Articles

Diabetes as a risk factor for stroke in women compared with

men: a systematic review and meta-analysis of 64 cohorts,
including 775 385 individuals and 12 539 strokes
Sanne A E Peters, Rachel R Huxley, Mark Woodward

Summary
Background Diabetes mellitus is a major cause of death and disability worldwide and is a strong risk factor for Lancet 2014; 383: 1973–80
stroke. Whether and to what extent the excess risk of stroke conferred by diabetes differs between the sexes is Published Online
unknown. We did a systematic review and meta-analysis to estimate the relative effect of diabetes on stroke risk in March 7, 2014
http://dx.doi.org/10.1016/
women compared with men.
S0140-6736(14)60040-4
See Editorial page 1945
Methods We systematically searched PubMed for reports of prospective, population-based cohort studies published
See Comment page 1948
between Jan 1, 1966, and Dec 16, 2013. Studies were selected if they reported sex-specific estimates of the relative risk
Julius Center for Health
(RR) for stroke associated with diabetes, and its associated variability. We pooled the sex-specific RRs and their ratio
Sciences and Primary Care,
comparing women with men using random-effects meta-analysis with inverse-variance weighting. University Medical Center
Utrecht, Utrecht, Netherlands
Findings Data from 64 cohort studies, representing 775 385 individuals and 12 539 fatal and non-fatal strokes, were (S A E Peters PhD);
Cardiovascular Epidemiology
included in the analysis. The pooled maximum-adjusted RR of stroke associated with diabetes was 2·28 (95% CI 1·93–2·69) Unit, Department of Public
in women and 1·83 (1·60–2·08) in men. Compared with men with diabetes, women with diabetes therefore had a Health and Primary Care,
greater risk of stroke—the pooled ratio of RRs was 1·27 (1·10–1·46; I²=0%), with no evidence of publication bias. This University of Cambridge,
sex differential was seen consistently across major predefined stroke, participant, and study subtypes. Cambridge, UK (S A E Peters);
School of Population Health,
University of Queensland,
Interpretation The excess risk of stroke associated with diabetes is significantly higher in women than men, Brisbane, QLD, Australia
independent of sex differences in other major cardiovascular risk factors. These data add to the existing evidence that (Prof R R Huxley DPhil); The
men and women experience diabetes-related diseases differently and suggest the need for further work to clarify the George Institute for Global
Health, University of Sydney,
biological, behavioural, or social mechanisms involved. Sydney, Australia
(Prof M Woodward PhD);
Funding None. Department of Epidemiology,
Johns Hopkins University,
Baltimore, MD, USA
Introduction account.6 Whether this sex difference also exists for (M Woodward); and The George
Diabetes mellitus is a global health concern; an estimated stroke—which shares many of the same risk factors— Institute for Global Health,
347 million people worldwide are affected, and in 2008 remains uncertain. Findings from previous studies have Nuffield Department of
diabetes accounted for 1·3 million deaths.1–3 The been inconsistent, with some investigators reporting Population Health, University
of Oxford, Oxford, UK
incidence of diabetes is projected to increase by more either a stronger,7,8 similar,9 or weaker effect of diabetes (M Woodward)
than 50% in the next decade because of rapid increases on stroke risk in women compared with men.10,11 Correspondence to:
in the prevalence of obesity and physical inactivity. As a In view of the substantial implications that any Prof Rachel Huxley,
result, diabetes is predicted to become the seventh clinically important sex difference in the association Epidemiology and Biostatistics
leading cause of death in the world by 2030.3,4 between diabetes and stroke risk would have, we did a Division, School of Population
Health, University of Queensland,
The burden of diabetes as a major cause of premature systematic review and meta-analysis of all available Herston, QLD 4006, Australia
illness and death is mostly caused by the associated prospective data to estimate the relative effect of diabetes r.huxley@uq.edu.au
increased risk of cardiovascular disease. Widely quoted on stroke risk in women compared with men.
estimates from WHO suggest that the cardiovascular
risk in people with diabetes is two to three times higher Methods
than in those without the disease, and that cardiovascular Search strategy and selection criteria
diseases cause between 50% and 80% of deaths in people We systematically searched PubMed for reports published
with diabetes.3 However, these estimates are based on the between Jan 1, 1966, and Dec 16, 2013, using a combined
assumption that diabetes confers the same degree of risk text and MeSH heading search strategy with the terms:
in women as in men, which is unlikely to be correct in “diabetes mellitus”, “diabetes”, “prediabetes”, “impaired
view of the accruing evidence that women and men fasting glucose”, “impaired glucose intolerance”,
experience the disease differently.4,5 Indeed, we have “borderline diabetes,” “blood glucose”, “hemoglobin A,
previously shown that the relative risk (RR) of diabetes- glycosylated”, “cohort studies”, “sex”, “gender”, “cardio-
related coronary heart disease is substantially higher in vascular disease”, “stroke”, “cerebrovascular disease”,
women than in men, even after differences in other “cerebrovascular attack”, “cerebral ischemia”, “brain
major cardiovascular risk factors have been taken into ischemia”, and “intracranial hemorrhage”. We also checked

www.thelancet.com Vol 383 June 7, 2014 1973

 Articles

the reference lists of identified reports for other potentially results were included in our primary analyses. In pooling
relevant studies. Prospective studies were included if RRs, multiple-adjusted results, the set of adjustments made
or equivalents, for the association between diabetes and were allowed to vary by study, but had to include at least
stroke in men and women were reported. Studies were one other risk factor for stroke in addition to age.
excluded if they had not adjusted at least for age, did not For each study, we extracted the sex-specific RRs (with
provide information about the variability around the point 95% CIs) for individuals with diabetes versus individuals
estimate, or were done in populations that predominantly without diabetes, from which we estimated the women-
consisted of individuals with a history of cardiovascular to-men ratio of RRs (and its 95% CI).6,16 After log
disease or other underlying pathological disorders. transformation of study-specific estimates, we generated
When duplicate reports from the same study were pooled estimates across studies using random-effects
identified, only the most recent publication, or the one meta-analysis. We used the inverse of the variance of the
with the longest follow-up period, was included. We also log RR to weight studies on the basis of an estimate of
used individual participant data from four studies: the statistical size.17 An identical approach was used for the
Scottish Heart Health Extended Cohort study (SHHEC),12 ratio of RRs.
the National Health And Nutrition Examination Survey We did several sensitivity analyses: fatal stroke only,
III (NHANES III),13 the Asia Pacific Cohort Studies region (Asia or non-Asia), baseline year of data collection
Collaboration (APCSC),14 and the Atherosclerosis Risk in (pre-1985 or 1985 onwards), method of diabetes
Communities study (ARIC).15 We contacted the authors of ascertainment (self-report or medical record), data source
all the studies that provided sex-specific RRs to ask them (aggregated data or individual participant data), and major
to provide data adjusted for a predefined set of stroke subtype (ischaemic or haemorrhagic). Using data
confounders. Authors of studies that had reported from APCSC, ARIC, NHANES III, and SHHEC, we also
prospective data, but that had adjusted for, rather than did analyses by age (<60 years vs ≥60 years) and smoking
stratified by, sex, were asked to provide separate results status (current smoker vs non-smoker).
for men and women. The I² statistic was used to estimate the percentage of
variability across studies due to between-study hetero-
Data extraction and statistical analysis geneity. We used random-effects meta-regression analyses
The primary endpoint was combined fatal or non-fatal to assess whether differences in the mean duration of study
stroke, and the primary metrics were the pooled adjusted follow-up, overall incidence of stroke, women-to-men ratio
RRs and the women-to-men ratio of RRs for individuals of stroke, prevalence of diabetes, or women-to-men ratio of
with diabetes versus those without diabetes. To include diabetes prevalence contributed to heterogeneity between
the largest set of individuals and stroke endpoints, studies studies.18,19 We used funnel plots to examine the presence of
that reported either age-adjusted or multiple-adjusted publication bias (ie, by plotting the natural log of the ratio
of RRs against its standard error).
Using individual participant data from APCSC, ARIC,
6120 potentially relevant studies assessed by title and abstract NHANES III, and SHHEC, we calculated stroke
incidence rates in men and women with and without
6028 studies excluded because they did not investigate the association
diabetes. Furthermore, to explore whether sex differences
between diabetes and stroke in asymptomatic individuals in other cardiovascular risk factors in individuals with
and without diabetes might contribute to any sex
difference seen in the RR for diabetes-associated stroke,
92 studies retrieved for more detailed assessment
we calculated the adjusted mean difference (with 95%
CI) for systolic blood pressure, total cholesterol, high-
79 studies excluded density lipoprotein (HDL) cholesterol, body-mass index
21 did not report sex differences in the association (BMI), and waist circumference in participants with and
15 used an inappropriate study endpoint
15 were done in populations with symptomatic individuals without diabetes, by sex. We pooled these data in meta-
14 were duplicate, cross-sectional, or retrospective studies* analyses and weighted by the inverse of the variance. All
14 did not assess diabetes as a risk factor
statistical analyses were done with Stata (version 11.0).

13 studies provided tabulated data† 4 studies provided individual participant data Role of the funding source
There was no funding source for this study. The
corresponding author had full access to all the data in the
study and had final responsibility for the decision to
17 studies (including one with separate results for two regions)
included in the main analysis
submit for publication.

Figure 1: Study selection

Results
*Including the four studies for which individual participant data were provided. †12 provided published data, and The systematic search identified 6120 articles, which
one unpublished data. were assessed by title and abstract. Of these, 92 articles

1974 www.thelancet.com Vol 383 June 7, 2014

 Articles

qualified for selection (figure 1). After full-text This database was extended with individual participant
assessment, 12 studies provided published, and one data from APCSC, ARIC, NHANES III, and SHHEC.12–15
study unpublished, summary data for sex differences in Table 1 shows the baseline characteristics of all
the association between diabetes and risk of stroke.7–11,20–28 64 cohorts included in the study. The year of the

Location Year of Study n Age Number with Ascertain- Number Fatal or Maximum adjustment
baseline duration (% women) range diabetes ment of of strokes non-fatal available
data (years) (years) (% women) diabetes (% women) strokes
collection included
APCSC Australia 1989–96 7 99 624 (45%) 20–104 4784 (31%) Self-report 1671 (41%) Both Age, SBP, smoking, BMI, total cholesterol
(Australia and and New or medical
New Zealand) Zealand
14
assessment
(nine
cohorts)
APCSC (Asia)14 Asia 1961–93 7 436 832 (33%) 20–107 17 763 (23%) Self-report 2872 (31%) Both Age, SBP, smoking, BMI, total cholesterol
(27 cohorts) or medical
assessment
ARIC15 USA 1987–89 18 15 732 (55%) 45–64 1610 (58%) Medical 930 (52%) Both Age, SBP, smoking, BMI, total cholesterol
assessment
DECODE10 Finland 1987–2002 5–21 9278 (55%) 40–69 826 (47%) Medical 530 (34%) Both Age, hypertension, BMI, total cholesterol,
and Sweden assessment HDL cholesterol, smoking
(seven
cohorts)
Dubbo28 Australia 1988–89 16 2006 (58%) >60 142 (49%) Medical 286 (55%) Both Age, SBP, smoking, BMI, total cholesterol
assessment
EPIC–Norfolk22 UK 1993–97 10 22 516 (55%) 40–79 441 (37%) Self-report 507 (49%) Both Age, SBP, smoking, BMI, total cholesterol,
triglycerides
Framingham USA 1988–90 14 2097 (50%) 50–81 99 (29%) Medical 130 (42%) Both Age, SBP, antihypertensive use, CVD,
Offspring25 assessment atrial fibrillation, LVH, smoking
Hisayama Japan 1988 14 2421 (57%) 40–79 291 (46%) Medical 132 (54%) Both Age, SBP, smoking, BMI, total cholesterol,
(2010)11 assessment HDL cholesterol, alcohol intake, physical
activity, ECG abnormalities
Hisayama Japan 1961 32 1621 (56%) >40 130 (34%) Medical 298 (52%) Both Age
(2000)9 assessment
Iso et al21 Japan 1975–80 17 10 582 (60%) 40–69 267 (48%) Medical 400 (45%) Fatal Age, hypertension, smoking, BMI, total
assessment cholesterol, HDL cholesterol, skinfold,
alcohol intake, community, menopause
JPHC7 Japan (two 1990–93 12 35 657 (63%) 40–69 2034 (63%) Medical 904 (47%) Both Age, SBP, antihypertensive use, smoking,
cohorts) assessment BMI, total cholesterol, HDL cholesterol,
triglycerides, alcohol intake, fasting
status, residential areas
Kuopio and Finland 1972–97 17 51 735 (51%) 25–74 1108 (46%) Self-report 917 (47%) Fatal Age, SBP, smoking, BMI, total cholesterol,
North Karelia8 (six cohorts) study year
NHANES III13 USA 1988 13 18 603 (46%) 18–90 1290 (38%) Self-report 329 (42%) Fatal Age, SBP, smoking, BMI, total cholesterol
or medical
assessment
Rancho USA 1972–74 12 3778 (54%) 50–79 320 (39%) Self-report 232 Both Age, SBP, total cholesterol, smoking,
Bernado24 obesity, family history, oestrogen use
Renfrew/ Scotland 1972–76 20 15 406 (54%) 45–64 NA Self-report 1029 (54%) Both Age
Paisley20 or medical
assessment
Sievers et al23 USA 1965–84 10 5131 (52%) 15–84 1266 (58%) Medical 30 (50%) Fatal Age
assessment
SHHEC12 Scotland 1984–87 16 13 287 (51%) 30–74 184 (46%) Medical 1083 (43%) Both Age, SBP, smoking, BMI, total cholesterol
assessment
Takayama26 Japan 1992 7 29 079 (54%) >35 1217 (35%) Self-report 259 (51%) Fatal Age, hypertension, smoking, BMI,
physical activity, education, energy
intake, vegetables, fat, alcohol intake

APCSC=Asia Pacific Cohort Studies Collaboration. SBP=systolic blood pressure. BMI=body-mass index. ARIC=Atherosclerosis Risk in Communities study. DECODE=Diabetes Epidemiology: COllaborative analysis
of Diagnostic criteria in Europe study. HDL=high-density lipoprotein. EPIC–Norfolk=European Prospective Investigation into Cancer, Norfolk. CVD=cardiovascular disease. LVH=left-ventricular hypertrophy.
ECG=electrocardiogram. JPHC=Japan Public Health Center study. NHANES III=National Health And Nutrition Examination Survey III. NA=not available. SHHEC=Scottish Heart Health Extended Cohort study.

Table 1: Characteristics of included studies

www.thelancet.com Vol 383 June 7, 2014 1975

 Articles

baseline survey ranged from 1961 to 2002, and duration had a fatal or non-fatal stroke. The incidence of stroke
of follow-up was between 5 and 32 years. Overall, data was greater in individuals with diabetes than in those
See Online for appendix were available from 775 385 individuals, of whom 12 539 without (appendix p 1). Table 2 shows the age-adjusted

Age-adjusted mean difference (95% CI) Multiple-adjusted mean difference (95% CI)
Men Women Men Women
Systolic blood pressure (mmHg) 5·33 (5·09 to 5·58) 6·79 (6·40 to 7·18) 4·18 (3·93 to 4·44) 4·70 (4·26 to 5·14)
Total cholesterol (mmol/L) 0·22 (0·20 to 0·23) 0·24 (0·22 to 0·27) 0·15 (0·13 to 0·17) 0·14 (0·11 to 0·17)
HDL cholesterol (mmol/L) −0·09 (−0·11 to −0·08) −0·16 (−0·17 to −0·14) −0·06 (−0·08 to −0·05) −0·11 (−0·13 to −0·09)
BMI (kg/m²) 1·01 (0·97 to 1·06) 2·00 (1·91 to 2·09) 0·64 (0·59 to 0·68) 1·66 (1·57 to 1·75)
Waist circumference (cm) 5·27 (4·79 to 5·75) 9·06 (8·53 to 9·59) 0·45 (0·21 to 0·70) 2·20 (1·90 to 2·50)

Pooled data from the Asia Pacific Cohort Studies Collaboration,14 Atherosclerosis Risk in Communities study,15 National Health and Nutrition Examination Survey III,13 and
Scottish Heart Health Extended Cohort study.12 Multiple-adjusted mean differences are, where appropriate, adjusted for age, systolic blood pressure, smoking, body-mass
index (BMI), and total cholesterol. HDL=high-density lipoprotein.

Table 2: Age-adjusted and multiple-adjusted mean difference in baseline risk factor levels among men and women with and without diabetes

Relative risk Weight

(95% CI)

Women
Takayama26 0·88 (0·36–2·16) 2·68%
APCSC (Australia and New Zealand)14 1·41 (0·95–2·08) 7·52%
NHANES III13 1·69 (0·90–3·15) 4·53%
Hisayama (2000)9 1·90 (1·20–3·00) 6·46%
APCSC (Asia)14 1·93 (1·45–2·58) 9·25%
Hisayama (2010)11 2·02 (1·07–3·81) 4·42%
Dubbo28 2·05 (1·14–3·66) 4·93%
EPIC–Norfolk22 2·12 (1·15–3·89) 4·66%
JPHC7 2·19 (1·53–3·12) 8·05%
Iso et al21 2·20 (1·20–4·00) 4·74%
Rancho Bernado24 2·20 (1·00–4·50) 3·50%
Sievers et al23 2·30 (0·70–8·30) 1·56%
DECODE10 2·37 (1·46–3·84) 6·11%
Framingham Offspring25 2·70 (0·80–9·16) 1·60%
Renfrew/Paisley20 2·83 (1·63–4·90) 5·29%
ARIC15 3·16 (2·55–3·91) 10·64%
SHHEC12 3·64 (2·29–5·79) 6·40%
Kuopio and North Karelia8 3·91 (2·68–5·72) 7·67%
Total (I2 =49·8%, p=0·009) 2·28 (1·93–2·69) 100·00%

Men
Sievers et al23 0·80 (0·30–2·40) 1·47%
Dubbo28 1·34 (0·74–2·45) 3·75%
APCSC (Asia)14 1·43 (1·23–1·66) 13·67%
EPIC–Norfolk22 1·45 (0·84–2·49) 4·35%
NHANES III13 1·49 (0·66–3·34) 2·30%
Renfrew/Paisley20 1·52 (0·72–3·21) 2·62%
APCSC (Australia and New Zealand)14 1·62 (1·18–2·22) 8·50%
JPHC7 1·64 (1·21–2·23) 8·77%
Takayama26 1·65 (0·99–2·76) 4·73%
Iso et al21 1·80 (1·00–3·20) 3·92%
Hisayama (2000)9 1·80 (1·20–2·60) 6·83%
Rancho Bernado24 1·80 (1·00–3·20) 3·92%
Framingham Offspring25 1·93 (0·86–4·33) 2·29%
ARIC15 2·00 (1·57–2·54) 10·68%
Hisayama (2010)11 2·54 (1·40–4·63) 3·76%
Kuopio and North Karelia8 2·66 (1·82–3·88) 7·01%
SHHEC12 2·80 (1·77–4·44) 5·50%
DECODE10 3·01 (1·95–4·64) 5·94%
Total (I2 =40·5%, p=0·039) 1·83 (1·60–2·08) 100·00%

0·5 1·0 1·5 3·0 4·0 5·0

Diabetes less risky Diabetes more risky

Figure 2: Maximum-adjusted pooled relative risk for any stroke, comparing individuals with diabetes to those without diabetes
Box sizes are in proportion to study weights. Asia Pacific Cohort Studies Collaboration (APCSC) provided separate estimates for cohorts from Asia and Australia and
New Zealand. NHANES III=National Health And Nutrition Examination Survey III. EPIC–Norfolk=European Prospective Investigation into Cancer, Norfolk. JPHC=Japan
Public Health Center study. DECODE=Diabetes Epidemiology: COllaborative analysis of Diagnostic criteria in Europe. ARIC=Atherosclerosis Risk in Communities study.
SHHEC=Scottish Heart Health Extended Cohort study.

1976 www.thelancet.com Vol 383 June 7, 2014

 Articles

and multiple-adjusted mean differences in cardio- (1·88–3·00) for women and 1·90 (1·61–2·25) for men
vascular risk factors for men and women with and in the multiple-adjusted models.
without diabetes from those studies for which The pooled maximum-adjusted RR for diabetes was
individual participant data were available. Compared significantly higher in women than in men (ratio of
with individuals without diabetes, those with diabetes RRs 1·27, 95% CI 1·10–1·46; figure 3). There was no
had higher systolic blood pressure, total cholesterol, evidence of publication bias (p=0·65; appendix p 6).
BMI, and waist circumference, and lower HDL Exclusion of the three studies that provided only age-
cholesterol. The mean difference for each of these risk adjusted estimates had no appreciable effect on the pooled
factors—particularly the anthropometric variables— ratio of RRs (multiple-adjusted ratio of RRs 1·26,
was greater for women than for men (table 2, appendix [1·09–1·47]). The pooled ratio of RRs did not vary
pp 2–3). substantially by duration of study follow-up (p for
The overall maximum-adjusted pooled RR based on heterogeneity=0·44); proportion of stroke events (p=0·93);
all available data for combined fatal and non-fatal stroke women-to-men ratio of stroke event rate (p=0·47); baseline
associated with diabetes was 2·28 (95% CI 1·93–2·69) prevalence of diabetes (p=0·59); or women-to-men ratio of
for women and 1·83 (1·60–2·08) for men (figure 2). diabetes prevalence (p=0·07; appendix p 7). Additionally,
The I² statistic for heterogeneity between studies was the ratio of RRs did not differ significantly by region age,
50% in women and 41% in men, suggesting substantial smoking status, year of study baseline, stroke subtype,
between-study heterogeneity. Exclusion of the three method of diabetes ascertainment, or data source
studies with results adjusted for age only did not reduce (figure 4).
the between-study heterogeneity and did not change The pooled sex-specific RR estimates for fatal stroke
the RR estimates (2·27 [1·88–2·74] for women and associated with diabetes were 2·29 (95% CI 1·73–3·04)
1·87 [1·61–2·16] for men; appendix p 4). To determine for women and 1·74 (1·45–2·08) for men (appendix
whether adjustment for other cardiovascular risk p 8). The corresponding ratio of RRs (women to men)
factors might have had a stronger effect in women was 1·32 (0·97–1·79; appendix p 9). The I² statistic for
than in men, we repeated the analyses using data heterogeneity between studies was 27%, with no
from 55 cohorts (738 082 individuals, and 10 311 strokes) evidence of publication bias (p=0·76; appendix p 10).
for which both age-adjusted and multiple-adjusted The pooled ratio of RRs did not vary by duration of
results were available (appendix p 5). Overall, study follow-up (p=0·34), proportion of stroke events
adjustment resulted in a similar amount of attenuation (p=0·79), women-to-men ratio of stroke event rate
(of less than 10%) in both women and men: the age- (p=0·84), baseline prevalence of diabetes (p=0·24), or
adjusted RR was 2·46 (1·83–3·32) for women and women-to-men ratio of diabetes prevalence (p=0·052;
1·96 (1·64–2·36) for men, decreasing to 2·37 appendix p 11).

Relative risk Weight

(95% CI)

Takayama26 0·53 (0·19–1·50) 1·90%

DECODE10 0·79 (0·41–1·51) 4·80%
Hisayama (2010)11 0·80 (0·33–1·90) 2·66%
APCSC (Australia and New Zealand)14 0·87 (0·53–1·44) 8·09%
Hisayama (2000)9 1·06 (0·58–1·92) 5·63%
NHANES III13 1·13 (0·41–3·14) 1·95%
Iso et al21 1·22 (0·53–2·82) 2·89%
Rancho Bernado24 1·22 (0·47–3·16) 2·24%
SHHEC12 1·30 (0·68–2·49) 4·76%
JPHC7 1·34 (0·84–2·14) 9·19%
APCSC (Asia)14 1·35 (0·98–1·87) 19·17%
Framingham Offspring25 1·40 (0·32–6·04) 0·95%
EPIC–Norfolk22 1·46 (0·65–3·31) 3·04%
Kuopio and North Karelia8 1·47 (0·86–2·51) 7·06%
Dubbo28 1·53 (0·66–3·53) 2·90%
ARIC15 1·58 (1·15–2·18) 19·64%
Renfrew/Paisley20 1·86 (0·74–4·71) 2·35%
Sievers et al23 2·88 (0·57–14·46) 0·78%
Total (I2 =0·0%, p=0·752) 1·27 (1·10–1·46) 100·00%

0·25 1·0 1·5 3·0 6·0

Higher relative risk in men Higher relative risk in women

Figure 3: Maximum-adjusted women-to-men ratio of relative risks for any stroke, comparing individuals with diabetes to those without diabetes
Box sizes are in proportion to study weights. Asia Pacific Cohort Studies Collaboration (APCSC) provided separate estimates for cohorts from Asia and Australia and
New Zealand. DECODE=Diabetes Epidemiology: COllaborative analysis of Diagnostic criteria in Europe. NHANES III=National Health And Nutrition Examination
Survey III. SHHEC=Scottish Heart Health Extended Cohort study. JPHC=Japan Public Health Center study. EPIC–Norfolk=European Prospective Investigation into
Cancer, Norfolk. ARIC=Atherosclerosis Risk in Communities study.

www.thelancet.com Vol 383 June 7, 2014 1977

 Articles

Ratio of relative p value for

lipoprotein (LDL) and HDL cholesterol, fasting glucose,
risks (95% CI) interaction and glycated haemoglobin.43,44 However, sex differences in
Region cardiovascular risk factors are unlikely to wholly account
Asia 1·20 (0·96–1·50)
0·51
for the greater excess risk of diabetes-related stroke seen
Non-Asia 1·32 (1·10–1·59)
Age
in women. In our analysis, adjustment for major
<60 years 1·44 (1·11–1·87) cardiovascular risk factors resulted in a similar degree of
0·35
≥60 years 1·11 (0·81–1·52) attenuation in the age-adjusted estimate, of less than
Smoking status
Non-smokers 1·24 (0·94–1·62) 10%, for both women and men. This result implies that
0·69
Current smokers 1·40 (0·87–2·26) although some residual confounding probably persists
Stroke subtype
Ischaemic 1·25 (1·01–1·54)
(as in all observational cohorts), it is marginal and
0·20 unlikely to differentially affect women more than men.
Haemorrhagic 1·83 (0·76–4·39)
Year of study baseline Sex differences in diabetes-related changes in the
Pre-1985 1·35 (1·08–1·69)
1985 onwards 1·21 (1·01–1·46)
0·47 progression of atherosclerosis or in more novel risk
Method of diabetes ascertainment factors (such as markers of coagulation and inflammation,
Self reports included 1·23 (1·00–1·52)
Medical diagnosis only 1·30 (1·08–1·58)
0·69 lipid peroxidation, and endothelial function), for which
Data source we had insufficient data to explore, might also be
Aggregated data 1·21 (0·98–1·49) involved.30,45–48 Mansfield and colleagues46 investigated sex
0·53
Individual participant data 1·33 (1·09–1·61)
differences in coagulation and fibrinolysis in individuals
0·5 1·0 1·5 3·0 4·0 5·0
with diabetes and reported that women had significantly
Favours higher relative Favours higher relative higher factor VII and plasminogen activator inhibitor 1
risk in men risk in women (also known as serpin E1) activity than men.46 Similarly,
Figure 4: Sensitivity analyses
data from the British Regional Heart Study and the
British Women’s Heart Health Study34 showed a greater
adverse effect of diabetes in women than in men on
Discussion markers of coagulation, fibrinolysis, lipids, and blood
In this pooled analysis of 64 cohorts, with data for more pressure, which were mediated by greater changes in
than three-quarters of a million individuals and more central adiposity and insulin resistance in women.34 In
than 12 000 fatal and non-fatal stroke events, diabetes was support of this finding, results from other studies suggest
a stronger risk factor for stroke in women than in men. that despite women having a greater body-fat percentage
Compared with men with diabetes, women with diabetes per unit of BMI than men,49 they conversely have greater
had a 27% greater RR for stroke when baseline differences insulin sensitivity at sites associated with insulin
in other major cardiovascular risk factors were taken into resistance, namely the liver and skeletal muscle.50 These
account. This sex difference in diabetes-related stroke data support the idea that women have further to travel
risk was apparent and consistent across a wide range of metabolically and physiologically than men to transition
prespecified subgroups. These findings add to the from normoglycaemia to a deranged glycaemic state;34 in
accumulating evidence for appreciable and clinically other words, the metabolic and vascular risk factor profile
relevant differences in how diabetes affects the risk of of women has to deteriorate to a greater extent than that
cardiovascular disease in men and women.6 of men before the onset of overt diabetes occurs.
In our analysis, as in previous reports,29–37 differences in Women have a much more favourable cardiovascular
major cardiovascular risk factors in individuals with and risk profile than men, but this pattern can be reversed
without diabetes were greater among women than among with deterioration in glycaemic control. A previous study51
men. Historically, men with diabetes or cardiovascular has shown that years before the manifestation of overt
disease were diagnosed earlier and more often treated diabetes, individuals who have impaired glucose
with aspirin, statins, and antihypertensive treatments tolerance—particularly women—have a more adverse
than were women, and were therefore more likely to cardiovascular profile than those with normal glucose
achieve recommended treatment targets for risk tolerance—the so-called ticking-clock hypothesis.
factors.38–42 As such, the excess risk of stroke in women Findings from studies in non-diabetic men and women
with diabetes might be expected to be an underestimate before they transition to a state of impaired glucose
of the real sex difference, especially in older studies in tolerance suggest that women who progress from
which sex disparities in treatment are most pronounced. normoglycaemia to prediabetes have greater endothelial
However, our analysis did not show a difference in the dysfunction than men, as well as more severe hypertension
excess risk of diabetes-related stroke in women between and more fibrinolysis and thrombosis.52 We propose that
cohorts with baseline data collection before 1985 and the diabetes-related excess risk of stroke in women is due
from 1985 onwards. Moreover, data from more recent to their having a chronically raised cardiovascular risk
studies suggest that even when treated similarly, women profile in the prediabetic state, which is more likely to go
with diabetes remain less likely than men to achieve undetected and therefore untreated than in men, rather
target values for systolic blood pressure, low-density than by any substantial sex difference in the effects and

1978 www.thelancet.com Vol 383 June 7, 2014

 Articles

complications of diabetes per se. If confirmed, the why diabetes confers a stronger coronary and cerebro-
implementation of sex-specific interventions before vascular hazard in women than in men.
diabetes becomes manifest—such as increased screening Contributors
for prediabetes in women—could have a substantial effect SAEP searched the scientific literature, did the statistical analyses,
on the prevention of vascular events. participated in data interpretation, and drafted the report. RRH and MW
conceived the study, contributed data, participated in data interpretation,
This meta-analysis provides the most definitive and and made important revisions to the draft report.
convincing evidence so far of a sex difference in diabetes-
Conflicts of interest
related risk of stroke. The findings were robust and We declare that we have no conflicts of interest.
applicable across a broad range of populations.
Acknowledgments
Limitations of our study are inherent to the use of SAEP is supported by a Niels Stensen fellowship and MW by an
published data, including the absence of standardisation Australian National Health and Medical Research Council fellowship. We
in study design, duration, endpoint definition, study thank Leon Simons and Shino Oba for contributing additional tabulated
populations, and the extent of adjustment for data from the Dubbo and the Takayama studies, respectively.

confounding across studies. Moreover, the studies References

1 Mendis S, Puska P, Norrving B (eds). Global atlas on cardiovascular
included in this meta-analysis generally did not report disease prevention and control. Geneva: World Health
on, or adjust for, the use of antidiabetic drugs. But Organization, 2011.
because women were historically more likely to be 2 Danaei G, Finucane MM, Lu Y, et al, for the Global Burden of
Metabolic Risk Factors of Chronic Diseases Collaborating Group
undertreated than men, the net effect is likely to have (Blood Glucose). National, regional, and global trends in fasting
been to underestimate the true excess risk of stroke in plasma glucose and diabetes prevalence since 1980: systematic
women with diabetes. Thus our estimate might be analysis of health examination surveys and epidemiological
studies with 370 country-years and 2·7 million participants.
regarded as a lower bound for this excess risk. Lancet 2011; 378: 31–40.
An important limitation of these data, which warrants 3 WHO. Diabetes (fact sheet No312). http://www.who.int/
further investigation by future studies, was our inability mediacentre/factsheets/fs312/en/index.html (accessed Feb 4, 2013).
to examine the potential contribution that a woman’s 4 Whiting DR, Guariguata L, Weil C, Shaw J. IDF diabetes atlas:
global estimates of the prevalence of diabetes for 2011 and 2030.
obstetric and gynaecological history might have on the Diabetes Res Clin Pract 2011; 94: 311–21.
excess risk of stroke. Additionally, because a standard 5 The Lancet. Taking sex into account in medicine. Lancet 2011;
definition of diabetes was not used across studies, 378: 1826.
6 Huxley R, Barzi F, Woodward M. Excess risk of fatal coronary heart
ascertainment of diabetes probably varied between—but disease associated with diabetes in men and women: meta-analysis
importantly, not within—studies. Hence, the use of the of 37 prospective cohort studies. BMJ 2006; 332: 73–78.
ratio of RRs (comparing women with men) remains 7 Cui R, Iso H, Yamagishi K, et al. Diabetes mellitus and risk of
stroke and its subtypes among Japanese: the Japan public health
valid. Finally, we were unable to assess whether the center study. Stroke 2011; 42: 2611–14.
excess risk of stroke in women with diabetes varied with 8 Hu G, Jousilahti P, Qiao Q, Katoh S, Tuomilehto J. Sex differences
different indices of glycaemic control or duration of in cardiovascular and total mortality among diabetic and non-
diabetes because of insufficient data for these variables. diabetic individuals with or without history of myocardial
infarction. Diabetologia 2005; 48: 856–61.
Future prospective cohort studies can address many of 9 Tanizaki Y, Kiyohara Y, Kato I, et al. Incidence and risk factors for
the limitations of this meta-analysis by exploring the role subtypes of cerebral infarction in a general population: the
of sex differences in individuals in the prediabetic state Hisayama study. Stroke 2000; 31: 2616–22.
10 Hyvärinen M, Tuomilehto J, Laatikainen T, et al. The impact of
and their contribution to vascular risk. Ideally, such diabetes on coronary heart disease differs from that on
studies would include a large number of men and ischaemic stroke with regard to the gender. Cardiovasc Diabetol
women free from diabetes and stroke at recruitment, 2009; 8: 17.
11 Doi Y, Ninomiya T, Hata J, et al. Impact of glucose tolerance status
with detailed information about risk factors, including on development of ischemic stroke and coronary heart disease in a
medical treatment, taken at baseline and several times general Japanese population: the Hisayama study. Stroke 2010;
during follow-up, which should continue until a 41: 203–09.
12 Woodward M, Brindle P, Tunstall-Pedoe H, for the SIGN group on
sufficient number of stroke events had occurred. Detailed risk estimation. Adding social deprivation and family history to
gynaecological histories should also be recorded to allow cardiovascular risk assessment: the ASSIGN score from the
the effect of complications in pregnancy and menopause Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;
93: 172–76.
to be studied.
13 Plan and operation of the Third National Health and Nutrition
Although costly, such studies would be free of the Examination Survey, 1988–94. Series 1: programs and collection
major drawbacks of our meta-analysis. They would allow procedures. Vital Health Stat 1 1994; 32: 1–407.
for detailed exploration of the sex-specific differences in 14 Woodward M, Barzi F, Martiniuk A, et al. Cohort profile: the Asia
Pacific Cohort Studies Collaboration. Int J Epidemiol 2006;
the cardiovascular risk trajectories between individuals 35: 1412–16.
who develop impaired glucose tolerance and overt 15 The ARIC investigators. The Atherosclerosis Risk in Communities
diabetes and those who remain normoglycaemic during (ARIC) study: design and objectives. Am J Epidemiol 1989;
129: 687–702.
follow-up, and how these sex differences contribute to 16 Huxley RR, Woodward M. Cigarette smoking as a risk factor for
stroke risk. Population-based big data studies, such as coronary heart disease in women compared with men: a systematic
the UK Biobank, are well suited to investigation of these review and meta-analysis of prospective cohort studies. Lancet 2011; For the UK Biobank website see
378: 1297–305.
issues and, on maturation, should be mined to find out http://www.ukbiobank.ac.uk/

www.thelancet.com Vol 383 June 7, 2014 1979

 Articles

17 Woodward M. Epidemiology: study design and data analysis, 37 Leosdottir M, Willenheimer R, Persson M, Nilsson PM. The
2nd edn. Boca Raton: Chapman & Hall and CRC, 2005. association between glucometabolic disturbances, traditional
18 Higgins JP, Thompson SG. Quantifying heterogeneity in a cardiovascular risk factors and self-rated health by age and gender:
meta-analysis. Stat Med 2002; 21: 1539–58. a cross-sectional analysis within the Malmö Preventive Project.
19 Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring Cardiovasc Diabetol 2011; 10: 118.
inconsistency in meta-analyses. BMJ 2003; 327: 557–60. 38 Ferrara A, Mangione CM, Kim C, et al, for the Translating
20 Hart CL, Hole DJ, Smith GD. Comparison of risk factors for stroke Research Into Action for Diabetes Study Group. Sex disparities in
incidence and stroke mortality in 20 years of follow-up in men and control and treatment of modifiable cardiovascular disease risk
women in the Renfrew/Paisley study in Scotland. Stroke 2000; factors among patients with diabetes: Translating Research
31: 1893–96. Into Action for Diabetes (TRIAD) study. Diabetes Care 2008;
31: 69–74.
21 Iso H, Imano H, Kitamura A, et al. Type 2 diabetes and risk of
non-embolic ischaemic stroke in Japanese men and women. 39 Krämer HU, Raum E, Rüter G, et al. Gender disparities in diabetes
Diabetologia 2004; 47: 2137–44. and coronary heart disease medication among patients with type 2
diabetes: results from the DIANA study. Cardiovasc Diabetol 2012;
22 Myint PK, Sinha S, Luben RN, Bingham SA, Wareham NJ,
11: 88.
Khaw KT. Risk factors for first-ever stroke in the EPIC-Norfolk
prospective population-based study. Eur J Cardiovasc Prev Rehabil 40 Wexler DJ, Grant RW, Meigs JB, Nathan DM, Cagliero E.
2008; 15: 663–69. Sex disparities in treatment of cardiac risk factors in patients with
type 2 diabetes. Diabetes Care 2005; 28: 514–20.
23 Sievers ML, Nelson RG, Knowler WC, Bennett PH. Impact of
NIDDM on mortality and causes of death in Pima Indians. 41 Gouni-Berthold I, Berthold HK, Mantzoros CS, Böhm M,
Diabetes Care 1992; 15: 1541–49. Krone W. Sex disparities in the treatment and control of
cardiovascular risk factors in type 2 diabetes. Diabetes Care 2008;
24 Barrett-Connor E, Khaw KT. Diabetes mellitus: an independent risk
31: 1389–91.
factor for stroke? Am J Epidemiol 1988; 128: 116–23.
42 Winston GJ, Barr RG, Carrasquillo O, Bertoni AG, Shea S. Sex and
25 Najarian RM, Sullivan LM, Kannel WB, Wilson PW, D’Agostino RB,
racial/ethnic differences in cardiovascular disease risk factor
Wolf PA. Metabolic syndrome compared with type 2 diabetes
treatment and control among individuals with diabetes in the
mellitus as a risk factor for stroke: the Framingham Offspring
Multi-Ethnic Study of Atherosclerosis (MESA). Diabetes Care 2009;
study. Arch Intern Med 2006; 166: 106–11.
32: 1467–69.
26 Oba S, Nagata C, Nakamura K, Takatsuka N, Shimizu H. Self-
43 Franzini L, Ardigò D, Cavalot F, et al, and the IT Study Group of the
reported diabetes mellitus and risk of mortality from all causes,
Italian Society of Diabetology. Women show worse control of type 2
cardiovascular disease, and cancer in Takayama: a population-based
diabetes and cardiovascular disease risk factors than men: from the
prospective cohort study in Japan. J Epidemiol 2008; 18: 197–203.
MIND.IT Study Group of the Italian Society of Diabetology.
27 Asia Pacific Cohort Studies Collaboration. The effects of diabetes on Nutr Metab Cardiovasc Dis 2013; 23: 235–41.
the risks of major cardiovascular diseases and death in the Asia-
44 Penno G, Solini A, Bonora E, et al, and the Renal Insufficiency And
Pacific region. Diabetes Care 2003; 26: 360–66.
Cardiovascular Events (RIACE) study group. Gender differences in
28 Simons LA, Simons J, Friedlander Y, McCallum J. A comparison cardiovascular disease risk factors, treatments and complications in
of risk factors for coronary heart disease and ischaemic stroke: patients with type 2 diabetes: the RIACE Italian multicentre study.
the Dubbo study of Australian elderly. Heart Lung Circ 2009; J Intern Med 2013; 274: 176–91.
18: 330–33.
45 Evans RW, Orchard TJ. Oxidized lipids in insulin-dependent
29 Göbl CS, Brannath W, Bozkurt L, et al. Sex-specific differences in diabetes mellitus: a sex-diabetes interaction? Metabolism 1994;
glycemic control and cardiovascular risk factors in older patients 43: 1196–200.
with insulin-treated type 2 diabetes mellitus. Gend Med 2010;
46 Mansfield MW, Heywood DM, Grant PJ. Sex differences in
7: 593–99.
coagulation and fibrinolysis in white subjects with non-insulin-
30 Howard BV, Cowan LD, Go O, Welty TK, Robbins DC, Lee ET, for dependent diabetes mellitus. Arterioscler Thromb Vasc Biol 1996;
the Strong Heart Study Investigators. Adverse effects of diabetes on 16: 160–64.
multiple cardiovascular disease risk factors in women. The Strong
47 Ossei-Gerning N, Wilson IJ, Grant PJ. Sex differences in
Heart Study. Diabetes Care 1998; 21: 1258–65.
coagulation and fibrinolysis in subjects with coronary artery
31 Rivellese AA, Riccardi G, Vaccaro O. Cardiovascular risk in disease. Thromb Haemost 1998; 79: 736–40.
women with diabetes. Nutr Metab Cardiovasc Dis 2010;
48 Steinberg HO, Paradisi G, Cronin J, et al. Type II diabetes abrogates
20: 474–80.
sex differences in endothelial function in premenopausal women.
32 Vaccaro O, Boemi M, Cavalot F, et al, for the MIND-IT Study Circulation 2000; 101: 2040–46.
Group. The clinical reality of guidelines for primary prevention of
49 Garaulet M, Pérex-Llamas F, Fuente T, Zamora S, Tebar FJ.
cardiovascular disease in type 2 diabetes in Italy. Atherosclerosis
Anthropometric, computed tomography and fat cell data in an
2008; 198: 396–402.
obese population: relationship with insulin, leptin, tumor necrosis
33 Walden CE, Knopp RH, Wahl PW, Beach KW, Strandness E Jr. factor-alpha, sex hormone-binding globulin and sex hormones.
Sex differences in the effect of diabetes mellitus on lipoprotein Eur J Endocrinol 2000; 143: 657–66.
triglyceride and cholesterol concentrations. N Engl J Med 1984;
50 Geer EB, Shen W. Gender differences in insulin resistance, body
311: 953–59.
composition, and energy balance. Gend Med 2009;
34 Wannamethee SG, Papacosta O, Lawlor DA, et al. Do women 6 (suppl 1): 60–75.
exhibit greater differences in established and novel risk factors
51 Haffner SM, Miettinen H, Stern MP. Relatively more atherogenic
between diabetes and non-diabetes than men? The British Regional
coronary heart disease risk factors in prediabetic women than in
Heart Study and British Women’s Heart Health Study. Diabetologia
prediabetic men. Diabetologia 1997; 40: 711–17.
2012; 55: 80–87.
52 Donahue RP, Rejman K, Rafalson LB, Dmochowski J, Stranges S,
35 Duggirala MK, Cuddihy RM, Cuddihy MT, Nyman MA, Naessens JM,
Trevisan M. Sex differences in endothelial function markers before
Pankratz VS. Women with diabetes have poorer control of blood
conversion to pre-diabetes: does the clock start ticking earlier
pressure than men. J Womens Health (Larchmt) 2005; 14: 418–23.
among women? The Western New York Study. Diabetes Care 2007;
36 Nilsson PM, Theobald H, Journath G, Fritz T. Gender differences in 30: 354–59.
risk factor control and treatment profile in diabetes: a study in
229 Swedish primary health care centres. Scand J Prim Health Care
2004; 22: 27–31.

1980 www.thelancet.com Vol 383 June 7, 2014

Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.