Antithrombotic medications (antiplatelets and anticoagulants) reduce the risk of cardiovascular disease (CVD), but with the disadvantage of increasing bleeding risk. Bleeding risk equations have been developed so that clinicians can balance the potential cardiovascular benefits of antithrombotic medication against their bleeding harms. We recently developed such an equation for people without CVD or atrial fibrillation (AF) to guide decision-making on the use of aspirin for the primary prevention of CVD [1, 2]. We found that ethnicity and socioeconomic status, in addition to clinical variables, were important independent predictors of major bleeds among women and men.

Neither ethnicity nor socioeconomic status are specified as factors associated with increased bleeding risk in US guidelines for the secondary prevention of CVD [3] or management of AF [4]. Bleeding risk is of particular importance among people with AF given the widespread use of anticoagulants to reduce the risk of thromboembolic disease, especially stroke, in this group. The purpose of this study was to determine whether ethnicity and/or socioeconomic status are also independent risk factors for a major bleed among people with a history of CVD or AF.

A total of 523,064 people who had a PREDICT CVD risk assessment were considered for the analysis (Fig. 1). Of these, 6,903 people were excluded because their first risk assessment was prior to 2007. Among the remaining 516,161 people, 28,054 were excluded because they met at least one of the exclusion criteria.

Among the remaining cohort of 488,107 people, 56.4% were men and the median age was 55 years (IQR 47–63). Although the majority were European (272,499, 55.8%), there were large numbers in the other ethnic groups: 64,420 (13.2%) Māori, 60,905 (12.5%) Pacific, 41,505 (8.5%) Indian, 28,948 (5.9%) Chinese and 19,830 (4.1%) Other Asian. There was a higher proportion of people living in areas of the highest (22.3%) and lowest (21.6%) quintiles of socioeconomic deprivation, than those living in intermediate quintiles (second quintile 19.4%, third quintile 18.0%, fourth quintile 18.7%).

We identified a total of 15,212 (3.1%), 43,790 (9.0%), and 429,105 (87.9%) people with AF (± CVD), CVD (no AF), and no CVD or AF, respectively (Table 1). There was a greater proportion of men, Europeans and Māori in those with AF or CVD compared with those without either. Those with AF or CVD were also older and more likely to be in the most deprived quintile of socioeconomic deprivation than those without either. The median follow-up (5.0 years, IQR 3.8–5.0) was similar across subgroups. During follow up there were a total of 9,873 bleeds (AF 1,218, CVD 2,105, no CVD/AF 6550), of which 6736 were gastrointestinal, 1384 were intracranial and 2024 were other bleeds. Cumulative incidence of bleeds (total and by type) for each of the subgroups is plotted in eFigure 1.

The proportionality assumption and linearity of the association between continuous variables (age, SBP, TC:HDL) and the outcomes were assessed for the three main models (outcome = all bleeds) and the nine models in the sensitivity analysis (outcome = gastrointestinal bleeds, intracranial bleeds, other bleeds). The proportionality assumption largely held for all variables across each of the models. While the association between age and TC:HDL and the outcomes was largely linear, there was some non-linearity in the association between SBP and the outcomes in some of the models at 140 mmHg, and therefore SBP was dichotomised for all models (< 140, ≥ 140 mmHg).

The adjHRs for ethnicity and socioeconomic deprivation in all of the models (main and sensitivity) are provided in Tables 2 and 3, respectively, and the adjHRs for all variables in each of the subgroups are provided in Additional file 1: Table S4 (main) and Additional file 1: Tables S5–S7 (sensitivity). The adjHRs for ethnicity in all of the models are also presented as forest plots in eFigure 2. In all three subgroups (AF, CVD, no CVD or AF), Māori (adjHR 1.63 [1.39–1.91], 1.24 [1.09–1.42], 1.57 [95% CI 1.45–1.70], respectively), Pacific people (adjHR 1.90 [1.58–2.28], 1.30 [1.12–1.51], 1.62 [95% CI 1.49–1.75], respectively) and Chinese people (adjHR 1.53 [1.08–2.16], 1.15 [0.90–1.47], 1.13 [95% CI 1.01–1.26], respectively) were at increased risk of a major bleed compared to Europeans, although for Chinese people the effect did not reach statistical significance in the CVD subgroup (Table 2). For Māori and Pacific people there was a fairly consistent increase in bleeding risk compared with Europeans across the three subgroups for each of the three bleed types. In contrast, for Chinese people the increase in bleeding risk in the three subgroups was mainly due to an increased risk of intracranial bleeds compared with Europeans. There was also an increased risk of intracranial bleeds among Other Asian people and, in the CVD and no CVD/AF subgroups, among Indian people.

Increasing socioeconomic deprivation was associated with increased risk of a major bleed in all three subgroups (adjHR 1.07 [1.02–1.12], 1.07 [1.03–1.10], 1.10 [95% CI 1.08–1.12], respectively, for each increase in socioeconomic deprivation quintile) (Table 3). There was a fairly consistent increase in bleeding risk with increasing socioeconomic deprivation across the three subgroups for each of the three bleed types.

We found that, after adjusting for multiple clinical variables, ethnicity and socioeconomic deprivation were independent predictors of major bleeds in those with AF, CVD and without either condition. Māori and Pacific people were at increased risk compared with Europeans across most bleed types / subgroups, whereas Indian, Chinese and Other Asians were at increased risk of intracranial bleeds compared with Europeans across most subgroups. Increasing socioeconomic deprivation was also associated with a higher risk of a major bleed across most bleed types / subgroups.

Strengths of this study are its considerable size overall and the substantial size of the included ethnic and socioeconomic groups. The size of the cohort overall and its heterogeneity enabled us to adequately address the research question by simultaneously adjusting for multiple demographic and clinical variables in distinct clinical cohorts.

There are a number of limitations to this study. The study is reliant on coded diagnoses, rather than clinical diagnoses, to identify major bleeds. This definition, though informed by clinicians (including a cardiologist, AK) and previous literature, and used in our previously published work [1, 2, 17], has not been validated against source documentation for sensitivity in identifying bleeding events. While some major bleeds may have been missed or misclassified in this study due to the reliance on coded data, this limitation is unlikely to have a substantive effect on our findings because it is unlikely that coding varies by ethnicity or socioeconomic deprivation. Another limitation of the study is that it was restricted to patients who had their CVD risk assessed. However, in included practices, data up to 2015 indicate that a CVD risk assessment had been conducted in approximately 90% of eligible persons [5]. Finally, while participants were censored at death, competing risk of non-bleeding death was not taken into account when risk of bleeding was estimated.

While we have noted the large sample size as a strength, a potential disadvantage of large sample size is the increased likelihood of identifying statistically significant differences between groups which may not have clinical relevance. We consider the magnitude of the independent effect of ethnicity and socioeconomic status on bleeding risk (which has taken into account multiple other predictors) to be clinically, as well as statistically significant. For example, after taking all other predictors into account, and depending on the specific clinical subgroup, Māori were 24–63% and Pacific people 30–90% more likely to have a major bleed than Europeans, and a person living in an area with the highest quintile of socioeconomic deprivation was 35–50% more likely to have a major bleed than a person living in an area with the lowest quintile of socioeconomic deprivation.

This study was reliant on secondary use of patient data, i.e. data that had already been collected. Further, in this study, patient data had already been collected through the process of health service delivery, and without imposing any additional burden on patients, clinicians or health services. It would have been impossible within our resource constraints (and those of any other research or health provider group in this and probably other countries) to obtain informed consent from included participants given the size of the study, which as we have noted above, was needed because our analyses required simultaneous adjustment for multiple demographic and clinical variables in distinct clinical cohorts. A waiver was granted by the national ethics committee to enable this and other similar research to proceed with their permission without obtaining informed consent from participants, because the study involved secondary use of routinely collected patient data, and patient data were anonymized prior to being received by the research team.

Our findings are largely consistent with those of other New Zealand [18‐20] and international [21‐40] studies that have also found higher rates of major bleeds (total, gastrointestinal or intracranial) in non-Whites compared with Whites and with increasing socioeconomic deprivation. In these studies, populations were not defined according to their history of CVD and AF, and there was adjustment for a limited number of variables, if any. Studies of people with AF have shown this group to have increased risk of major bleeds with more deprived socioeconomic status [41], and increased risk of intracranial haemorrhage if they were Asian, Hispanic or Black (compared with White people) [42].

We identified two studies with findings inconsistent with our study [43, 44]. A systematic review of population-based studies of intracranial haemorrhage that pooled incidences in a random-effects binomial meta-analysis found that the incidence of intracranial haemorrhage was greater for East and Southeast Asian people than for White people, whereas there was no statistically significant differences in incidence for Black, Hispanic, Indian or Māori compared with White people [43]. There are a number of significant limitations to this study, including inconsistency in case definition and ascertainment across included studies and the inability to adjust for individual participant characteristics such as age because of a reliance on aggregated data [43].

The QBleed sex-specific risk equations for upper gastrointestinal and intracranial bleeds were developed among people in primary care irrespective of their history of CVD or AF, using multiple variables from routinely collected data from general practices in the United Kingdom [44]. Increasing socioeconomic deprivation was associated with an increase in bleeds, both upper gastrointestinal and intracranial. In QBleed there was no statistically significant difference in the risk of upper gastrointestinal bleeds between those in any of the ethnic groups (Indian, Pakistani, Bangladeshi, Other Asian, Caribbean, Black African, Chinese, Other [including mixed]) compared with the reference group (Whites or people in whom ethnicity was not recorded), with the exception of Black African men who were at lower risk of an upper gastrointestinal bleed than the reference group [44]. The following ethnic groups were at increased risk of an intracranial bleed: Bangladeshi women, Other Asian men, Caribbean men and Black African men. Chinese people were at increased risk of an intracranial bleed but the difference was not statistically significant. There are a number of reasons that might explain why differences by ethnicity were more consistently observed in our study than in QBleed. Ethnicity was missing from 26% of the QBleed participants, the reference ethnic group in QBleed combined Whites and those in whom ethnicity was missing, and the proportions in the non-reference groups was low (Indian 2.1%, Pakistani 1.0%, Bangladeshi 0.8%, Other Asian 1.4%, Caribbean 1.0%, Black African 1.8%, Chinese 1.1% and other or multiple 2.3%).

We consider that a plausible explanation for the finding of higher rates of major bleeds in non-Whites compared with Whites and with increasing socioeconomic deprivation to be that ethnicity and socioeconomic deprivation are surrogate markers for longitudinal, lifetime exposure to risk factors that are not able to be captured by other predictors either because these are cross sectional or not measured at all. Ethnicity itself has been directly associated with adverse health outcomes through racism [45, 46].

These findings indicate that clinical decision-making regarding the use of antithrombotic medication for the management of AF and CVD, as well as the primary prevention of CVD, should take ethnicity and socioeconomic deprivation into account. In order to be able to do this, clinicians need multivariable bleeding risk assessment tools. Clinicians need such tools to enable them to take into account ethnicity and socioeconomic deprivation at the same time as other independent predictors of bleeding risk. Ideally such bleeding risk assessment tools would be validated in the local population.

US guidelines for the secondary prevention of CVD recommend antiplatelet therapy in specific groups, largely irrespective of bleeding risk [3]. More recent US and European post-acute coronary syndrome guidelines [47, 48], have moved towards the use of multivariable equations to estimate the risk of bleeds, but none of these equations include ethnicity or socioeconomic status [49‐51]. Similarly, while US guidelines for the management of AF [4] recommend a multivariable equation to estimate the risk of bleeds, that equation [52] also does not include ethnicity or socioeconomic status.

Given the significant independent effects of socioeconomic status and ethnicity on major bleeds, it would be appropriate for future guidelines to consider both factors in bleeding risk assessments, ideally using multivariable equations, to guide the use of antithrombotic medication for the management of AF and CVD.

We found that ethnicity and socioeconomic deprivation were independent predictors of major bleeds in those with AF, CVD and without either condition. Clinical decisions regarding the use of antithrombotic medication for the management of AF and CVD, as well as the prevention of CVD, should take ethnicity and socioeconomic deprivation into account. Clinicians need locally validated multivariable bleeding risk assessment tools to enable them to take into account ethnicity and socioeconomic deprivation, along with other independent predictors, when considering and discussing the balance of benefits and harms of antithrombic medication with their patients.