Evidencing the gap between health expectancy and life expectancy for ethnic groups in Scotland

Background: Recent evidence has shown that ethnic minorities live longer than the majority population in Scotland. This mortality advantage in ethnic minorities is not unique  to  Scotland. However, whether morbidity  patterns by  ethnicity  align with mortality patterns by ethnicity is unknown. Thus, this study explores ethnic differ‐ ences in health expectancies (HE) in Scotland and contrasts HE with life expectancy (LE) findings. Methods: The  Scottish  Health  and  Ethnicity  Linkage  study  anonymously  links  the Scottish Census 2001 for 4.6 million people to mortality records. The Scottish Census 2001 collected  two measures of  self‐assessed health,  self‐declared ethnicity,  age, and sex. Utilising the life tables used to calculate life expectancy by ethnicity and sex in Scotland, the Sullivan method was employed to calculate two measures of health expectancy (healthy life expectancy and disability‐free life expectancy) by ethnicity and sex. 95% confidence intervals were calculated to detect significant differences compared to the majority White Scottish population, taken as reference. Results: Longer health expectancies were found in males and females of Other White British, Other White,  and Chinese origins  as well  as  in  Indian males  compared  to White Scottish populations. Any Mixed Background and Pakistani populations had the  shortest  healthy  life  expectancies.  Patterns of health  expectancy by  ethnicity mostly aligned with patterns of life expectancy by ethnicity with the clear exception of the Pakistani population who showed among the longest life expectancies with the  shortest health expectancies. Contrasting HE with  LE  findings,  the number of years  in an unhealthy  state was greater  in  females  than  in males  for each ethnic group. In relation to ethnicity, Pakistani and Indian populations had the highest num‐ ber of years in an unhealthy state in Scotland. Pakistani females showed the strong‐ est disadvantage in this respect. Geneviève Cézard 137 Conclusion: Pakistani populations had the shortest health expectancies contrasting with the longest life expectancies in Scotland. Future research should aim to under‐ stand why such a discrepancy occurs while policy makers ensure that fair and adapt‐ ed care is provided to offer better quality of life for the most vulnerable.

Population and Health Research Group (PHRG), School of Geography and Sustainable Development (SGSD), University of St Andrews, UK.

Introduction
Longer life expectancies have been found in most minority ethnic groups compared to the majority White Scottish population in Scotland (Gruer et al., 2016). Assuming greater deprivation in ethnic minorities compared to the majority population, one might not expect minority populations to fare so well in this regard. However, the Scottish context has its specificities and some evidence suggests that ethnic minorities do not necessarily experience lower socio-economic conditions compared to the majority in Scotland (Walsh, 2017). Furthermore, evidence in the Scottish context supports lower all-cause mortality in ethnic minorities beyond socio-economic differences (Bhopal et al., 2018). This mortality advantage phenomenon in ethnic minorities relative to the majority population is not unique to Scotland. For example, a mortality advantage has been shown in Hispanic populations in the US comparative to their low socio-economic status, often referred to as the «Hispanic mortality paradox» (Abraído-Lanza et al., 2005;Arias et al., 2010;Elo et al., 2004;Markides, Eschbach, 2005;Ruiz et al., 2013). In Europe, there is also a growing body of evidence on a «migrant mortality advantage» although with variation across countries and migrant groups (Aldridge et al., 2018;Anson, 2004;Brahimi, 1980;Deboosere, Gadeyne, 2005;Gadd et al., 2006;Ikram et al., 2016;Khlat, Darmon, 2003;Khlat, Guillot, 2017;Marmot et al., 1984;Moncho et al., 2015;Syse et al., 2018;Uitenbroek, Verhoeff, 2002;Wallace, Kulu, 2015). The reasons for this phenomenon are not fully understood. Key explanations offered as underlying the mortality advantage observed in Hispanics in the US and in specific migrant groups in Europe range from health selection hypotheses such as the «healthy migrant effect» (positive in selection) and the «salmon bias» (negative out selection) (Abraido-Lanza et al., 1999;Norredam et al., 2015;Rubalcava et al., 2008;Tarnutzer, Bopp, 2012;Turra, Elo, 2008;Ullmann et al., 2011;Wallace, Kulu, 2018), culturally protective health behaviours (Abraído-Lanza et al., 2005;Fenelon, 2013) and data artefacts (Elo et al., 2004; Nevertheless, this mortality advantage in specific ethnic minorities and migrant groups in varied contexts is particularly surprising as evidence based on general morbidity points to a disadvantage in some of those same groups in Scotland (Scottish Government, 2004Whybrow et al., 2012), in the UK (Becares, 2013;Darlington et al., 2015;Sproston, Mindell, 2006) and other high income countries (Hayward et al., 2014;Lindström et al., 2001;Solé-Auró, Crimmins, 2008). For example, Pakistani women have the longest life expectancy in Scotland (Gruer et al., 2016) while initial general morbidity evidence from Scottish government reports points to poorer reported health in Pakistani populations (Scottish Government, 2004. This discrepancy in the evidence related to morbidity and mortality patterns should direct researchers to question whether living longer aligns with living longer healthily. Furthermore, to reduce health inequalities and ensure better quality of life for all, there is a need to better characterise ethnic inequalities in morbidity and mortality simultaneously which can then inform policy makers and feed into planning appropriate care services for the most vulnerable. Some researchers have started to gather evidence and explore more generally whether the mortality advantage observed in specific populations aligns with their morbidity patterns. So far, a morbidity-mortality contrast appears in Greek and Italian populations in Australia (Kouris-Blazos, 2002;Kouris-Blazos, Itsiopoulos, 2014;Stanaway et al., 2018) and in Mediterranean migrants both in Belgium (De Grande et al., 2014;Deboosere, Gadeyne, 2005) and in France (Khlat, Guillot, 2017). In these studies, the observed mortality patterns and morbidity patterns were mostly gathered from different sources and did not necessarily refer to the same individuals. Some recent studies have also explored differences in healthy life expectancy between a specific ethnic or migrant group and the majority population (Carnein et al., 2014;Hayward et al., 2014). Although their aim was not necessarily to identify a morbidity-mortality contrast, these studies showed that both Turkish migrants in Germany and Hispanics in the US had similar or longer life expectancy compared to the majority population while showing more years with limitations or disability.
A systematic literature review of studies investigating health expectancies (HE) inequalities in older ages found that HE measures were used to compare the health of different subpopulations using variables such as gender, social class, race/ethnicity and health behaviours (Pongiglione et al., 2015). The review highlighted that evidence related to the study of HE differences by ethnic or race groups came uniquely from the US context. In addition to studies identified in the review, a few recent European studies have explored HE by ethnic or migrant groups (Carnein et al., 2014;Reus-Pons et al., 2017;Wohland et al., 2015). Thus, evidence of ethnic inequalities in HE remains limited worldwide. Most of this body of research used the Sullivan method. This method requires cross-sectional data which can combine morbidity and mortality from different sources. Indeed, available studies of HE by ethnicity gathered data on morbidity and mortality from different data and population sources. Mortality data came from official life tables or death registry while health data came primarily from surveys in both the US and European studies.
This study aims to identify whether a morbidity-mortality contrast appears in specific minority ethnic groups in Scotland. Identifying those who are living longer but in poorer health is key to planning for adapted care and services. This paper uses the Sullivan method to calculate two health expectancy (HE) measures: healthy life expectancy (HLE) and disability-free life expectancy (DFLE). HE measures by ethnicity enable us to assess how long ethnic groups are expected to live in good health. HE estimates are also useful if contrasted with life expectancy (LE) estimates. Doing so will provide us with a quantification of the contrast between reported health and mortality.
Previously published LE estimates by ethnicity in Scotland were derived from the Scottish Health and Ethnicity Linkage Study (SHELS) (Gruer et al., 2016). The HE calculation in this study uses the same data source, SHELS, since it holds individual level data from the Scottish Census 2001, containing two measures of self-assessed health (SAH), linked to mortality data. Thus, SHELS contains both reported morbidity and mortality data for 4.6 million people who responded to the Scottish Census in 2001. SAH is widely used and accepted as a reliable measure of health due to its strong association with other health indicators such as measures of physical and mental health, physician ratings of health, health care usage and mortality (Cohen et al., 1995;Idler, Benyamini, 1997;Idler, Kasl, 1995;Larue et al., 1979;Miilunpalo et al., 1997;Mossey, Shapiro, 1982;Wannamethee, Shaper, 1991). SAH has also been shown to be consistently associated with other measures of morbidity across ethnic groups in the UK setting (Chandola, Jenkinson, 2000).
As well as addressing hitherto unanswered questions about ethnic differences in contrasting morbidity-mortality patterns in Scotland, this study makes methodological contributions. Gathering data on health, mortality and ethnicity at the national level to enable health expectancy calculation by ethnicity is a challenge. In England and Wales, one study investigated 140 health expectancies by ethnicity but HE estimations were based on an indirect method of mortality estimation by ethnicity due to the absence of data gathering mortality and ethnicity together (Wohland et al., 2015). In the US and Europe, a few studies have calculated HE by ethnicity (Carnein et al., 2014;Hayward et al., 2014) or by migrant status (Reus-Pons et al., 2017) but they used different samples and sources to gather morbidity and mortality data. Thanks to the linkage of death records to census data at a national level in the SHELS data, this is the first time that health expectancies by ethnicity have been calculated using a direct method and with morbidity and mortality data based on the same population source.
The research questions are as follows: -What are the magnitude and direction of ethnic differences in health expectancies based on a direct method using individually linked data? -Do patterns of ethnic differences in health expectancies differ from patterns of ethnic differences in life expectancy in Scotland (i.e. is there a morbiditymortality paradox)?
In line with initial evidence of ethnic differences in SAH in Scotland (Scottish Government, 2004Government, , 2015, ethnic differences in HLE and DFLE are expected to go in both directions compared to the majority White Scottish population. For example, evidence based on the Scottish censuses of 2001 and 2011 consistently showed comparatively worse reported health in the Pakistani population and better reported health in the Chinese population. Therefore, we might expect to find shorter health expectancies in the Pakistani population and longer health expectancies in the Chinese population compared to the White Scottish population. If such a disadvantage in HLE and DFLE is found in the Pakistani population in Scotland, it will contrast with their LE advantage and translate into more years in an unhealthy state. Disaggregation is preferred to disentangle the patterns and specificities of each ethnic group in the most refined way possible. However, when SAH was combined with mortality to produce health expectancies (HLE and DFLE), some ethnic groups were aggregated due to a low number of deaths. The HE analysis shared the same constraint of low death events for some groups as the LE published results (Gruer et al., 2016). Consequently, ethnic groups followed the same aggregation process. Bangladeshis were grouped with Other South Asians. Caribbean, Black African and Black Scottish and Other Black were combined into one group, labelled «African Origin». The results for the «All Other Ethnic Groups» are not reported due to the heterogeneity of this category.

Self-Assessed Health measures
The Scottish Census 2001 collected two SAH indicators: Self-Reported Health (SRH) and Limiting Long Term Illness (LLTI). The question related to SRH in the Scottish Census 2001 was as follows «Over the last twelve months would you say your health has on the whole been» with the opportunity to answer «good», «fairly good» or «not good». The question related to LLTI asked: «Do you have any long-term illness, health problem or disability which limits your daily activities or the work you can do? Include problems which are due to old age». This question could be answered with «yes» or «no». SRH (categories «good» and «fairly good» combined) and LLTI («no») were used to calculate the rate of good health which was included into an augmented life table to estimate health expectancies. Healthy life expectancy was estimated using SRH and disability-free life expectancy using LLTI.

Life table
Life expectancy by ethnicity was previously published and its method of calculation explained (Gruer et al., 2016). It followed the Office of National Statistics recommendations in relation to life expectancy estimation for small populations using 3 years of data (Toson, Baker, 2003 (Jagger et al., 2006). In addition to LE at birth, results were also presented at 65 years of age to assess whether ethnic inequalities are consistent in older ages.
To provide a better picture of potential discrepancies between LE estimates and HLE/DFLE estimates, the number of years in poor health (LE minus HLE) and the number of years lived with disability (LE minus DFLE) and their CIs were calculated for each sex and ethnic group. The proportion of years lived in good health (HLE divided by LE) and the proportion of years lived without disability (DFLE divided by LE) were also provided as an indication of relative length of life lived in a healthy state. Figure 1 shows HLE and DFLE estimates at birth with 95% CI by ethnicity and sex. Precise estimates are also available in Table 1, in column 4 for HLE at birth and column 5 for DFLE at birth. In the White Scottish population (taken as reference: dotted line in Figure 1)   Contrary to published LE estimates at birth where most minority ethnic groups were expected to live longer than the White Scottish majority (Gruer et al., 2016), ethnic patterns in HLE and DFLE showed differences going both ways. As expected the Chinese population experienced an advantage in health expectancies and the Pakistani population a disadvantage. Shorter health expectancies were also found in White Irish and Any Mixed Background males compared to White Scottish males. The use of DFLE (but not HLE) revealed a significant disadvantage (shorter disability-free life expectancy) in Any Mixed Background and Indian females compared to White Scottish females.  In summary, patterns of ethnic differences in LE at birth and in HLE and DFLE at birth are mostly consistent. However, Pakistani males and females and Indian females had among the longest LE at birth contrasting with the shortest HLE and DFLE at birth.  Evidencing the gap between health expectancy for ethnic groups in Scotland

FIGURE 2
Number of years in poor health or with disability by ethnicity and sex

Ethnic inequalities in LE and HLE/DFLE at 65 years
This section considers whether the estimates of LE and HE at 65 years are consistent with findings described previously using estimates at birth. In line with patterns at birth, results at 65 years (Table 2)  Such a LE advantage in older ages in minority ethnic groups compared to the majority ethnic group of Scotland again contrasted with health expectancies findings. A significant disadvantage in both HLE and DFLE was found in Indian and Pakistani females. Chinese males and females no longer showed an advantage in DFLE at older ages but had similar levels of DFLE compared to the White Scottish population.
In line with the findings at birth, the quantification of the contrast between LE and HLE/DFLE at 65 years highlighted a disadvantage in males and females of Pakistani and Indian origins. At 65 years, Pakistani females had the highest estimated number of years in an unhealthy state with 11.4 years in poor health and 17.9 years with disability compared to, respectively, 4.4 years and 10.3 years in White Scottish females. A disadvantage also emerged in Chinese females. The findings showed that Chinese females had longer LE at 65 years (21.6 years) along with similar DFLE at birth to that of White Scottish females, and thus live more years with disability (13.6 years).

Discussion
This study has shown ethnic differences in health expectancies in Scotland in both directions compared to the majority White Scottish population: an advantage was observed in males and females of Other White British, Other White and Chinese origins as well as in Indian males, while Any Mixed Background and Pakistani males and females had a clear health expectancy disadvantage. Overall, patterns of ethnic differences in HLE and DFLE at birth were in line with previously published life expectancy patterns by ethnicity in Scotland (Gruer et al., 2016).
However, the findings of this paper have highlighted a clear contrast between the LE advantage and the corresponding HLE and DFLE patterns for Pakistani and Indian populations. The number of years lived in poor health and with disability were greater in females than males for all ethnic groups. If considered with ethnicity, results were particularly concerning for Indian and Pakistani populations, especially for females: they were living longer but in poorer health. The worst outcome was found in Pakistani females: 20.4 years lived in poor health, 30.7 years with disability. Results at birth and at 65 years consistently showed that individuals spent longer periods in an unhealthy state in Pakistani and Indian populations compared to the majority White Scottish population.
Advocating the usefulness of healthy life expectancy as a measure of health inequalities (beyond the sole use of life expectancy), Scotland has been producing HE figures for the last few decades (Scottish Government, 2019;Wood et al., 2006). Based on multiple sources, the latest figures produced and released by the Scottish Government report HE by sex, Scottish council areas, health boards and area-level deprivation (Scottish Government, 2019). However, monitoring progress in tackling health inequalities requires the investigation of various dimensions of social inequalities. No measure of healthy life expectancy by disability, ethnicity, origin, religion or individual socio-economic circumstances is available in the Scottish context. Hence, this study is the first to provide HLE and DFLE estimates by ethnicity in Scotland. It also contributes to bridging the gap in quantifying health inequalities between social subgroups in Scotland.
Indeed, a major strength of this study is that it was based on the national sample size of SHELS linking reported morbidity to mortality data at an individual level. Therefore, the calculation of health expectancy by ethnicity was based on a direct method, gathering data from the same cohort of individuals and at a national level. Analysis was done on 10 ethnic groups in Scotland with a fine ethnic granularity. Due to low number of deaths, smaller ethnic groups were aggregated into Other South Asian and African origin groups. No difference in HE were found for these aggregated ethnic groups in comparison to the White Scottish population, potentially hiding divergent HE patterns for their subpopulations. LE and HE estimates could be directly contrasted by ethnicity which allowed the investigation of whether some ethnic groups have longer lives in worse health.
The unique strength of the data and methods of this paper has enabled it to offer clear initial evidence of a Pakistani morbidity-mortality paradox in the Scottish context. Pakistani populations lived longer but had rela-tively shorter healthy lives and consequently had more years in an unhealthy state than the White Scottish majority. This phenomenon was also seen in Indian populations in Scotland although to a lesser extent. This is the first time that evidence of such a morbidity-mortality discrepancy in particular ethnic groups has been demonstrated at a country level based on the same cohort. Previous research focusing on the morbiditymortality gap in specific ethnic groups gathered evidence from different sources and population samples or were based on small sample sizes (Deboosere, Gadeyne, 2005;Khlat, Guillot, 2017;Kouris-Blazos, 2002;Stanaway et al., 2018).
Doubts can be expressed about the accuracy of the morbidity and mortality data.
Firstly, the observed mortality advantage in most ethnic groups in Scotland could be due to a data artefact, the effect of selective moves on health (e.g. unhealthy return migration) and unrecorded emigration and death. For example, migrants, when getting older and sick, could decide to migrate back to their country of origin to finish their lives back home with their relatives. This unhealthy return migration phenomenon, often referred to as the «salmon bias», if combined with unrecorded emigrations and deaths, could bias mortality estimates by producing a mortality advantage where there is none. However, a salmon bias hypothesis seems unlikely in Pakistani and Indian populations in Scotland. In the SHELS cohort, around 60% of the Pakistani and half of the Indian populations were born in the UK, highlighting the well-settled nature of these populations. Furthermore, these ethnic groups have a strong sense of national belonging in the UK. Indeed, more than half of people from ethnic minorities in the UK describe their national identity as some form of UK identity with the highest proportions seen in South Asian and Black Caribbean populations (up to 84% in Pakistani populations) (Jivraj, Simpson, 2015). National Health Services in Scotland also offers health care services that are «free at the point of use». Hence, if South Asians living in Scotland are well-settled and can access free health care services when becoming ill, they are likely to stay to benefit from the health care they need. If most of their family and descendants are also settled in the UK, it would reduce even more their likelihood of returning to their country of origin in older ages. Little is known about the prevalence of the salmon bias in the UK and its contribution to the observed mortality advantage in ethnic minorities. One recent study showed that there was an unhealthy return migration phenomenon for specific ethnic groups in the UK but this phenomenon was not strong enough to explain the mortality advantage observed in South Asian populations in the UK (Wallace, Kulu, 2018).
Secondly, the reliability of the reported morbidity data can be challenged. One could argue that there can be language barriers or cultural differences in the reporting and meaning of health. In that case, we could assume that SAH as a subjective measure of health might not accurately reflect the objective health status of minority ethnic groups. Hence, such a reporting bias could result in biased morbidity estimates and possibly lower-than-expected HE estimates. Based on the Health Survey for England, Chandola and Jenkinson showed that worse reported health was associated with greater morbidity in all ethnic groups and for a range of more objective measures of health (Chandola, Jenkinson, 2000). Their analysis also supported no differential association between reported health and more objective morbidity across ethnic groups. Their findings supported a strong and consistent association between subjective morbidity and more objective morbidity across ethnic groups. However, on account of the present findings, further research is warranted to support claims of the consistency of the subjective-objective health link between ethnic groups in the Scottish context.
If we assume that a salmon bias cannot explain the mortality advantage in Pakistani and Indian populations in Scotland and that their reported morbidity reflects their objective health status, we could argue that there is a real morbidity-mortality paradox (i.e. living longer but in poorer health) in theses populations. Higher morbidity in certain groups could be due to differences in access and quality of care. Although, evidence from the Health Survey for England does not suggest unequal access to GP services for minority ethnic groups (Nazroo et al., 2009), available evidence both in England and Scotland suggests a more complex picture of potential unequal access operating at different level of healthcare and healthcare settings (Katikireddi et al., 2018;McFarland et al., 1989;Nazroo et al., 2009;Worth et al., 2009). Morbidity inequalities are less likely to be solely due to an ethnic differential in health services engagement. Nevertheless, this is an area for policy to consider. To explain the morbidity-mortality contrast, the sex morbidity-mortality paradox literature offers additional alternatives. Some groups could suffer from specific conditions that contribute to reporting higher morbidity as well as have lower risk of mortality (Case, Paxson, 2005). In the case of the Pakistani populations in the UK, they have a particular disease profile including higher risk of metabolic syndrome related diseases such as diabetes, renal disease and cardiovascular disease (CVD) (Bansal et al., 2013;Bhopal et al., 2011;Dreyer et al., 2009;Forouhi et al., 2006;Hull et al., 2011;Sproston, Mindell, 2006), higher risk of respiratory disease such as asthma (Sheikh et al., 2016) and lower risk of cancer (Bhopal et al., 2012) in comparison to their white counterparts. This disease profile (particularly their higher risk of CVD) does not fully fit with the expected set that contributes to higher morbidity and lower mortality. However, an emerging literature in the UK shows that South Asian populations, once diagnosed with either diabetes, renal disease or CVD survive longer than their white counterparts diagnosed with the same disease (Bansal et al., 2013;Davis et al., 2014;Mathur et al., 2018). In relation to the sex morbiditysurvival paradox, Oksuzyan and colleagues highlight that potential mechanisms can be complex and multifactorial including biological, social and psychological (Oksuzyan et al., 2018;Oksuzyan et al., 2008). Understanding mechanisms will require further research along these avenues.
Policy makers should aim to improve the quality of life of Pakistani and Indian populations in Scotland and ensure that fair and culturally-adapted care is provided in primary and secondary care settings while the root causes of this paradox are pinpointed and better understood. Further research should investigate the underlying mechanisms of the morbiditymortality contrast observed and aim for a better characterisation of the diseases that drive a morbidity disadvantage but do not necessarily lead to worse survival rates.
me with advices and feedbacks on my research and reviewed this manuscript prior to submission. Mary Carr provided English language support and proofread the manuscript. This paper including analysis and interpretation is my responsibility alone.
The SHELS study was approved by the Multicentre Research Ethics Committee for Scotland (REC 13/SS/0225) and the Privacy Advisory Committee of NHS National Services Scotland (PAC 36/13). The present work was conducted in addition to the initial SHELS research plan. Therefore, further ethical approval was sought and approved by the University Teaching and Research Ethics Committee, School of Geography and Geosciences, University of St Andrews in August 2017 (reference GG13084). An amendment to PAC 36/13 was also submitted and approved by the Public Benefit and Privacy Panel for Health and Social Care in November 2017. Individual consent was not sought. The anonymously linked SHELS data were accessed in a secure environment thanks to the Administrative Data Research Network (PROJ-208). Appropriate security clearance was required to carry out the analyses.