key: cord-0936831-2du80q3c authors: Stafford, Mai; Knight, Hannah; Hughes, Jay; Alarilla, Anne; Mondor, Luke; Pefoyo Kone, Anna; Wodchis, Walter P.; Deeny, Sarah R. title: Associations between multiple long-term conditions and mortality in diverse ethnic groups date: 2022-04-01 journal: PLoS One DOI: 10.1371/journal.pone.0266418 sha: e36641a3078e5568fa4d239a7d23feef69fe4830 doc_id: 936831 cord_uid: 2du80q3c BACKGROUND: Multiple conditions are more prevalent in some minoritised ethnic groups and are associated with higher mortality rate but studies examining differential mortality once conditions are established is US-based. Our study tested whether the association between multiple conditions and mortality varies across ethnic groups in England. METHODS AND FINDINGS: A random sample of primary care patients from Clinical Practice Research Datalink (CPRD) was followed from 1(st) January 2015 until 31(st) December 2019. Ethnicity, usually self-ascribed, was obtained from primary care records if present or from hospital records. Long-term conditions were counted from a list of 32 that have previously been associated with greater primary care, hospital admissions, or mortality risk. Cox regression models were used to estimate mortality by count of conditions, ethnicity and their interaction, with adjustment for age and sex for 532,059 patients with complete data. During five years of follow-up, 5.9% of patients died. Each additional condition at baseline was associated with increased mortality. The direction of the interaction of number of conditions with ethnicity showed a statistically higher mortality rate associated with long-term conditions in Pakistani, Black African, Black Caribbean and Other Black ethnic groups. In ethnicity-stratified models, the mortality rate per additional condition at age 50 was 1.33 (95% CI 1.31,1.35) for White ethnicity, 1.43 (95% CI 1.26,1.61) for Black Caribbean ethnicity and 1.78 (95% CI 1.41,2.24) for Other Black ethnicity. CONCLUSIONS: The higher mortality rate associated with having multiple conditions is greater in minoritised compared with White ethnic groups. Research is now needed to identify factors that contribute to these inequalities. Within the health care setting, there may be opportunities to target clinical and self-management support for people with multiple conditions from minoritised ethnic groups. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 The number of people with multiple long-term conditions, or multimorbidity, is rising. Though there is no accepted international definition of multimorbidity [1] , recent large-scale studies using electronic health records in the UK estimate 23-27% of people have two or more long-term conditions [2, 3] . Multimorbidity has been consistently associated with poorer outcomes for patients, with risk of death increasing with each additional condition [4, 5] and some studies suggesting that the association may be multiplicative [6] , and more pronounced when conditions concern different body systems (complex multimorbidity) [7] [8] [9] [10] [11] though the association between higher risk of mortality and multimorbidity is weaker at older ages [12, 13] . Multimorbidity is also associated with higher use of health care [2] and reduced quality of life [14, 15] . Many health and care systems are designed to care for patients with single conditions, but there is growing recognition that if they are to improve outcomes for patients, health care systems must be adapted to address the challenge of multimorbidity [16, 17] . There is an established body of evidence that the prevalence of multimorbidity is socially patterned. Previous studies have demonstrated an association between the prevalence of multiple conditions and socioeconomic disadvantage in households [18] and local areas [2, 3, 19] . The prevalence of multiple conditions is also higher in some minoritised ethnic groups [20] . People from some minoritised ethnic groups are more likely to have experienced discrimination and multiple disadvantage over their life course, leading to an increased risk of experiencing material deprivation, living in deprived areas and an associated higher prevalence of downstream behavioural risk factors including smoking and obesity [20, 21] . Poorer experience of healthcare services has also been reported by some minoritised ethnic groups, and they are less likely than their counterparts in the majority population to report that they are able to manage their own health [22] . Whether minoritised ethnic groups experience disadvantage or discrimination over the course of their lives will differ between countries and over time, and as a result, evidence on mortality risk for minoritised ethnic groups varies. For example while excess deaths in Black populations in the US have remained high for many years [23] , in the UK, the link between ethnic minority status and mortality risk varies by cause of death [24] and migration status [25] . Research in the UK on this topic has been hampered due to lack of ethnicity data on death certificates and historically poor recording of ethnicity in medical records, though the latter has improved markedly in recent years. A previous study in the United States of America found that having multiple chronic conditions resulted in reduced life expectancy but the impact did not differ between African Americans and non-Hispanic White people [26] . Analysis of the Health and Retirement Study, on the other hand, found that Black and Hispanic Americans were more likely to have multisystem multimorbidity and more likely to die during follow-up compared with their White American counterparts [27] . We are not aware that this has been assessed in the UK context, though here studies have investigated ethnic differences in long-term survival for people with a single or index condition of interest. These point to the possibility of ethnic differences in survival across a range of conditions, for example lower two-year survival for Black women in England with breast cancer [28] and higher survival for people in London with unipolar depression from Black Caribbean, Black African, South Asian and Chinese ethnic backgrounds [29] . It is plausible that many of the factors contributing to high prevalence of multiple conditions may also contribute to poorer survival in minoritised ethnic groups once multiple conditions are established. Likewise, the potential for differences across ethnic groups in the association between long-term conditions and mortality may be greater where the organisation of health care and the recommended treatments and lifestyle changes are especially complex, as is the case with multiple conditions. The aims of our study were to estimate the mortality risk of having multiple conditions, assess whether this risk is seen for complex multimorbidity, and examine whether the magnitude or direction of these risks varies across ethnic groups, compared with people of White ethnicity living in England. Our sample was drawn from primary care records. Over 95% of the England population are registered with a general practice. A random sample of 600,000 adults (age 18 and over) was drawn from the Clinical Practice Research Datalink (CPRD Aurum [30] ). This research database of pseudonymised routinely collected primary care records captures diagnoses, symptoms, prescriptions, referrals and tests and includes over 40 million patients (13 million currently registered as of June 2021). CPRD Aurum comprises GP practices using the EMIS Web software (one of four main general practice IT systems in operation) that have agreed to contribute data. Eligible adults were in a CPRD practice on 1 st January 2014 (to ensure records were up to date at least one year before the study start), were alive and still registered at the study start on 1 st January 2015, and were eligible for linkage to Hospital Episode Statistics (HES) and Office for National Statistics mortality data. Data linkage was carried out by a Trusted Third Party, namely NHS Digital, the organisation with responsibility for standardising, collecting and publishing data from across the health care system in England. They were followed until the study end (31 st December 2019) or death if this was earlier and were censored if they left the CPRD practice or the practice stopped providing data to CPRD. The study was reviewed for ethical and methods content and approved by the CPRD team (eRAP protocol number 20_000239). Patients cannot be identified from CPRD so GPs are not required to seek individual patient consent. However, patients may opt out of having their patient information being shared for research purposes through the national data opt-out scheme. Survival time was calculated from 1 st January 2015 to death or censoring. The number of long-term conditions was counted at study start. We used a list of 32 physical and mental health conditions (S1 Table) that have previously been associated with higher mortality risk, poorer functioning, and requiring primary care input [2.3] . They were taken from a list of 37 conditions that had been both previously validated based on CPRD [31] and included in published code lists [32, 33] . This was repeated to calculate number of conditions at the study end or censoring date, which may be fewer than at study start as we allowed for three conditions (anxiety/depression, asthma and cancers) to resolve. Complex multimorbidity was defined as having three or more long-term conditions in three or more different body systems [34] (S1 Table) . Ethnic identity, usually self-ascribed, was obtained from SNOMED codes recorded by the GP or, where that was missing or incomplete (29.8%), from linked HES (Hospital Episode Statistics) records. Where multiple values of ethnicity have been recorded, we selected the modal value where this was unique, or the most recent value [35] . Categories from the England 2011 census were used in our analysis but we combined White British, White Irish and other White because these separate categories were not available in HES. Socioeconomic deprivation was captured by 2015 Index of Multiple Deprivation (IMD) decile in the patient's area of residence based on lower-level super output area boundaries. The analytical sample included those with complete data on sex, age (n = 0 excluded), ethnicity (n = 67524 excluded), or deprivation (n = 417 excluded). Excluded patients were younger, more likely to be men, over-represented in less deprived areas, and had fewer conditions (S2 Table) . The association between survival time and baseline number of long-term conditions was modelled using a multilevel Cox proportion hazards model with adjustment for baseline age (centred at age 50 to aid interpretation), sex and ethnicity. Number of conditions was included as a continuous variable after confirming its association with survival time was linear (S1 Fig). Interactions of age by ethnicity and age by number of conditions were included to improve model fit. Age was included as a continuous variable in all models due to small numbers of patients and deaths in some age group-ethnicity cells. We present hazard ratios estimated at age 50. A two-level model was used to allow for the clustering of patients within GP practices (model 1). We assessed model 1 for violations of the proportional hazards assumption. The association between sex and mortality hazard was found to depend on follow-up time (p = 0.004), with a marginally higher hazard for men after 2.5 years of follow-up. The association between baseline age and mortality hazard also depended on follow-up time (p = 0.02), with a marginally higher hazard with advancing age after 2.5 years of follow-up. However, the differences across follow-up time were small. Furthermore, allowing for time-varying estimates for sex and age did not alter the estimates for the main variables of interest (namely, ethnicity and number of long-term conditions) so we elected to present the simpler model without time-varying estimates. To examine whether the association between survival time and long-term conditions varied by ethnicity, we added ethnicity by number of conditions interaction terms (model 2). A likelihood ratio test was used to test the combined statistical significance of these interactions (model 2 vs model 1). Where this test was statistically significant at the 5% level, we also present mortality hazard ratios from ethnicity-stratified models. In these models, the reference group comprises people with no long-term conditions of the same ethnicity. We examined two possible factors that could explain survival differences across ethnic groups, if any were observed (model 3). We added number of long-term conditions at end of follow-up. Patterns and rate of long-term condition acquisition vary across ethnic groups [20, 27, 36, 37] . We also added socioeconomic deprivation. There is a well-established relationship between ethnic minority identity and greater socioeconomic deprivation, driven by longstanding structural factors that disadvantage people from minoritised ethnic groups in multiple domains including housing, education and employment. We hypothesised that any ethnic differences in the association between survival time and baseline number of conditions would be smaller in models that included number of long-term conditions at end of follow-up and deprivation. In sensitivity analysis, we repeated model 2 replacing baseline number of conditions with presence or absence of complex multimorbidity. Cancers, circulatory, endocrine and respiratory system conditions are leading causes of death [38] . We also repeated model 2 replacing baseline number of conditions with presence or absence of a condition in these body systems. During the 5-year follow-up period, 5.9% of patients died ( Table 1 ). The majority of patients were of White ethnic background (85.4%) and these were older and over-represented in less deprived areas compared with all other ethnic groups (S2 Table) . The unadjusted mean number of long-term conditions at baseline was highest in the White ethnic group (1.23) and lowest in the Chinese ethnic group (0.33). People of Chinese ethnicity also had the lowest mean number of long-term conditions and lowest prevalence of complex multimorbidity across all age groups (S3 Table) . The initial model (model 1) addressed the first objective, to assess whether number of longterm conditions is associated with mortality. This model is based on the assumption that the association between number of conditions and mortality is consistent across ethnic groups. Each additional long-term condition at baseline was associated with increased mortality. For example, the estimated hazard ratio (HR) for a man of White ethnicity at age 50 was 1.80 with two conditions and HR = 3.25 with four conditions, compared to their counterpart with no conditions (see Table 2 for hazard ratios estimated at age 50 and S4 Table for full set of estimates). A statistically significant negatively signed interaction for number of conditions by baseline age shows that the relative difference in mortality for those with more versus no conditions was smaller at older ages. Ethnicity was associated with mortality and this association depended on age. At all ages, the mortality rate was significantly lower for those of Indian or Chinese ethnicity than for those of White ethnicity. To address the second objective, interaction terms for ethnicity by number of conditions were added. Their addition improved model fit (likelihood ratio test 0.05