Background Place of cancer death varies ethnically and internationally. Palliative care reviews highlight limited ability to demonstrate equal access due to incomplete or unreliable ethnicity data.
Aim To establish place of cancer death by ethnicity and describe patient characteristics.
Design We linked census, hospital episode and mortality data for 117 467 persons dying of cancer, 2001–2009. With White Scottish population as reference, prevalence ratios (PR), 95% CIs and p values of death in hospital, home or hospice adjusted for sex and age were calculated by ethnic group.
Results White Scottish group and minority ethnic groups combined constituted 91% and 0.4% of cancer deaths, respectively. South Asian, Chinese and African Origin patients were youngest at death (66, 66 and 65.9 years). Compared with the Scottish White reference, the White Irish (1.15 (1.10 to 1.22), p<0.0001) and Other White British (1.07 (1.02 to 1.12), p=0.003) groups were more likely to die at home. Generally, affluent Scottish White patients were less likely to die in hospital and more likely to die at home or in a hospice regardless of socioeconomic indicator used.
Conclusions Cancer deaths occur most often in hospital (52.3%) for all ethnic groups. Regardless of the socioeconomic indicator used, more affluent Scottish White patients were less likely to die in hospital; existing socioeconomic indicators detected no clear trend for the non-White population. Regardless of ethnic group, significant work is required to achieve more people dying at home or the setting of their choice.
- Home care
- Hospice care
- Hospital care
- Terminal care
Statistics from Altmetric.com
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.
Cancer patient and carer decisions regarding where to die are complex and influenced by many factors. These include presence of carer support, the health status and emotional capacity of the main carer, the availability and use of specialist and community-based palliative care services, the need for specialist symptom control, tumour type, distance to services, the pattern of functional decline before death, attitude to dying and the strength of patient and carer preference for a particular place of death that may change over time.1–3 Many of these influencing factors are culturally and ethnically shaped. In the UK, one study4 found that Chinese patients prefer to die in a hospital environment reflecting consideration of the effect of their illness and death on the family and the perception that dying at home leads to ‘contamination’ of the family home. A further study5 based on the experience of Bangladeshi migrants in London proposed that differing ethnic groups favour home deaths due to the relative isolation of hospital care resulting from poor communication, differing food requirements and desire to fulfil religious duties. Others suggest that place of death differences between Chinese migrants living in the UK who prefer a hospital death compared with those of their compatriots living in Taiwan who prefer home death may reflect the shorter term living accommodation found in UK cities compared with ancestral accommodation in their home country.4
A systematic review of primary research on minority ethnic groups and end-of-life care in the UK reports that Asian patients with cancer were half as likely to die in a hospice as non-Asian patients while similar proportions of Black and White patients who wanted to die at home did so.6 Compared with the White population in southeast England, death in hospice was significantly less likely for Pakistani, Indian and Bangladeshi patients; death at home was significantly less likely for Black African, Black Caribbean and Chinese patients; while Pakistani, Indian, Bangladeshi, Black African, Black Caribbean and Chinese patients were all significantly more likely than White patients to die in hospital.7
A systematic review of studies conducted in the USA and covering other Western industrialised nations identified that home death was consistently more likely among White patients and those living in less racially diverse areas compared with Black, Latino or non-White patients. Furthermore, native-born patients were more likely to die at home than foreign-born patients. Studies examining an area or individual socioeconomic indicator reported that higher socioeconomic status was associated with home death.8
In New Zealand, Asians aged 65 years or younger with cancer and admitted initially to hospice for respite care were more likely to die in a hospice setting.9 In California, a multilevel study established for US-born Asians and Hispanics, age, marital status, nursing home residence and distance to a hospital explained differences in place of death compared with the White population while for African-Americans, availability of hospital beds accounted for most of the difference.10
Recent reviews of palliative care services highlight the inability to demonstrate equal access to services as almost no access data are gathered.7 ,11 As a result, few studies have focused on ethnicity differences at this last stage of life in the UK.7 ,12 Living and Dying Well, the strategy addressing palliative care services provision in Scotland, advocated an approach to care that was person centred, based on patient or carer needs, and responsive to the diversity of life circumstances of those in need of end-of-life care regardless of age, disability, gender, race, ethnicity, religion or belief, or sexual orientation.13
Data linkage of the Scottish census with previously linked hospital episode, cancer registry and mortality data through the Scottish Health Ethnicity Linkage Study (SHELS) provided access to self-declared ethnicity data and the opportunity to evaluate ethnicity and place of death from cancer in Scotland for the first time.14 ,15 Our objective was to establish the pattern of place of death from cancer by ethnic group to describe socioeconomic, demographic and clinical characteristics of patients in order to understand any differences and to evaluate several socioeconomic variables as possible confounders.
The University of Edinburgh, Information Services Division (ISD) of the National Health Services (NHS) National Services Scotland and National Records Scotland (NRS) collaborated on SHELS to link specified fields (eg, ethnic code) of the 2001 Scottish census with an already linked data set maintained by ISD, including hospital episode (Scottish Morbidity Record) and NRS mortality data.7 ,14 ,16
All cancer patients aged 25 years or older who participated in the 2001 Scottish census and died from cancer in Scotland between May 2001 and December 2009 were included in the study. Cancer diagnosis was grouped into all cancers (including C44, non-melanoma skin cancer) and the four main incident cancers in Scotland: lung (ICD10 C33, C34; ICD9 162), colorectal (ICD10 C18-C21; ICD9 153, 154), breast (ICD10 C50; ICD9 174, 175) and prostate (ICD10 C61; ICD9 185).
Death data (place of death and underlying cause of death) were based on the death certificate and related system managed by NRS. Place of death was categorised as hospital, home or hospice based on the location code managed by ISD and used by NRS to record place of death on the death certificate. Scottish Morbidity Records for inpatients and day cases and relevant fields (discharge code and specialty) were used to identify hospice units housed within a hospital and without a unique location code. For the period in question (2001–2009), the ‘place’ operated at least part of the period.
Self-assessed ethnicity was based on 14 ethnic codes recorded in the Scottish 2001 census.16 After reviewing each ethnic group separately to confirm no important differences, the ethnic groups used were White Scottish; White Irish; Other White British; Other White; South Asian (Pakistani, Indian and Other South Asian, including Bangladeshis, given small numbers); Chinese, African Origin (Scottish Black, African, Caribbean or other Black) and Other non-White (other ethnic group and mixed).
Age at census was based on the Scottish 2001 census. Age at death in 5-year-age bands was derived from age at census along with month and year of death from the death record.
NHS board region of residence was based on postcode of residence reported on the 2001 census. Highest education level for individuals 25–74 years old was based on the 2001 Scottish census and grouped as none, lower (Scottish Standard or higher grade or equivalent) or higher (Higher National Certificate or equivalent, degree level or professional).
Association between eight indicators of socioeconomic status and place of death from cancer relative to the remaining place of death categories by ethnic group was examined to establish if any of the socioeconomic indicators were potentially valid confounding factors with direction of association consistent across all the ethnic groups. The indicators were postcode-based Scottish Index of Multiple Deprivation, car ownership, highest qualification of the individual, highest qualification in the household, National Statistics Socioeconomic Classification at individual and household levels, and economic activity in the week preceding census completion.
The Pearson χ2 test of association was used to assess the statistical significance of differences in proportions using two-sided p values and a significance level adjusted for the number of comparisons (Bonferroni correction) to account for the overall probability of a false significant result when making multiple comparisons.17 Fisher's exact test was performed where the number of cases totalled less than 10. Prevalence ratios and 95% CIs were calculated using three univariate Poisson regression models for death in hospital, death at home and death in hospice versus elsewhere (all other places of cancer death combined). The models were adjusted for sex and age; no adjustment was made for socioeconomic status as none of the socioeconomic indicators were consistently associated with the outcome across ethnic groups. Due to small numbers in each of the separate ethnic groups, South Asian, African Origin, Chinese and Other non-White ethnic groups were combined representing the non-White population. The Other White ethnic group was excluded in the prevalence ratio analyses given the disparate nature of this group compared with the Other White British and White Irish ethnic groups. The White Scottish group was the reference in all models. Age and sex-adjusted prevalence ratios and 95% CIs for each place of death compared with all other places by ethnic group and socioeconomic indicator were also calculated. All models and analyses were conducted using SAS V.9.2 (SAS Institute Inc., Cary, North Carolina, USA).
Multicentre Research Ethics Committee for Scotland and the Privacy Advisory Committee of the NHS National Services Scotland approved this study. The ethics, other permissions and related issues have been reported in detail.14 ,15 ,18 Data Protection Act and safe setting rules required the data set only contained cancer outcomes. Analysis was conducted on a secure standalone computer, following strict disclosure protocol. Outputs leaving the safe setting (including this paper) were screened by an NRS disclosure committee.
The White Scottish group constituted 91% (106 898) of all cancer deaths (117 467) occurring between 2001 and 2009. South Asian, Chinese, African Origin and Other non-White ethnic patients combined who died of cancer were small in number (473) and represented a small proportion (0.4%) of all cancer deaths (table 1).
White Scottish patients were older at death (72.4 years) than the South Asian, Chinese, African Origin and Other non-White ethnic groups (65.9–69.0 years) but younger than the remaining White ethnic groups (73.3–73.6 years) (table 1).
Just over 14% of White Scottish patients dying from cancer held a higher qualification; only the Chinese group had a lower proportion (11.5%) while the African Origin proportion was greatest at 48.7%. A lower proportion of cancer deaths was due to one of the four main cancers (lung, colorectal, breast, prostate) among South Asian and Chinese groups (42.7% and 42.3%, respectively) when compared with the White Scottish group (50.2%), while the greatest proportion of African Origin patients died from these cancers (64.7%).
The White Scottish (52.5%), Other White British (64.3%) and Other White (47.1%) groups all had the largest proportion of their respective decedent populations living in All Other Board areas. Most White Irish (43.9%) and South Asian (49%) cancer decedents lived in Greater Glasgow & Clyde NHS Board. Around a third of the Chinese decedents lived in All Other Board areas, the dominant region for this group (table 1).
Regardless of group, the proportion of cancer deaths occurring in hospital was greatest (47.9%–66.7%), followed by home (excluding Chinese patients where hospice was the second most prevalent place of death) and ultimately hospice. The African Origin group had the largest proportion of hospital cancer deaths (66.7%). The Other non-White group had the largest proportion of cancer deaths occurring at home (37.2%), followed by the White Irish group (34.7%). The Chinese group had the largest proportion of hospice cancer deaths (29.9%) (table 2).
After adjustment for sex and age, prevalence ratios, CIs and p values (using Bonferroni-corrected level of significance to account for multiple testing) indicate that differences remain only for death in home versus elsewhere. Compared with the Scottish White reference, the White Irish (1.15 (1.10 to 1.22), p<0.0001) and Other White British (1.07 (1.02 to 1.12), p=0.003) groups were more likely to die at home (table 3).
Generally, higher socioeconomic status among the White Scottish group was predictive of hospice or home death with a lower risk of a hospital death (table 4). This trend was evident for the other White ethnic groups in Scotland but harder to discern for the minority ethnic groups (see online supplementary appendix 1).
No socioeconomic indicator was consistently associated in the same direction with selected place of cancer death compared with other places in each ethnic group for either sex. These indicators, therefore, did not meet the requirement of a confounding variable for our purposes (see online supplementary appendix 1).
We established that cancer deaths occur most often in hospital, home and hospice in that order for all cancer deaths and almost all ethnic groups. After allowing for multiple testing and compared with the White Scottish group, White Irish and Other White British cancer patients were more likely to die at home.
Differences in place of death distributions before adjustment for age and sex were not significant for most non-White minority ethnic groups, suggesting little variation between groups and the Scottish White population; only the Chinese group showed differences (p=0.0021). Given the small number of cancer deaths among the Chinese, this result requires confirmation by other studies. Nevertheless, our findings may reflect preference among Chinese migrants living in the UK to die in a clinical setting.4 Alternatively, this finding may reflect the younger age profile of the Chinese group. Younger age at census and death for the Chinese group may reflect Scotland's younger non-White ethnic population19 while the lower proportion being diagnosed with the four main cancers may reflect this younger age profile, as well as lower incidence rates of these cancers in the country of origin.
Compared with the White Scottish population, we identify that Other White British and White Irish cancer patients were more likely to die at home. A greater or substantially larger proportion of these groups lived in the more rural and remote areas of Scotland, described as Other Board areas. In these areas, hospice provision is more limited.11 As a result, home death may reflect service availability rather than patient or carer preference, information that was not available to us.
Findings in relation to the literature
An English study established that Pakistani, Indian, Bangladeshi, Black African, Black Caribbean and Chinese patients are all more likely to experience a hospital death from one of the four main cancers compared with White patients.7 In contrast, we found no difference in the likelihood of hospital death between the groups. Palliative care (NHS and voluntary sector) adult beds per head of population for England and Scotland separately suggest that hospice provision is less available in England compared with Scotland (5.0 and 6.8 beds per 100 000 population, respectively); this difference in provision might explain at least some of the observed differences between Scotland and England.20
A prospective longitudinal qualitative study interviewing South Asian Sikh and Muslim patients concluded that recent migrants have the poorest access to palliative services.12 However, our findings suggest little difference between South Asian and the White Scottish reference with respect to hospice death. Our population-based results may suggest that the differences identified through interview of a relatively small number of patients, carers and health professionals may not be reflective of all patients of similar ethnicity in Scotland. However, it should also be acknowledged that place of death is a relatively crude indicator of access to appropriate palliative care services.
Although not quite statistically significant by p value after allowing for multiple testing, compared with the White Scottish reference, the minority ethnic groups combined were less likely to die at home. This may be explained by the greater proportion of minority groups residing in urban areas where institutional provision is more concentrated.11 In addition, the proportion residing in the most deprived areas was twice that of the White Scottish reference. Others have established that home death is associated with higher socioeconomic status and lower availability of hospital beds.21 However, compared with the reference group, a greater proportion of the combined group held higher qualifications; others have identified that higher education is associated with death at home.22 The explanation for our findings requires further research.
Although time periods differ slightly, our results indicating 31.7% of all cancer deaths occur at home is comparable to Southeast England (30.3%)7 and is slightly above the average (29.8%) of WHO estimates for selected countries.23 Recent work in England and Wales highlighted a reversal in the declining trend in proportion of patients dying at home since 2004.24 In Scotland, the number of hospital cancer deaths has consistently declined from a peak of 8064 in 2003 to 7623 in 2008 while home deaths have increased slightly over the same period (from 3628 to 3761) and hospice deaths have increased steadily (2477 to 2839).25 On the face of it, these changes are in the desired direction, but the pace of change appears slow.
Consistent with another SHELS study evaluating ethnicity and cancer risk,16 we were unable to identify any socioeconomic variable that behaved as a confounding factor in a consistent manner across ethnic groups. The variables available reflect the Scottish context and consequently the dominant White Scottish population. As a result, these indicators appear less suitable for measuring socioeconomic circumstances of the non-White population. It is generally accepted there is no single measure of socioeconomic circumstances that comprehensively reflects all (related) aspects of socioeconomic stratification.26 Given our focus on place of cancer death, it is logical that multiple socioeconomic variables measured throughout the life course are required. Our results suggest the added dimension of ethnicity also requires to be considered.
Strengths of the study
The SHELS study provided a population-based cohort of patients diagnosed with and dying from cancer where ethnicity was self-reported. As a result, our study uniquely provides a more accurate ethnicity assessment and comprehensive coverage of the Scottish population.16 This is particularly relevant given an ethnic population in Scotland that is small both in absolute number and proportionally. Linkage to the census provided the opportunity to explore the potential confounding effect of individual, as well as area socioeconomic variables not otherwise available. Use of information available through the Scottish Morbidity Record enabled us to identify deaths occurring in palliative care units embedded in hospitals, thereby more accurately reflecting the prevalence of hospice place of death.
Limitations of the study
A limitation of our study is the representation of ethnic groups.14 With respect to cancer deaths in particular, the minority ethnic groups contributed only 0.4% (473) of all cancer deaths. Consequently, for modelling results, we aggregated South Asian, African Origin, Chinese and Other non-White groups despite knowing that beliefs regarding death can differ between these ethnic groups.27 More cases over a longer time period would be required to overcome this limitation. While this would increase total ethnic minority patient numbers, it would also introduce underlying period and migration cohort factors. Other than the modelling, our analysis has been conducted on the more disaggregated findings presented in table 2.
Even when the ethnic minority groups were combined, the number of events was low, increasing the risk of a type II error (failure to reject a null hypothesis) due to lack of statistical power. To some extent, we may have compounded this shortcoming by allowing for multiple testing in an effort to reduce the risk of a type I error (inappropriate rejection of a null hypothesis). Others might argue that this approach is too conservative; as recommended in these circumstances, relevant analyses are shown in full.28
It is generally accepted that some death certificate information is less accurate as a result of reliance on family members for information at a stressful point in time. Age at death and patient's occupation29 are two examples. Our approach to deriving age at death from age at census overcomes this accuracy concern; in addition, relatively broad age categories were used. As a result, any error rate is likely to be very low with accuracy potentially improved over what may have been reported for age at death on the certificate.
Information on the death certificate regarding location of death is coded by NRS based on an algorithm. The category ‘non-institutional location’ has been interpreted in this study as ‘home’, and in the case of cancer patients it is unlikely to include places such as ‘on mountain’, which may occur for individuals dying of, for example, myocardial infarction but are less relevant for cancer patients and therefore unlikely to affect results reported here.
A possible scenario that we have not been able to investigate is the so-called ‘salmon bias effect’, where foreign-born individuals may leave Scotland at the time of declining health to return to their home country to die in the presence and care of family left behind.30 One English study5 revealed strong preferences of Bangladeshi migrants for death and burial to occur in Bangladesh, particularly for recent migrants. A number of factors are likely to influence the decision of where to die. Given Scotland's healthcare is provided on the basis of need, not ability to pay, it is possible that free access to healthcare in Scotland may in part mitigate the possible effect of ‘salmon bias’. In the absence of any data on this phenomenon, as well as a very small minority ethnic population, we have not been able to quantify the potential impact, if any, on our results. In any case, the Scottish minority ethnic population is relatively settled and stable with a smaller proportion having recently migrated.16
Conclusions and policy implications
We conclude that cancer deaths occur most often in hospital, home and hospice in that order for all Scotland; differences in place of death by ethnic group found in England6 ,7 and internationally8 ,9 ,10 are limited in Scotland, in part reflecting very small ethnic populations, but also perhaps implying more equitable access to palliative care services regardless of ethnicity. However, regardless of socioeconomic indicator used, more affluent Scottish White patients were less likely to die in hospital; existing socioeconomic indicators detected no clear trend for the non-White population. Nevertheless, the dominant proportion of cancer deaths for all Scotland occurred in hospital (52.3%), demonstrating there is significant progress required to meet policy objectives of enabling death out with the acute hospital setting, reflecting the preferences of patients and carers alike.11 ,31
We thank the Chief Scientist's Office for a grant (CZH/4/432), NHS Health Scotland for a supplementary grant and the Equality and Diversity Information Programme of the Information Services Division (ISD) of NHS National Services Scotland for support. ISD and the National Records Scotland both made ‘in-house’ contributions to the work. Joan Jamieson (ISD) was co-investigator in the early stages and general advisor. Helen Brown was senior research fellow in the earlier stages of the study. Colin Fischbacher was co-principal applicant and Chairman of the Steering Committee. Chris Povey was a co-investigator and had the idea of linking the census data to the data held by ISD and performed most of the linkage work, including developing linkage methods. Jim Chalmers was a co-investigator and had the original idea for the use of one-way encryption. Ganka Mueller was a collaborator and was key in linking census data to health data.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online appendix
Contributors KHS initiated discussions regarding extending the Scottish Health Ethnicity Linkage Study to explore ethnicity and place of death from cancer, defined analysis plan, coordinated and provided input to the statistical modelling design, attended the NRS safe haven to review preliminary results to shape future iterations, performed further analysis on hospice bed provision for country comparisons and prepared all drafts of the manuscript and submitted the final agreed version, having reviewed and approved the final version. KHS had full access to all the data in the study. GC advised on the statistical analysis, performed all statistical analyses at the NRS safe haven, prepared presentations of tables, has reviewed and commented on interim and final drafts of the manuscript and has approved the final version. GC had full access to all the data in the study. NB facilitated extension of the Scottish Health Ethnicity Linkage Study approvals to include place of death from cancer and ethnicity, reviewed statistical analyses performed, reviewed and commented on interim and final drafts of the manuscript and has approved the final version. NB had full access to all the data in the study. RSB initiated the Scottish Health Ethnicity Linkage Study, reviewed and commented on the drafts and final version of the manuscript for critically important intellectual content and interpretation and approved the final version. RSB had full access to all the data in the study and final responsibility for the decision to submit for publication and is co-guarantor. DHB reviewed and commented on the drafts and final version of the manuscript for critically important intellectual content and interpretation and approved the final version. DHB had full access to all the data in the study and final responsibility for the decision to submit for publication and is co-guarantor.
Funding This work was supported by the Chief Scientist's Office (grant number CZH/4/432), a supplementary NHS Health Scotland grant and support from Equality and Diversity Information Programme of the Information Services Division (ISD) of NHS National Services Scotland.
Competing interests None.
Ethics approval Multicentre Research Ethics Committee for Scotland and Privacy Advisory Committee of the NHS National Services Scotland.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Due to the strict data protection and organisational procedures required to ensure confidentiality of the data of those who participated in the census, as well as the sensitive nature of health data, sharing the data used in this study is likely to be limited if at all possible. Safe haven procedures restrict access to named investigators who have been approved by the data guardians, in this case National Records Scotland and NHS National Services Scotland; for these individuals access to data was limited to the secure standalone computer located in NRS premises. The data may be open to researchers to apply to access the data through the lead investigator (Prof. Bhopal), but this would also depend on the data guardian approvals and would also require a safe haven approach.