Ethnic variations in asthma hospital admission, readmission and death: a retrospective, national cohort study of 4.62 million people in Scotland

We obtained ethical approval from the Scotland A Research Ethics Committee. We also obtained approvals from Scotland’s Privacy Advisory Committee and the Community Health Index Advisory Group, which are charged with considering the appropriate use of potentially identifiable patient data. Access to data, analyses and release of outputs followed a pre-specified protocol, which included rounding of all count numbers to the nearest five to ensure confidentiality and to prevent disclosure of any potentially identifying data. This included suppression of data if there was any risk that individual patients may be identifiable (n <6). All outputs were reviewed by a National Records of Scotland data disclosure committee.

The methods of the Scottish Health and Ethnicity Linkage Study (SHELS) have previously been described in detail and we therefore confine ourselves here to providing a brief overview of the procedures used [13]. In summary, we linked National Health Service (NHS) Scotland’s hospital discharge and death records (SMR01) to the 2001 census using probabilistic linkage methods creating a unique link between the encrypted Community Health Index and census number, thereby creating a national retrospective cohort. An overview of the linkage techniques is shown in Fig. 1. All personal identifiers were then removed from the linked dataset.

Ethnic categories were derived from the 14 groups in the 2001 Scottish Census (Box 1) and reported following the principles of the census (including capitalisation of ethnic group labels) [14]; when discussing the work of others, however, we used the labels assigned by authors in their publications. Given small populations, we combined Bangladeshis with Other South Asians and the Caribbean, African and Black Scottish or Other Black as an African origin group, thus giving rise to the nine categories of ethnicity that were studied and reported on.

We had three outcomes of interest: (1) first hospitalisation or death from asthma following inception of the cohort in May 2001, which was our primary outcome and is henceforth referred to as ‘first asthma events’ (group 1); (2) asthma readmission among survivors of group 1; and (3) asthma death among survivors of group 1. We identified hospitalisations and deaths for asthma during the period May 2001–April 2010 using relevant International Classification of Disease codes (i.e. ICD-9: 493 and 493.9, and ICD-10: J45 and J46). To take account of the chronic nature of asthma, we did not exclude those with admissions before May 2001 [15]. Our ‘first asthma events’ were therefore initial episodes within the cohort study, which may or may not correspond with the first ever episodes of hospitalisation.

Census data included age, sex, Scottish Index of Multiple Deprivation (SIMD), and country of birth (CoB). SIMD is an area-based socioeconomic measure that uses 38 indicators across nine domains including health to rank areas based on full postcode [16]. The SIMD is known to be associated with health outcomes whether it includes, or excludes, the health domains.

Our detailed analysis plan specified that, because of concerns around small numbers and the associated risk of disclosure, analysis of hospitalisation and death would only be separately undertaken if deaths were at least 20 % of the total numbers. We also pre-specified that if this was not the case, the first of either of these two events would be included, thereby ensuring that we included community deaths.

Using Poisson regression models with robust variance we calculated age-adjusted rates of first asthma events (i.e. outcome (1) specified above) per 100,000 person years and incidence rate ratios (IRR) with 95 % confidence intervals (CI) by sex, this being undertaken because of the known sex variations in asthma epidemiology and outcomes. The person-years-at-risk denominators were calculated after censoring for deaths, asthma-related hospitalisations and transfers of patients out of NHS Scotland (i.e. to other parts of the UK), as appropriate.

The age-adjusted sex-stratified analysis was the primary analysis. We then examined the effect of adjustment by socioeconomic variables and CoB separately and in combination. We explored the association between eight measures of socioeconomic status (i.e. individual educational level, household educational level, individual National Statistics Socioeconomic Classification, household National Statistics Socioeconomic Classification, SIMD quintile, car ownership, household tenure, and economic activity) and health outcomes following our published approach, and we selected SIMD (quintiles), which was available for everyone; in contrast, other relevant variables were not collected for younger and older age groups [17]. Our analysis used the quintile of socioeconomic deprivation. We adjusted for birth within or outside the UK to explore whether this altered IRRs (there were not enough events for stratified analysis by CoB and ethnicity).

We undertook survival analysis and calculated hazard ratios (HRs) for death or readmission with asthma following hospitalisation (outcomes (2) and (3) as specified above, censoring for death) using Cox proportional hazard models, checking the proportional hazards assumption with the graphical method using log-log (ln{-ln(survival)}) plots. Analysis was undertaken using SAS v9.3 (SAS Institute Inc., Cary, North Carolina, USA).

Scotland had an enumerated population of 4.9 million on census day (29 April 2001); 52 % of the population was female and 89 % was White Scottish (Table 1). We were able to link data on 4.62 million individuals (94.9 % of the eligible population), yielding 38,216,530 patient-years of data over the 9-year study period (Fig. 2). We successfully linked data on at least 85 % of people from each ethnic group of interest (Tables 1 and 2).

There was a total of 113,790 first asthma admissions (n = 67,080; 59 % in females) and 505 deaths with asthma diagnosis and no prior hospital admission for asthma (59 % in females). There were substantial ethnic variations in first asthma events in both males and females (Fig. 3). Age-adjusted incidence rates for a first asthma event in the total population were 256.6 (95 % CI, 208.1–316.3) per 100,000 person years in males and 338.1 (288.7–395.9) in females.

When compared to the White Scottish reference population, the highest age-adjusted rates were in Pakistani males (IRR = 1.59; 95 % CI, 1.30–1.94) and females (IRR = 1.50; 95 % CI, 1.06–2.11) and Indian males (IRR = 1.34; 95 % CI, 1.16–1.54), and the lowest were seen in Chinese males (IRR = 0.62; 95 % CI, 0.41–0.94) and females (IRR = 0.49; 95 % CI, 0.39–0.61) followed by ‘Other White’ males (IRR = 0.77; 95 % CI, 0.64–0.92) and females (IRR = 0.81; 95 % CI, 0.70–0.93). Further adjustment for SIMD and CoB did not substantially modify the observed patterns (Table 3).

There were 107,710 readmissions in the cohort; 40,075 of these readmissions were for asthma (37 % of all readmissions for asthma), with the majority of these occurring in females (n = 24,660; 61.5 %). Increased rates of readmission were seen in Any Mixed Background males (HR = 1.37; 95 % CI, 1.02–1.83) and African Origin females (HR = 1.56, 95 % CI, 1.11–2.20). Other than this, no clear patterns were apparent, this possibly reflecting the relatively small number of readmissions within most of the minority ethnic groups and the associated imprecision of the HR estimates obtained (Table 4).

In total, there were 1,340 asthma deaths recorded in those who survived the first asthma event. As death was an infrequent occurrence (0.4 % of first asthma events) data on survival were suppressed for most ethnic groups because of the possible risk of disclosure. The available data on Other White British males (HR = 0.69; 95 % CI, 0.45–1.06) and Other White British (HR = 0.93; 95 % CI, 0.70–1.23), White Irish (HR = 1.08; 95 % CI, 0.58–2.01) and Other White (HR = 1.27; 95 % CI, 0.63–2.55) females was imprecisely estimated.

This national investigation into risk of poor asthma outcomes has found substantial ethnic variations in the rates of hospital admission and death between ethnic groups in Scotland. When compared to the reference White Scottish population, people of South Asian descent (i.e. Pakistani, Indian and Other South Asians) had 20–50 % increased rates of hospitalisation from asthma, whereas people of Chinese origin had 30–40 % lower rates. Comparing the Chinese with the Pakistani populations revealed a striking 2–3-fold variation in rates of admission. However, despite accruing over 38 million person-years of data during the 9-year study period, we obtained only imprecise estimates of the rates of death and readmission in survivors, this reflecting the relative infrequency of these outcomes in many of the populations studied.

This is, as far as we are aware, the first national investigation into within-country ethnic variations in asthma outcomes and it has provided some of the first estimates of asthma outcomes in a number of sub-sections of the UK population (e.g. White Irish and Chinese) [11]. Additional key strengths of this work include the substantial number of patient-years of data available, which offered the possibility to study a large number of hospitalisations for asthma – the main focus of our study. Through linking to the 2001 Census we were able to simultaneously adjust for age, CoB and socioeconomic status [18]. The decision to stratify reporting by sex was important given the marked overall increased rate of admission, readmission and mortality in females. Furthermore, in keeping with best practice, our analysis strategy was written and agreed prior to data analysis.

Nonetheless, we acknowledge limitations, including the fact that we were not able to link data on the entire Scottish population, differential rates in linkage between ethnic groups which may have introduced bias, the fact that we were not fully able to take into account emigration from Scotland, that event numbers in some non-White ethnic groups were small (particularly limiting for survival analysis), and that, because analysis of data on small groups potentially risked disclosing identity, we were not able to report on mortality data for the majority of the groups studied. Missing data may also have introduced the possibility of bias. It is impossible to know how much missing data there was overall, but we believe that this is likely to be small. This missing data could have arisen for three possible reasons: (1) those who did not participate in the census (estimated <5 %); (2) those for whom it was not possible to link census and hospitalisation and mortality data (estimated <5 % at date of census, and increases with length of time from the census due to immigration and births); and (3) those admitted into private hospitals that did not then convey this information to the Information Services Division of NHS National Services Scotland (<1 %). We attempted to censor those who migrated, but it was not possible to identify the likely small proportion of people who were not registered with NHS Scotland who emigrated or those who were registered with NHS Scotland but who emigrated outside of the UK. Finally, we were only able to adjust for a limited number of potential confounders in the relationship between ethnicity and asthma outcome and, as discussed below, future work needs to consider the impact of wider organisational, care-related and behavioural factors which may help to explain these variations.

These findings confirm the poor outcomes previously noted for UK South Asians, but were not as marked, possibly reflecting the relatively high rate of hospitalisation in the reference Scottish White population and/or the reduced risk of bias resulting from the fact that we were able to study the entire population [11]. These data also reveal that not all ethnic minority groups experience poor outcomes when compared with the reference population. In particular, the rate of asthma admission in Chinese-origin males and females were markedly lower than that seen in the White Scottish population. Given that good outcomes in the Scottish Chinese population have also been noted for a range of other long-term conditions such as heart failure [19], there is a need to investigate whether this is due to Chinese populations having comparatively better health, lower use of health services, or a combination of both of these factors.

There is now a need to understand the reasons underpinning these striking variations in outcomes. Possibilities include biological differences in disease severity, the impact of migration (although we did not find any effect of adjusting for CoB) [18], and the greater prevalence of atopy noted in South Asian children in the UK [20]. Another potentially important explanatory consideration is differences in over-crowding with the associated risk of admissions triggered by viral upper respiratory tract infections. Differences in health professional management including access to and quality of primary care (such as vaccination rates, prescribing and use of supported self-management plans) [21] and cultural factors, including differences in personal beliefs, self-management (in particular, concordance with preventive therapies) and health-seeking behaviour, may also be important contributory factors [22]. US data have, for example, shown marked variations in the quality of ambulatory and emergency department asthma care, but comparable hospital-based care [23]. There is also a need to consider the possibility of systematic biases through investigating the impact of differential rates of linkage and emigration between ethnic groups [24].

Data on a number of potential explanatory variables are available in general practitioner records, and our plan is therefore to link the SHELS cohort with primary care data across Scotland. We have undertaken a feasibility pilot (data being prepared for publication) and other primary care data extractions [25], which shows that this is technically possible and that relevant data on, for example, smoking, vaccination and prescribing patterns can be obtained for future studies. A key priority for future work will be to investigate to what extent these variations in outcome can be explained by differences in lifestyle factors, health-seeking behaviours and/or the quality of primary care [21, 22, 26, 27]. Future work should also explore the possibility of interactions between ethnicity and socioeconomic status.

We do not have permission to share the underlying dataset as, due to security considerations, this needs to be analysed within the confines of a safe haven. Researchers should contact the principal investigator if they wish to explore the possibility of collaborative work.