Sex differences in the association between diabetes and risk of cardiovascular disease, cancer, and all-cause and cause-specific mortality: a systematic review and meta-analysis of 5,162,654 participants

Background Studies have suggested sex differences in the mortality rate associated with diabetes. We conducted a meta-analysis to estimate the relative effect of diabetes on the risk of all-cause, cancer, cardiovascular disease (CVD), infectious disease, and respiratory disease mortality in women compared with men. Methods Studies published from their inception to April 1, 2018, identified through a systematic search of PubMed and EMBASE and review of references. We used the sex-specific RRs to derive the women-to-men ratio of RRs (RRR) and 95% CIs from each study. Subsequently, the RRR for each outcome was pooled with random-effects meta-analysis weighted by the inverse of the variances of the log RRRs. Results Forty-nine studies with 86 prospective cohorts met the inclusion criteria and were eligible for analysis. The pooled women-to-men RRR showed a 13% greater risk of all-cause mortality associated with diabetes in women than in men (RRR 1.13, 95% CI 1.07 to 1.19; P < 0.001). The pooled multiple-adjusted RRR indicated a 30% significantly greater excess risk of CVD mortality in women with diabetes compared with men (RRR 1.30, 95% CI 1.13 to 1.49; P < 0.001). Compared with men with diabetes, women with diabetes had a 58% greater risk of coronary heart disease (CHD) mortality, but only an 8% greater risk of stroke mortality (RRRCHD 1.58, 95% CI 1.32 to 1.90; P < 0.001; RRRstroke 1.08, 95% CI 1.01 to 1.15; P < 0.001). However, no sex differences were observed in pooled results of populations with or without diabetes for all-cancer (RRR 1.02, 95% CI 0.98 to 1.06; P = 0.21), infectious (RRR 1.13, 95% CI 0.90 to 1.38; P = 0.33), and respiratory mortality (RRR 1.08, 95% CI 0.95 to 1.23; P = 0.26). Conclusions Compared with men with the same condition, women with diabetes have a 58% and 13% greater risk of CHD and all-cause mortality, respectively, although there was a significant heterogeneity between studies. This points to an urgent need to develop sex- and gender-specific risk assessment strategies and therapeutic interventions that target diabetes management in the context of CHD prevention. Electronic supplementary material The online version of this article (10.1186/s12916-019-1355-0) contains supplementary material, which is available to authorized users.


Background
According to the Global Burden of Disease Study (GBD), non-communicable diseases (NCDs) are the main cause of premature deaths amongst the world's population [1]. As one of four main NCDs, diabetes affected an estimated 387 million people throughout the world and caused around 1.3 million deaths worldwide in 2010 alone [2][3][4]. With the increasing prevalence of physical inactivity and obesity, the burden of diabetes is predicted to increase to 592 million by 2035, making it a major contributor to the global burden of disease [5].
Type 2 diabetes mellitus is associated with an approximate twofold increase in the risk of all-cause mortality as well as death from cardiovascular disease (CVD), kidney disease, infectious disease, respiratory disease, and several specific forms of cancer [6]. Previous metaanalyses, through internal, within-study comparisons of female and male participants, have observed that women with diabetes are at substantially higher risk of coronary heart disease (CHD), stroke, and gastric cancer compared to affected men. On the other hand, no sex differences were found between diabetes and the risk of esophageal cancer, colorectal cancer, and pancreatic cancer [7][8][9]. However, the magnitude of the excess risk of these and other cause-specific outcomes that are conferred by diabetes for men and women is unknown. Furthermore, it is unclear whether important confounders (e.g., age) and methodological heterogeneity (duration of follow-up, method of diabetes classification or assessment) would modify any such sex differential in the association between diabetes and mortality. It is also unclear whether such a difference might be more pronounced in recent years with the growing obesity epidemic (e.g., year of publication).
Accordingly, we sought to conduct a meta-analysis of prospective cohort studies in order to (i) calculate any sex differential in the association between diabetes and risk of cardiovascular disease, cancer, and all-cause and cause-specific mortality for the general population and (ii) to determine whether these associations are modified by demographics, setting, length of follow-up, diabetes measurement, and recency of publication.

Search strategy
The meta-analysis was performed in accordance with the Meta-Analysis of Observational Studies in Epidemiology guidelines [10] and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement [11] (Additional file 1: Table S1). We searched the PubMed and EMBASE databases from their inception to April 1, 2018. Details of the search strategy using a combined text word and medical subject heading are displayed in Additional file 1. The articles were restricted to English language studies. Moreover, the reference lists of the retrieved publications and reviews were checked for other potentially relevant studies.

Study selection
Studies were included if they met the following criteria: (1) the study was a prospective cohort design; (2) the outcomes included all-cause mortality, cancer mortality, CVD mortality, CHD mortality, stroke mortality, infectious disease mortality, and/or respiratory disease mortality; (3) the studies provided odds ratio (OR), relative risk (RR), or hazard ratio (HR) with 95% confidence intervals (CI) for the associations between diabetes and mortality disaggregated for men and women participants; and (4) when multiple publications reported on the same population or subpopulation, the study with the most recent or most informative data was included. The exclusion criteria were as follows: (1) matched prospective cohort study design, (2)

Data extraction and study quality assessment
Two investigators (YFW and YRJ) independently reviewed all potentially eligible studies using predefined criteria and extracted the data from each paper. In case of incomplete or unclear data, the authors were contacted where possible. The cohort study quality was estimated using the nine-star Newcastle-Ottawa quality assessment Scale (NOS) ranging from zero to nine stars [12]. Disagreements were resolved by consensus between the authors.

Statistical analysis
The RR was used as a measure of the association between diabetes and outcome risk. For individual participant data, we used Cox proportional hazards regression to obtain HRs (regarded as RRs). If the included studies did not report the RRs, the HRs were directly considered as RRs and the ORs were converted into RRs using the formula: RR = OR/[(1 − Po) + (OR × Po)], in which Po was the incidence of the outcome of interest in the nondiabetes group [13]. For studies that reported RRs in different age groups, we pooled these RRs with inverse variance random-effect models, and then we used combined estimates for that study. For the primary analysis, we used the sex-specific RRs to derive the womento-men ratio of RRs (RRR) and 95% CIs from each study, as previously described [14]. Subsequently, the RRR for each outcome was pooled with random-effects metaanalysis weighted by the inverse of the variances of the log RRRs. We also pooled RRs for men and women separately, using an identical approach. The heterogeneity among the included studies was evaluated by the Q test and I 2 statistic [15].
Subsequently, where the number of included studies was more than 10 for each outcome of interest, sensitivity analyses were performed by mean age (≤ 60 versus > 60 years), region (Asia versus Europe versus America versus others), publish year (≤ 2000 versus 2001-2009 versus ≥ 2010), length of follow-up (≤ 10 versus > 10 years), and ascertainment of diabetes (known diabetes versus newly diagnosed diabetes versus both). Random-effects meta-regression analyses were used to evaluate whether the differences in the mean/medium duration of study follow-up and mean age of participants at baseline contributed to the heterogeneity between the studies. Publication bias was assessed by Begg's rank correlation test and its funnel plots of the natural log of the RRR against its standard error [16]. Where publication bias was detected, trim and fill analyses were used to adjust the RRs or ratio of RRs. All statistical analyses were performed with Stata version 13.0 (StataCorp, College Station, TX, USA).

Results
Of the 24,303 references identified through the systematic search, 375 were examined in the full-text review ( Fig. 1). In addition, 6 articles were retrieved from the reference lists of relevant articles and reviews. Subsequently, individual participant data from NHIS were added to these published results. Finally, 49 studies with 86 prospective cohorts met the inclusion criteria and  were eligible for analysis ( Table 1). The characteristics of the included studies are described in Table 1. Baseline surveys were conducted between 1950 and 2014, and the number of participants ranged from 379 to 1,298,358. The mean/median duration of follow-up ranged from 6.0 to 21.4 years, while the average baseline age was between 33.3 and 80.0 years. The quality of all included studies based on NOS was high (Additional file 1: Table S2). All studies adjusted for age and most of the studies also controlled for smoking (n = 77), hypertension (n = 71), and body mass index (n = 68).
Twenty-eight studies with 3,887,585 participants were included to assess the sex-specific association between diabetes and all-cause mortality. For cause-specific mortality, 14 studies with 4,482,501 reported on cancer mortality, 23 studies with 2,067,486 reported on CVD mortality, 23 studies with 2,050,929 reported on CHD mortality, 15 studies with 2,292,387 reported on stroke mortality, 4 studies with 1,633,520 reported on respiratory disease mortality, and 3 studies with 1,638,651 reported on infectious disease mortality.
Sex-specific association between diabetes and risk of allcause, cancer, CVD, infectious disease, and respiratory disease mortality The pooled multiple-adjusted RRs of all-cause mortality associated with diabetes compared with no diabetes were 1.93 (95% CI 1.80 to 2.06; Fig. 2) in women and 1.74 (1.67 to 1.82) in men. The pooled women-to-men RRR showed a 13% greater risk of all-cause mortality associated with diabetes in women than in men (RRR 1.13, 95% CI 1.07 to 1.19; P < 0.001; Figs. 3 and 4). There was, however, a significant heterogeneity between the studies (I 2 = 60%, P < 0.001; Fig. 2).
The pooled multiple-adjusted RRs showed that diabetes was associated with a 26% (1.16 to 1.36) increased risk for cancer mortality in women and a 29% (1.18 to 1.42; Additional file 1: Figure S1) increased risk in men. There was no evidence of a sex difference in the association between diabetes and cancer mortality; the pooled multiple-adjusted RRR of cancer mortality for diabetes was 1.02 (0.98 to 1.06; P = 0.21; Fig. 4; Additional file 1: Figure S2). No evidence of significant between-study heterogeneity was found (I 2 = 0%; P = 0.60).

Subgroup, meta-regression, and publication bias analyses
We performed subgroup analyses for cancer, CHD, stroke, CVD and all-cause mortality outcomes. Results showed no evidence of heterogeneity between the subgroups stratified by study characteristics including age, geographical location, duration of follow-up, publish year, and method of diabetes ascertainment (Table 2). For the method of diabetes ascertainment, sex differences for CVD, CHD, and all-cause mortality conferred by diabetes were only significant in self-reported diagnosis (all-cause mortality: RRR 1.17, 95% CI 1.07 to 1.27, P < 0.001; CVD mortality: RRR 1.20, 95% CI 1.02 to 1.42, P < 0.001; CHD mortality: RRR 1.52, 95% CI 1.20 to 1.92, P < 0.001). The pooled RRR for CHD, stroke, CVD, and all-cause mortality did not vary by mean age of the participants at baseline, mean/medium duration of follow-up, baseline prevalence of diabetes, and women-to-men ratio of diabetes prevalence (all P > 0.1). We found no evidence of publication bias for cancer, CHD, stroke, CVD, respiratory disease, infectious disease, and allcause mortality (P > 0.10).

Discussion
This systematic review and meta-analysis of 49 studies with 86 prospective cohorts found that diabetes conferred  a greater risk for almost all outcomes of interest. Diabetes appears to be a stronger risk factor for CHD, CVD, and all-cause mortality in women than in men. Of note, compared to men with diabetes, women with the same condition had 57% excess risk for CHD. Although diabetes was associated with a higher risk of cancer mortality, infectious disease, and respiratory disease mortality, we did not observe a sex difference between diabetes and mortality. Interestingly, however, these results were only upheld in studies that used self-reporting measures to identify diabetes cases.
Diabetologists and epidemiologists have long been aware that diabetes has pronounced cardiovascular consequences for women, irrespective of diabetes type [10,28,64]. Indeed, CVD is the leading cause of morbidity and mortality for individuals with diabetes, which accounts for > 50% of all deaths [65]. We found that for women, diabetes confers a 54% excess risk of CHD death. While such sex-specific differences are of increasing interest in cardiology and medical fields, the underpinning mechanisms driving this association are not entirely clear. The pathogenesis seems to be multifactorial with contributions from sex differences in genetic and biological factors, gender disparities from cultural and environmental factors, and the well-documented differences in the diagnosis, management, and treatment of DM and CVD of women and men [66][67][68].
The putative biological mechanisms have centered on the effects of estrogen which can deplete during menopause to elevate women's CHD risk [69]. Testosterone Fig. 3 Pooled women-to-men RRRs for risk of all-cause mortality may be involved in different mechanisms attributed to sex differential in CHD risk [70][71][72]. In men, higher total testosterone levels are associated with reduced risk of future CHD and ischemic stroke. Testosterone has anabolic effects, promoting muscle mass and strength [73]. The recent prospective cohort study of half a million UK Biobank participants showed that higher grip strength was associated with a lower risk of incidence of and mortality from CVD [74]. Compared with men, women with lower testosterone levels have low mass and strength of muscle, which also partially explain greater risk for CHD death conferred by diabetes in women compared with men.
Women with diabetes are more likely to have poor risk factor profiles and suffer greater disease risk owing to the effects of individual risk factors. A recent metaanalysis showed that smoking conferred 25% excess risk for CHD in women than in men [7]. In addition, women with diabetes remain less likely to achieve high-density lipoprotein cholesterol targets and have a higher prevalence of obesity than men [75][76][77]. Whether existing sex differences in diabetic heart disease are magnified by sex differences in traditional and modifiable cardiac risk factors requires consideration. Recently, a meta-analysis of individual data from 68 prospective studies showed that body mass index, blood pressure, and total cholesterol each had continuous log-linear associations with CHD or stroke mortality that were similar in strength among those with and those without diabetes, irrespective of sex [78]. Our other study found that compared with men with metabolic syndrome, women with metabolic syndrome had a significant 16% higher risk of CHD incidence (RRR 1.16, 95% CI 1.01 to 1.34; P = 0.04), and the significant sex difference disappeared in non-diabetes population (RRR 0.92, 95% CI 0.73 to 1.17; P = 0.50). This partly supported the hypothesis that the stronger detrimental effects of diabetes for women than for men in CVD could not be explained by the different levels of established major CVD risk factors and their clusters. Differences in the clinical manifestation of diabetes warrants further consideration. Prediabetes is associated with an increased risk of cardiovascular disease [79], and the sex differences in the nonphysiological effects can be partly accounted for the diabetes-related excess risk of CVD in women. In the prediabetic state, impaired glucose tolerance may be more serious in women than in men [80,81].   Biases embedded within health service need to be considered. There is evidence that women, compared to their male counterparts, are less likely to have their risk factors assessed by physicians when they present in primary care. Compared to older women at high risk of CVD, younger women at high risk were less likely to receive preventative treatment [82]. Indeed, women with diabetes or CVD are diagnosed later and have a lower frequency of statin therapy, aspirin use, and ACE inhibitor and β-blocker use than men [83]. Some studies observed lower medication adherence in women than in men [84,85]. Where medication is adhered to, women do not always benefit to the same extent as men given the well-documented issues with under-representation of women in clinical trials [66]. What is more, younger women's symptoms often present differently to those of men of the same age. There may be less myocardial ischemic preconditioning in women, and subsequently greater susceptibility to ischemia. Therefore, sex and gender disparities in treatment may exacerbate the sex differences in CVD owing to diabetes [86,87].
Some studies show that the proportion of undiagnosed diabetes to total diabetes in men is higher than that in women [88,89]. In studies that used self-reported measures to identify diabetes, there was a greater proportion of undiagnosed diabetes in men. It is possible that this concealed the true excess risk of mortality conferred by diabetes in men and subsequent sex-specific relative risk estimates that were calculated for women and men.
Our finding that diabetes elevates the risk of all-cancer mortality is in general agreement with previous reviews [90]. However, most have looked at site-specific cancers; sexspecific associations from which results have been inconsistent. One meta-analysis indicated that diabetes conferred a stronger positive relationship with kidney cancer mortality and gastric cancer risk in women than in men [91,92]. Others have found that diabetes increased the risk of esophageal cancer and leukemia in men, but not in women [93,94]. Prospective studies showed that HRs for noncancer, non-vascular deaths among participants with diabetes, as compared to those without diabetes, were also significantly higher among women (women: HR 2.20, 95% CI 1.91 to 2.52; men: HR 1.58, 95% CI 1.41 to 1.76; P interaction < 0.001). The absence of sex disparities for infectious disease and respiratory disease mortality did not contribute to the sex difference for non-cancer, non-vascular deaths [95]. Therefore, future research is needed to distinguish whether and to what extent the excess risk of cause-specific mortality from non-cancer, non-vascular deaths conferred by diabetes differs between the sexes, such as kidney disease mortality.

Strengths and limitations
The present meta-analysis has several main strengths. Firstly, the large number of participants ensured greater statistical power to detect sex differences than some previous individual studies. Secondly, using within-study comparisons to estimate sex-specific relationships between diabetes and cause-specific outcomes can minimize the role of extraneous, between-study factors. Thirdly, the study comprehensively evaluated the sex-specific associations for a range of important health outcomes: all-cause, all-cancer, CVD, and other cause-specific mortality. This has the potential to be more informative in aiding our understanding of the sex-specific burden of disease from diabetes. Fourthly, the detailed subgroup, sensitivity, and influence analyses ensure the robustness of the study findings.
There are also some specific limitations of this review that merit consideration. Firstly, there was some heterogeneity across studies for outcomes such as all-cause, CVD, and CHD mortality, but subgroup analyses and meta-regression analyses on study characteristics including age, geographical location, duration of follow-up, publish year, and method of diabetes ascertainment did not provide any evidence of a substantial effect of these differences on the results. Secondly, the present metaanalysis is based on prospective cohort studies, and the observational design is open to biases due to the residual confounding from incompletely measured factors and cannot elucidate causal relationship. Thirdly, the present meta-analysis did not include non-fatal events, which limited the ability to assess the presence of sex differences in risk for the incidence. Fourthly, differences in definition of diabetes, diabetes duration, duration of follow-up, and populations might have contributed to the sex differences in the association of diabetes with risk of death and CVD; although subgroup, metaregression, and sensitivity analyses were conducted to explore the potential between-study heterogeneity, lack of individual participant data limited more in-depth sensitivity analyses than were reported here. Fifthly, our analysis cannot ascertain the underlying cause of the sex differences in the relationship between diabetes and the risk of CVD mortality. Finally, the potential publication bias was also a concern. Although we did not observe any apparent publication bias in our statistical tests, it was still difficult to completely rule this out.

Conclusions
Our study demonstrated that women with diabetes have a greater risk of all-cause mortality, particularly from CHD, compared with men with the same condition. An increased understanding and appreciation of sex differences in the relationship between diabetes and risk of all-cause and CHD mortality is required given the substantial global and regional burden of NCDs. Women with diabetes should be treated and managed throughout their life course with the view to reduce the burden of other diseases related to diabetes. In the future, indepth sex-specific analyses from randomized trials and other studies using approaches like Mendelian randomization are needed to clarify the biological, behavioral, or social mechanisms involved.

Additional file
Additional file 1: contains additional information and analysis. Table S1. Study protocol: PRISMA 2009 Checklist. Table S2. Quality of included studies assessed with Newcastle-Ottawa Scale. Figure S1. Pooled RRs for the risk of cancer mortality. Figure S2. Pooled women-to-men RRRs for the risk of cancer mortality. Figure S3. Pooled RRs for the risk of CVD mortality. Figure S4. Pooled women-to-men RRRs for the risk of CVD mortality. Figure S5. Pooled RRs for the risk of CHD mortality. Figure S6. Pooled women-to-men RRRs for the risk of CHD mortality. Figure S7. Pooled women-to-men RRRs for the risk of stroke mortality. Figure S8. Pooled RRs for the risk of respiratory mortality.