slightly smaller for the risk ratios compared with directly age standardised incidence rates and ratios, the latter maximising study power and providing a cross check for analysis of the age group 30–74 years. The Disclosure Control Committee (see ethics below) ruled that publishing the numerators by ethnic group for both analyses risked inadvertent disclosure of identity. So numerators are suppressed for this second analysis. The reduction in cases was least in the younger non-White populations (eg, the cases in White Scottish men declined from 37 265 to 23 635 but for Pakistani men from 120 to 108).

Ethics and disclosure

The work was approved by the Multicentre Research Ethics Committee (for Scotland) and the Privacy Advisory Committee of NHS National Services Scotland. The ethical and other permissions and related issues have been reported in detail previously,16 including an independent assessment by an ethicist.19 The analysis was conducted on a standalone computer in a locked room in the General Register Office for Scotland (GROS) now called National Records of Scotland, by named researchers (HB, NB, GB, MS), following a strict protocol. Outputs were screened by a GROS disclosure committee before release from GROS.


Table 1 shows that with the exception of Chinese, age standardised incidence rates of first hospitalisation and/or mortality for heart failure were higher in men than in women. In men, Other White British (RR=86.4) and Chinese (RR=54.2) had less heart failure than White Scottish, while Pakistani (RR=134.9) men had more. In women, the patterns were similar: Other White British (RR=86.0) and Chinese (RR=62.9) had comparatively low risks but in Pakistani women the 95% CI included 100.

Table 1

Heart failure: directly standardised annual rates (per 100 000), rate ratios, and age and education adjusted risk ratios* (with 95% CIs) in people ≥30 years, by ethnic group and sex

As the risk ratio analysis shows, some differences were attenuated and others increased by adjusting for highest educational qualification. The difference between Other White British and White Scottish men was reduced but not abolished (risk ratio changed from 75.8 to 85.4). Differences now emerged in Indian men (risk ratio after education adjustment=131.7). The risk ratios for Chinese and Pakistani men hardly changed. In women, the difference in the Other White British was only slightly attenuated. The differences between Pakistani and White Scottish women were clearcut and little affected by adjustment for education (risk ratio=158.1).

While numbers of outcomes were small, there were no notable variations for African men or women, with and without adjustment. The upper limit of the 95% CI around the RR was 173.9 in men and 112.4 in women.


Principal findings

This study provides both an innovative approach and new data to shed light on the increasingly important issue of ethnic variations in the incidence of heart failure hospitalisation/mortality. Ethnic variations were demonstrated, and were potentially important, but they did not corroborate previous work indicating 3–20-fold differences.3 6 11 Rather, the data point to comparatively modest, although still clinically important, variations. A recent cross sectional study based on detailed clinical and laboratory investigations also reported a similar surprising result for both South Asian and Black African–Caribbean populations.5 Together, the studies call for a critical reappraisal of previous conclusions, at least in the UK setting.

Strengths and limitations of the study

The strength of the study is the overall size (4.65 million people, about 3 million ≥30 years of age), national population setting, availability of an ethnic code completed either by the householder on behalf of household members or by individual members, information on a wide range of ethnic groups and linkage of the census data to both hospital morbidity and community/hospital mortality so differences do not merely reflect differences in hospitalisation.16 In Scotland, because of free healthcare at the point of delivery and free ambulance transport, people with severe illnesses are usually managed in hospital. Audits show high completeness (99%) and quality (94.8% diagnostic coding accuracy) of the SMR01 file for CVD (cerebrovascular) diagnoses and 94.2% accuracy for CHD diagnoses.20 Heart failure has not been audited separately. All deaths are certified by a doctor in Scotland and all public hospitals are required to submit data. The private hospital sector is small in Scotland especially for non-elective problems.

The validity of available indicators of socioeconomic position is not yet established in multiethnic studies.21–23 If indicators were acting as confounding variables they would be statistically associated with the outcome (all CVD here, used to maximise the statistical power of the analysis). As socioeconomic variables are co-correlated, adjustment for all would be inappropriate, not least because mismeasurement errors in confounding factors may lead to either diminution or exaggeration of associations.24 25 The choice of education was based on first, visual examination of graphs showing the relationship between each of eight indicators (in each ethnic group by sex) and all CVD directly standardised death/hospitalisation rates. We quantified associations using relative risks. A committee of five people agreed, unanimously, that highest educational qualification was the superior indicator. Examination of the role of social and economic factors is proving complex internationally, but others have also found education to be a potentially good indicator, and in one study this was extracted from the census, as here.26 27

These data are novel in Europe, both in terms of findings and in linkage methods.15 28 Previous linkage studies have not simultaneously combined census, death and hospitalisation data to examine heart failure.8 13

The weaknesses of the study include the small population size for some non-White populations and hence small number of events and imprecision of estimates; the inability to capture events that may have occurred outside the UK; the fact that the cases are those admitted to hospital or dying, excluding those managed at home; and the unavailability of CVD risk factor data. There is no easy way of capturing events internationally, but deaths of UK residents are reported back via several channels, including embassies and consulates, and the primary care registration systems. Such reports, however, may not give accurate cause of death. There are no similar linkage or prospective cohort studies on a range of ethnic groups in other parts of Europe where ethnic minority populations are larger so we cannot directly compare our results. For small populations, the 95% CIs indicate the likely extent of variations, and even for the smaller populations these are fairly narrow (eg, for African women the CI for the rate ratio ranged from 31.5 to 112.7 and for men from 66.8 to 152.2, making a great difference in incidence rates in the Scottish context unlikely).

Our linkage does not provide cardiovascular risk factor data. Furthermore, it is not possible to identify the specific cause of heart failure from the data we have since ICD code I50 does not distinguish valvular, rheumatic, ischaemic or other causes. In the UK setting, valvular and rheumatic heart disease are uncommon causes of heart failure.4 5

Findings in relation to the scientific literature

In judging whether an ethnic group has a low, high or average rate of a disease, we need to consider the reference/comparison population. Usually, as here, this is the majority (often White) population, notwithstanding the limitations of this approach.29 The Scottish population has, by international standards, relatively high rates of CVDs, with major variations according to socioeconomic status, and high levels of sudden cardiac death.30 Even so, data showing extremely high heart failure rates in ethnic minority populations, especially those of African3 8 or South Asian origins,12 were not corroborated in Scotland. The African origin populations in Scotland are comparatively young, as were those in the CARDIA study reviewed by the Canadian Cardiovascular Society.6 There is no suggestion in our data of the 20-fold risks reported there or of the doubled risks found by Armstrong et al in California.8 The observation of 3–5-fold excesses of heart failure in South Asians in Leicestershire and Birmingham,7 12 both in England, were not supported here, thereby potentially justifying criticisms, especially absence of a population denominator, of the Leicester data made by Fischbacher and Bhopal.31 Even comparing results using the standardised rates found in other White British men (379.5) (predominantly English)—436.5 in Indian men, 592.0 in Pakistani men (table 1)—however, does not show the kind of variations seen in Leicestershire11 and Birmingham.7 Crucially, the cross sectional study by Galasko et al in Harrow, England, showed no substantial difference between South Asian and White populations.4 The 2011 results of the prevalence study in Birmingham also help interpret the conflicting findings.5 They indicate no differences in left ventricular systolic dysfunction or heart failure between the South Asian and Black African–Caribbean group, and the overall prevalence figures were comparable to the national quality and outcomes framework prevalence figures.

While Alexander et al reported a low rate of heart failure in Asians,8 their population was not likely to be the same as our South Asians, but East Asian. Our East Asian population, the Chinese, also had low heart failure rates, in line with those in California.8 The highest risk of heart failure was in Pakistani men and women, with a clear separation from the Indian population, as forewarned from an examination of risk factor patterns published in 1999.32

Given controversies over whether ethnic variations in disease arise largely from socioeconomic variations—a hypothesis that is advocated by researchers including Nazroo—33 34 or largely because of other factors such as culture and lifestyle, or access to healthcare or adherence to medication, studies adjusting for socioeconomic measures are needed.22 Our analysis showed socioeconomic circumstance, as measured by highest educational qualification—the best performing potential indicator of eight tested—was a potential factor in the explanation for the comparatively low heart failure risk in Other White British (mainly in men) but not so for other ethnic groups. Such analyses, including potential confounding factors that alter the strength of associations, signal but do not in themselves explain the causes of ethnic variations. Our data, based on linkage of administrative/clinical databases, cannot disentangle socioeconomic, lifestyle, cultural and genetic causes. They do point to the need for more detailed heart failure studies such as the Ethnic-Echocardiographic Heart of England Screening (E-ECHOES) study, which is, however, cross sectional.5 7

Heart failure is usually secondary to underlying cardiovascular problems, principally IHD, but also chronic hypertension and heart valve defects, including rheumatic heart disease. Ethnic variations in heart failure in the UK are not likely to result from differences in valvular defects, as these have been shown to be uncommon compared with other underlying causes—for example, in Birmingham, 16 of 3442 South Asian people (0.47%) had valvular problems while 1570 (45.6%) had hypertension, 328 (9.5%) had angina and 299 (8.7%) had myocardial infarction or related procedures.5 Heart failure in African–American populations is, compared with White comparison populations, less likely to be a result of IHD and more likely to result from hypertension, a result that also held in the African–Caribbean population studied in the UK based E-ECHOES study.5 7 Our results are likely to reflect the underlying pattern of IHD which is known to be relatively more common in South Asian populations and the Scottish born population and less common in Chinese populations and the other White British/England and Wales born populations.1 By contrast, blood pressure levels are not raised comparative to the UK White populations in South Asians combined35 or Chinese populations,36 and seem to be comparatively low in specific South Asian groups (eg, Pakistani and Bangladeshi populations).35

Detailed clinical studies are providing deeper understanding of ethnic variations in heart failure. In West London, Indian Asians (Indians, Pakistanis and Bangladeshis combined) had attenuated left ventricular function, higher left ventricle filling pressure and more concentric remodelling than a comparison group of white Europeans. These differences were not affected by adjustment for potential confounding factors.37 In the US based Multi-Ethnic Study of Atherosclerosis, important subclinical differences in systolic function were found, with Chinese–Americans having the most, and African–Americans and Hispanic Americans the least, favourable profiles.38 These clinical studies are in line with epidemiological patterns, including those shown here.


Ethnic variations in heart failure clearly exist, although their direction and dimension still requires close scrutiny. Our study indicates that differences are likely to vary from place to place and that they do not seem to merely reflect socioeconomic factors. As such, ethnicity needs to be given more prominence within European policies and interventions designed to reduce inequalities.39 40 Our linkage approach could be applied in countries where the population register or census holds ethnic group or proxy variables such as country of birth. It has potential, demonstrated here and in Holland,13 for studying the association between ethnicity and disease. Given the rising burden of heart failure, and the increasing ethnic diversity of European and North American populations, clinical strategies and healthcare guidelines need to incorporate the ethnic dimension, as recommended in Canada.6 Clinical research and practice, and healthcare delivery, will be informed by incorporating knowledge about such variations.41 We also need to redouble efforts to prevent heart failure by reducing IHD through control of risk factors such as hypertension, smoking, hyperlipidaemia and diabetes.

Key messages

The research advances the somewhat neglected subject of ethnic variations in heart failure in four ways:

  • It demonstrates a method for obtaining cohort data in large populations and a wide range of ethnic groups at low cost.

  • It provides unique data in a European setting on ethnic variations in heart failure.

  • It corroborates previous observations that ethnic variations exist and are important.

  • It does not corroborate prior findings of large differences—up to 20-fold in some populations—possibly because of the greater reliability of our methods.


Anne Houghton gave secretarial help to prepare the paper. Referees' feedback led to substantial improvements to the submitted version.



  • Funding The work was funded by the Chief Scientist's Office (grant CZH/4/432). NHS Health Scotland provided a supplementary grant. ISD and the General Register Office for Scotland both supported this work. The researchers acted independently of the funding body and the study sponsor (the University of Edinburgh) at all stages of the work.

  • Competing interests None.

  • Ethics approval The work was approved by the Multicentre Research Ethics Committee (for Scotland) and the Privacy Advisory Committee of NHS National Services Scotland.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.