figure b

Introduction

Diabetes mellitus is one of the leading causes of mortality worldwide, accounting for more than 4 million deaths in 2019, representing 11.3% of the global mortality rate [1]. Compared with healthy individuals, people with diabetes have a shortened lifespan that is directly attributable to the development of macrovascular (including coronary artery disease, peripheral arterial disease and stroke) and microvascular (neuropathy, nephropathy and retinopathy) complications, resulting from hyperglycaemia, excess NEFA and insulin resistance.

Diabetes mellitus disproportionately affects racial/ethnic minorities (e.g. Hawaiian/Pacific Islander, Asian and Hispanic people), who commonly present with worse prognosis and higher mortality risk than their White counterparts. For instance, some racial/ethnic minority groups are more likely to develop coronary artery disease, peripheral arterial disease, retinopathy or end-stage renal disease (ESRD) [2], although White people with diabetes seem to be at higher risk for diabetic macrovascular complications than racial minorities [3]. In addition, the racial/ethnic distribution of obesity (highest among non-Hispanic Black people and lowest among Asian people) does not mirror that of diabetes, which is highest among Asian people and lowest among White people [4, 5].

Within this context, it is incumbent on practitioners to identify high-risk populations to target public health interventions that may help delay and/or prevent a considerable proportion of future diabetic complications. There remains, however, a paucity of evidence on the influence of race or ethnicity on future complications or mortality risk in people with diabetes, and current guidelines and position statements make no differentiation on its clinical management according to race/ethnicity [6]. A possible reason for this lack of consideration might be that the quantitative association of race/ethnicity for diabetic complications and mortality risk is not well established. Previous reviews have found higher mortality rates and higher risk of diabetes complications among diabetic individuals from ethnic minorities [2, 3]. However, much of the research up to now has been descriptive in nature (i.e. qualitative reviews). To address these critical evidence gaps, we aimed to quantify racial/ethnic differences for future diabetic complications and all-cause mortality by performing a meta-analysis of prospective studies.

Methods

The present systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and recommendations and was conducted following the 24-step guide design [7]. The study was registered in the International Prospective Register of Systematic Reviews (PROSPERO; registration no. CRD42021239274).

Data sources and searches

Two investigators (YE and AG-H) independently searched PubMed and EMBASE for studies listed from database inception to May 2021. Additionally, a grey literature search of abstracts and conference proceedings from the ADA, International Diabetes Association, British Diabetes Association (Diabetes UK) and the World Diabetic Foundation conventions in the last few years was undertaken. The following string of terms was used to identify studies investigating racial differences in mortality risk and future complications in individuals with diabetes: ‘race’; ‘racial’; ‘ethnic*’; ‘white’; ‘black’; ‘Asian’; ‘minorit*’; ‘diabetes’; ‘mortality’; ‘diabetes outcomes’; ‘stroke’; ‘amputation’; ‘death rate’; ‘survival’; ‘mortality’; ‘nephropathy’; ‘retinopathy’; ‘end-stage renal disease’; ‘lower extremity’; and ‘complication*’. Only publications written in the English language were included (see electronic supplementary material [ESM] Methods). A medical librarian was consulted to audit the quality of the search strategy. Reference lists of eligible articles were manually examined for further identification of relevant articles. Any disagreement was resolved by consensus with a third author (RR-V).

Study selection

After reviewing the title and abstract, two investigators (YE and AG-H) systematically assessed the full text of identified articles for eligibility. To be eligible for inclusion in the meta-analysis, studies reporting clinical outcomes in individuals with diabetes across different racial/ethnic groups (including, but not limited to, Black, Hispanic, White, Native American, Pacific Islander, East Asian, South Asian or Māori individuals) were selected according to the following criteria: (1) exposure: racial/ethnic group; (2) participants: individuals with type 1 and/or type 2 diabetes; (3) outcomes analysed: all-cause mortality (including the following specific cause-of-mortality categories: CVD, ischaemic heart disease, renal cause, heart failure, cancer, respiratory disorders, postoperative mortality after lower-limb amputation, and all other causes) and clinical complications (i.e. lower-limb amputation, dementia, ESRD, anaemia, cardiovascular event, stroke); and (4) study design: prospective cohort studies with at least 6 months of follow-up. Studies were excluded if they did not report data regarding the variables of interest and/or reported insufficient information for calculating HRs or RRs and 95% CIs. In the case of duplicate studies, only the most comprehensive or the most recent was included for qualitative appraisal. Conflicts over study inclusion were resolved by consensus with a third author (RR-V).

Data collection process and data items

Data collection was conducted independently by two investigators (YE and AG-H), using an Excel spreadsheet specifically designed for the present study. The following information was extracted from each study that met the selection criteria: (1) study characteristics (first author’s name, publication year, study location, sample size, number of participants across racial groups, follow-up duration, information on source of data); (2) participants’ information (sex, age, number of clinical events); (3) race assessment details (self-reported, extracted from primary care records, surname analysis, observation by the administrative staff); and (4) statistical analysis and study results (confounding factors, outcome of interest and main results). Missing data from the studies were requested from the corresponding authors of the original published papers; however no response was obtained.

Risk of bias in individual studies

The Quality Assessment Tool for Observational Cohort and Cross-sectional Studies was used to determine the risk of bias of each study. The checklist comprised 14 items for longitudinal studies. Each item of methodological quality was classified as ‘yes’, ‘no’ or ‘not reported’.

Summary measures

We estimated the race-specific prevalence in relation to all-cause mortality and clinical events (e.g. amputation, heart failure, stroke, nephropathy, etc.). The a priori plan was to conduct a one-step individual participant data meta-analysis. All analyses were carried out using Stata (version 16.1; Stata Corporation, College Station, TX, USA). Meta-analysis was performed using the random-effects inverse-variance model with the Hartung–Knapp–Sidik–Jonkman adjustment [8]. HRs were pooled, comparing racial differences using White individuals with diabetes as a reference group in relation to all-cause mortality and clinical events. It is important to clarify the following aspects regarding our statistical analyses: (1) meta-analyses were only performed for outcomes that were included in three or more studies; (2) when different sexes were included in studies, their data were analysed as independent samples; (3) when two or more tests for measuring the same variable were included in a study, we calculated the pooled HR according to its event (i.e. amputation, CVD, ERSD or all-cause mortality) using fixed effect models; (4) where original studies reported a referent group other than White individuals with diabetes, risk estimates were recalculated to guarantee that estimates were each relative to the reference group of interest; and (5) when studies presented several statistical risk-adjustment models, only HRs associated with the statistical models that contained the highest number of additional covariates were considered.

Synthesis of results

The percentage of variation across studies that is due to heterogeneity rather than chance was estimated using the heterogeneity index (I2), derived from the Cochran Q statistic, considering I2 values of 25%, 50% and 75% as cut-off points for low, moderate and high degrees of heterogeneity, respectively [9].

Risk of bias across studies

The presence of potential small-study effects due to publication bias, poor methodological quality in smaller studies, artefactual associations, true heterogeneity or chance, was analysed using the Luis Furuya-Kanamori (LFK) index and the Doi plot, respectively. Both tests have been shown to be more robust than the traditional funnel plot and Egger’s regression intercept test [10]. An LFK index value greater than 1 or less than −1 indicates minor asymmetry, and a value greater than 2 or less than −2 indicates major asymmetry [10].

Additional analysis

A sensitivity analysis was done to assess the robustness of the summary estimates and to determine whether or not a particular study accounted for the heterogeneity. To examine the effects of each result from each study on the overall findings, results were analysed with each study deleted from the model once. Subgroup analyses according to specific event (i.e. stroke, heart failure) were conducted. Finally, random-effects meta-regression analyses using method of moments (DerSimonian and Laird method) were estimated to separately evaluate whether results were different by length of follow-up (years) [11].

Results

Study selection

The electronic search strategy retrieved 3897 studies. After removing duplicates and screening titles and abstracts, 1779 studies were assessed for eligibility based on full text. A reference list of excluded articles and reasons for exclusion based on the full text is detailed in ESM Results. The PRISMA flow diagram illustrating the number of studies excluded at each stage of the systematic review and meta-analysis is shown in ESM Fig. 1.

Study characteristics

Table 1 summarises the characteristics of included studies. All studies were published in English. Twenty-three prospective cohort studies fulfilled eligibility criteria and were included in the systematic review, comprising 2,416,516 individuals diagnosed with diabetes (White 59.3%, Black 11.2%, Asian 1.3%, Hispanic-American 2.4%, Native American 0.2%, East Asian 1.9%, South Asian 0.8%, Pacific Islander 2.3%, Māori 2.4% and others 18.2%). The estimated number of events stratified by race/ethnicity is depicted in ESM Table 1. Overall, estimations of total events included 37,530 deaths, 5121 cardiovascular events, 90,125 cases of nephropathy and 69,107 amputations. Two studies reported HRs for racial/ethnic differences in dementia [12] and anaemia [13] but because there were no additional studies investigating these outcomes, these studies could not be included in the meta-analysis. Thus, 21 studies were included in the quantitative synthesis.

Table 1 Summary of prospective cohort studies included
Full size table

In most studies, micro- and macrovascular complications were extracted from electronic medical records and were identified using the ICD codes recorded in the course of routine care by the patients’ treating physicians. One study extracted information from the Australia and New Zealand Dialysis and Transplant Registry [14]. Another study extracted data for lower-limb amputation from discharge codes or via a claims reimbursement database for outside medical services [15]. All studies including mortality as an outcome extracted data from National Registries, death files and hospital records.

Nine studies were based on data from follow-up periods prior to year 2000. Fifteen studies were from North America (USA and Canada), one from the UK and seven from New Zealand. Fifteen studies included a multi-ethnic cohort, seven cohorts analysed only Black and White populations [16,17,18,19,20,21,22] and one study analysed Māori and White populations [14].

Participants had a mean age of 64.4 years and 43.6% were female. Mean time since diagnosis was 7.15 years (although this was only reported by seven studies [12, 15, 17, 18, 23,24,25]). Sample sizes across studies ranged from 267 [17] to 429,918 [26] individuals. The mean follow-up length was 8.15 years and ranged from 2 [13,14,15, 27] to 20 [17] years.

Summary measures (exposure)

Race/ethnicity was self-reported in most of the studies [12, 13, 15, 16, 21, 23, 27,28,29]. Two studies [26, 30] assessed race or national origin by extracting information from clinical documentation and/or observation of administrative staff and did not consider self-report. One study extracted race from primary care ethnicity records [24] and another study used a validated surname analysis to categorise participants as South Asian (from Pakistan, India or Bangladesh) or East Asian (from China, Taiwan, or Hong Kong) using surnames recorded in provincial registries [31]. Ten studies did not explain the race assessment method used [14, 17,18,19,20, 22, 25, 32,33,34].

Risk of bias within studies

All studies met at least ten out of 14 items included in the Quality Assessment Tool for Observational Cohort and Cross-sectional Studies and were considered to have fair-to-good methodological quality. The mean score was 10.61/14 (ESM Table 2).

Synthesis of results

All-cause mortality

Māori people with diabetes had a higher mortality risk than White people with diabetes (HR 1.88 [95% CI 1.61, 2.21]; p = 0.001; Q = 3.23; I2 = 7.1%; Fig. 1), with major asymmetry suggestive of small-study effects (LFK = −2.04; ESM Fig. 2). We observed no significant risk for mortality in other racial/ethnic groups when compared with White individuals (Fig. 1). Sensitivity analyses indicated no modifications in the results after removing one study at a time, with HRs ranging from 1.59 (95% CI 1.23, 1.95) [25] to 1.94 (95% CI 1.80, 2.09) [33].

Fig. 1
figure 1

Forest plot showing HRs (95% CIs) for all-cause mortality in Black, Hispanic-American, Māori and Pacific Islander vs White individuals with diabetes

Full size image

Based on meta-regression analyses, no significant associations were found between length of follow-up (β −0.026 [95% CI −0.152, 0.101]; p = 0.690) and the HR estimates.

CVD

Hispanic-American participants had a lower risk of CVD than White participants (HR 0.66 [95% CI 0.53, 0.81]; p = 0.013; Q = 1.31; I2 = 0%; Fig. 2), with major asymmetry suggestive of small-study effects (LFK = −4.55; ESM Fig. 3). No differences in the results were found in the sensitivity analysis after removing one study at a time, with HRs ranging from 0.59 (95% CI 0.48, 0.71) [27] to 0.66 (95% CI 0.60, 0.73) [23].

Fig. 2
figure 2

Forest plot showing HRs (95% CIs) for CVD in Black, Hispanic-American and South Asian vs White individuals with diabetes

Full size image

Regarding specific events, pooled HRs (95% CI) for coronary artery disease, heart failure, lower-extremity amputation and stroke stratified by Black vs White people, and Hispanic-American vs White people are shown in ESM Table 3. Hispanic-American individuals had a lower risk of coronary artery disease (HR 0.66 [95% CI 0.49, 0.89]; p = 0.027; Q = 1.43; I2 = 0%; LFK = −4.41; ESM Fig. 4), heart failure (HR 0.63 [95% CI 0.42, 0.95]; p = 0.041; Q = 0.38; I2 = 0%; LFK = 3.54; ESM Fig. 5) and stroke (HR 0.68 [95% CI 0.46, 0.99]; p = 0.048; Q = 1.99; I2 = 0%; LFK = −4.84; ESM Fig. 6) than White individuals. Black participants with diabetes were at lower risk for coronary artery disease (HR 0.76 [95% CI 0.66, 0.89]; p = 0.002; Q = 319.11; I2 = 96.6%), with no small-study effect (LFK = 1.30; ESM Fig. 7). In addition, meta-regression analysis did not reveal significant effects of length of follow-up (β −0.006 [95% CI −0.074, 0.062]; p = 0.872).

ESRD

Black individuals with diabetes had a higher risk for ESRD (HR 1.54 [95% CI 1.05, 2.24]; p = 0.033; Q = 109.83; I2 = 95.4%) than White individuals (Fig. 3), with major asymmetry suggestive of small-study effects (LFK = 4.58; ESM Fig. 8). Sensitivity analyses indicated no modifications in the results after removing one study at a time, with HRs ranging from 1.37 (95% CI 1.11, 1.63) [27] to 1.56 (95% CI 1.19, 1.93) [20].

Fig. 3
figure 3

Forest plot showing HRs (95% CIs) for ESRD in Black and Hispanic-American vs White individuals with diabetes

Full size image

Based on meta-regression analyses, no significant effects of length of follow-up were found (β −0.024 [95% CI −0.052, 0.004]; p = 0.097).

Amputation

No significant differences were observed in any racial/ethnic group compared with White participants (Fig. 4).

Fig. 4
figure 4

Forest plot showing HRs (95% CIs) for amputation in Black and Hispanic-American vs White individuals with diabetes

Full size image

Discussion

Our results show that, relative to White individuals, Māori individuals with diabetes have a 1.88-fold increased risk for all-cause mortality, Hispanic-American individuals have a significantly lower risk for macrovascular complications and Black individuals are at significantly higher risk for ESRD. Other racial/ethnic populations (East and South Asian, Native American or Pacific Islander) were not at significantly higher risk for diabetes complications than their White counterparts.

Our findings of higher mortality risk in Māori people than in White people are consistent with previous studies, which also reported Indigenous disparities in mortality risk among individuals with diabetes [35,36,37]. These relationships may partly be explained by poorer glycaemic control, albuminuria and higher obesity and smoking rates [38], which are suggested to render Māori individuals with diabetes more prone to adverse outcomes than their White peers. Moreover, socioeconomic disadvantage [39], in combination with poorer health-seeking behaviour, may contribute to a faster progression of the disease, and affected individuals may be sicker on arrival at hospital or may present at later stages of illness. These factors, however, do not explain the magnitude of the difference between Māori and other minority groups, such as Pacific Islanders, especially when considering that they share the same Polynesian genetic and cultural ancestry, and that access to care is similar to that for other racial groups in the same geographical area [29]. Other possible explanations for the excess risk in Māori individuals include lower social and familial support, dietary factors or even the existence of a different optimal ‘therapeutic window’ in response to glucose-lowering agents, which differs from that of White people on whom most guidelines are based. The modifiable risk factors and health barriers contributing to the health disparities among Māori individuals with diabetes should, therefore, be identified and explored.

Regarding macrovascular complications such as CVD, our findings revealed that Hispanic-American individuals had 34% lower risk for coronary artery disease, 32% lower risk for stroke and 37% lower risk for heart failure compared with White people. These findings are in line with those of previous systematic reviews [2, 3], although, paradoxically, Hispanic-American individuals present with more insulin resistance, and higher rates of central obesity and associated cardiovascular risk factors [40]. A possible lower degree of atherosclerosis and a lower internal carotid artery plaque thickness in Hispanic compared with Black or White people [41, 42] have been suggested as possible explanations for some of the observed differences. However, survivor bias may have influenced the overall results, as Hispanic-American patients with diabetes who developed CVD complications and died before becoming eligible for the medical insurance programme (i.e. Medicare claims) might not have been included in many studies. Nonetheless, as no significant difference in CVD rates between Hispanic-American and White people in the general US population exists [43], and it is unknown whether a protective effect against the development of macrovascular complications is present in Hispanic-American individuals, intensive control of cardiovascular risk factors is recommended as in other racial/ethnic groups.

Racial disparities in microvascular complications such as nephropathy have been previously described, with disproportionately higher rates of ESRD in Black individuals with diabetes than in White people with diabetes [44]. Consistent with the literature, our results showed that Black individuals with diabetes in the USA had a 56% higher risk of developing ESRD, even after adjusting for confounders such as age, sex, BMI, smoking, diabetes risk factors, comorbidities, type of diabetes and glucose-lowering medications. Several biological mechanisms have been proposed to explain the present findings. First, risk variants in the APOL1 gene on chromosome 22, which were initially discovered in African Americans, are associated with a considerably increased risk of kidney disease. These variants rose to high frequency in Western Africa as a protection against trypanosomal infection [45]. African Americans carrying two APOL1 risk alleles (G1 and G2) have shown to present heightened risk for glomerular disease, and APOL1-associated focal segmental glomerulosclerosis seems to appear earlier and progress to ESRD faster [46]. Second, renal disease risk can be augmented by preterm birth, which is more common in African American people than in White people according to data from the USA [47]. Adults born preterm may present with reduced nephron mass, leading to hampered renal pressure natriuresis and augmented vulnerability to BP factors (i.e. sodium intake and obesity) [48], potentially affecting end-organ damage and fuelling the progression to ESRD. Screening for early kidney disease, rigorous management of risk factors and further research on the aetiology of renal disease among Black individuals with diabetes is, therefore, warranted.

Evidence of racial disparities in diabetic amputations is present in the literature. For example, diabetic individuals from racial/ethnic minorities have a higher risk for amputation than White counterparts [49]. Conversely, a lower risk of diabetic foot ulcers and amputations among South Asian compared with White people has been identified [17, 50], and has been partly explained by better skin microvascularisation, lower smoking rates and shorter height [51]. Our study, however, was unable to find a significant higher risk for amputations in any racial/ethnic group compared with White people.

Overall, barriers to health when considering race include socioeconomic status, discrimination, transportation, criminal history, documentation status, language proficiency and neighbourhood violence [52, 53]. Racial/ethnic minorities are frequently exposed to the stress of racial discrimination, which is often managed by the adoption of detrimental health behaviours, such as smoking, excessive alcohol use, or illicit substance use [54]. Moreover, medical care, health insurance and their associated costs, can be significant challenges for individuals with lower socioeconomic status, thereby perpetuating the ‘diabetic complications cycle’. Important questions remain about the cumulative effects of excess body weight (i.e. BMI >25 kg/m2) over several decades, its effect across racial/ethnic groups, and interactions with diabetes risk factors [55]. Likewise, BMI cut-offs for obesity and for being overweight may differ in non-White populations [56]. Other unresolved questions relate to the most appropriate measure of ethnicity in terms of future diabetic complications, the mechanisms that underpin sex differences, and disparities in terms of access to health services in the life-course of individuals. Further research is necessary to identify the mechanisms underlying racial disparities in diabetes outcomes in high-risk populations.

Limitations and strengths

A number of caveats need to be noted regarding the present study. First, race and ethnicity are not synonyms; individuals who are of ‘Black’ race may be of ‘Hispanic’ ethnicity, and many studies did not define what they meant by specific racial or ethnic groups. Similarly, race and ethnicity are intrinsically imprecise, being estimations of societally defined groups (and in most instances, self-reported). Consequently, our results might be sensitive to how studies assessed and defined racial/ethnic groups. Second, all studies included were from New Zealand, the UK, the USA or Canada. While these countries have ethnically diverse populations, the generalisability of our findings to other countries is limited as the management of patients (such as healthcare access, insurance coverage or health-system infrastructure) may differ. Further research should be undertaken in other geographical settings and in diverse income-level countries, particularly in countries where individuals of White race do not characterise the majority of the population. Finally, subgroup analyses to explore whether factors such as level of adjustment, percentage of male/female participants included in the studies, country/location, age, BMI and risk of bias have an impact in the associations could not be performed because of the limited number of studies.

Conclusion

Given the small number of studies from four countries and bearing in mind potential bias due to possible race misclassification, caution is warranted regarding the strength of the existing evidence base. Nevertheless, our current findings increase substantially our understanding of the relevance of race and ethnicity in the development of future diabetic complications and mortality risk, lending support for the inclusion of such categories in international diabetes clinical guideline recommendations, at least until better predictors are available. Because of the importance of the management of risk factors, self-management education should be tailored according to the lifestyles and beliefs specific to racial/ethnic groups. A key policy priority should be to plan to address and overcome disparities in diabetes outcomes by improving identification of high-risk groups, and by an intensive control of cardiovascular risk factors across ethnically diverse populations. Further study of the biological and societal mechanisms underlying these disparities is, therefore, warranted.