Epidemiology/Health Services/Psychosocial Research

O R I G I N A L A R T I C L E

The Relationship Between Glucose and

Incident Cardiovascular Events
A metaregression analysis of published data from 20 studies of 95,783
individuals followed for 12.4 years
MARIO COUTINHO, MD, PHD YONG WANG, PHD without regard to the risk of cardiovascular
HERTZEL C. GERSTEIN, MD, MSC SALIM YUSUF, MBBS, DPHIL disease. If there is a glucose threshold for
cardiovascular disease, it may therefore be
lower than that for diabetes or even possi-
bly impaired glucose tolerance. We
explored this possibility by doing a sys-
OBJECTIVE — To assess the relationship between nondiabetic glucose levels and cardio- tematic overview and metaregression
vascular risk. analysis of cohort studies that assessed the
relationship between glucose and the inci-
RESEARCH DESIGN AND METHODS — Three independent searches using MED- dence of cardiovascular disease.
LINE (1966–1996), followed by a manual search of the references from each retrieved article,
were conducted by two physicians and one medical librarian. Data had to be reported in at least
three quantiles or intervals so that the nature of the relationship between glucose and cardio- RESEARCH DESIGN AND
vascular events (i.e., linear or nonlinear ) could be explored, and to ensure that any incremental METHODS — All prospective studies of
cardiovascular risk was consistent across quantiles or intervals. the risk of cardiovascular disease according to
baseline glucose determinations were sought.
RESULTS — Analyzed studies comprised 95,783 people (94% male) who had 3,707 car- Three independent searches using MED-
diovascular events over 12.4 years (1,193,231 person-years). Studies reporting fasting glucose LINE (1966–1996), followed by a manual
levels (n = 6), 2-h glucose levels (n = 7), 1-h glucose levels (n = 5), and casual glucose levels
search of the references from each retrieved
(n = 4) were included. The glucose load used varied from 50 to 100 g. The highest glucose
interval for most studies included glucose values in the diabetic range. The relationship article, were conducted by two physicians
between glucose levels and the risk of a cardiovascular event was modeled for each study and and one medical librarian. The medical sub-
the -coefficients were combined. Compared with a glucose level of 4.2 mmol/l (75 mg/dl), a ject headings of “blood glucose” or “glucose
fasting and 2-h glucose level of 6.1 mmol/l (110 mg/dl) and 7.8 mmol/l (140 mg/dl) was asso- intolerance” were combined with the head-
ciated with a relative cardiovascular event risk of 1.33 (95% CI 1.06–1.67) and 1.58 (95% CI ing of “coronary artery disease,” “stroke,”
1.19–2.10), respectively. “cerebrovascular disorders,” “mortality,” or
“heart disease,” as well as “prospective study,”
CONCLUSIONS — The progressive relationship between glucose levels and cardiovascu- “cohort study,” or “follow-up.” In addition,
lar risk extends below the diabetic threshold. any other cohort studies, which reported
cardiovascular disease (myocardial infarction
Diabetes Care 22:233–240, 1999
or stroke) and mortality, were analyzed to
identify additional studies in which glucose
was measured at baseline.
ndividuals with type 2 diabetes have a total mortality is decreased by lowering

I
Published studies were included if 1)
two- to fourfold increased risk of blood glucose with insulin after an acute they included nondiabetic participants who
cardiovascular disease compared with myocardial infarction (7). Thus, in patients were not selected on the basis of preexisting
nondiabetic individuals (1–3). Recent epi- with type 2 diabetes, the blood glucose disease or health status, 2) data were
demiological studies suggest that in people concentration is a continuous and possibly reported by quantiles (defined by tertiles,
with type 2 diabetes, cardiovascular mor- modifiable cardiovascular risk factor. quartiles, etc.) or intervals of glucose values
tality is related to the degree of hypergly- The glucose thresholds used to define at baseline, and 3) the number of fatal or
cemia (4–6); one randomized controlled diabetes were chosen to identify people at nonfatal cardiovascular outcomes (sudden
trial has shown that in diabetic patients, risk for eye and kidney diseases (8–10), death, stroke, or acute myocardial infarc-
tion) and the number of people at risk
within each glucose quantile or interval
were prospectively recorded. Data had to
From the Preventive Cardiology and Therapeutics Research Program (M.C., H.C.G., Y.W., S.Y.), Hamilton
Civic Hospitals Research Center; and the Divisions of Cardiology (M.C., S.Y.) and Endocrinology and Metab- be reported in at least three quantiles or
olism (H.C.G.), McMaster University, Hamilton, Ontario, Canada; and the Department of Clinical Medicine intervals so that the nature of the relation-
(M.C.), Federal University of Santa Catarina, Florianopolis, Brazil. ship between glucose and cardiovascular
Address correspondence and reprint requests to Dr. H.C. Gerstein, Department of Medicine, Room 3V38, events (i.e., linear or nonlinear) could be
1200 Main St. West, Hamilton, Ontario, L8N 3Z5, Canada. E-mail: gerstein@fhs.csu.mcmaster.ca.
Received for publication 2 June 1998 and accepted in revised form 7 October 1998. explored, and to ensure that any incre-
A table elsewhere in this issue shows conventional and Système International (SI) units and conversion mental cardiovascular risk was consistent
factors for many substances. across quantiles or intervals.

DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999 233

Relationship of glucose to cardiovascular risk

Table 1—Characteristics of abstracted studies

Study Cardiovascular Highest glucose

Age Men Number Diabetes Number duration Cardiovascular events plus Glucose Glucose interval
Reference (years) (%) screened exclusions followed (years) deaths (n) deaths (n) method intervals (mmol/l)

Fasting
Scheidt-Nave 40–79 43 3,458 DM Hx, 3,458 14 217 217 Plasma 7 7.2–7.7
et al. (19) BCH
Shaten 35–57 100 12,866 DM meds 12,694 10.5 426 426 Serum 6 7.8
et al. (24)
Yarnell 45–59 100 4,860 DM Hx 4,349 4 NA 209 Plasma 5 7.8
et al. (29)
Schroll and 50 100 388 DM Hx 383 10 18 18 Blood 5 5.3–12.5
Hagerup (30)
Ohlson 73 100 973 DM Hx, 832 17 N/A 141 Whole 5 5
et al. (31) BCH blood
Pyorala 40–59 100 1,059 DM Hx 845 10 42 42 Blood 5 5.2–15.2
et al. (16,65)
2-h (dose)*
Stamler et al. 42–58 100 1,694 DM Hx 1,694 15 166 166 Plasma 5 6.7–23.3
(14) (100 g)
Pyorala et al. 40–59 100 1,059 DM Hx 845 10 42 42 Blood 5 5.6–18.5
(16,65) (75–90 g)
Fuller et al. 40–64 100 18,403 DM Hx 18,106 15 915 915 Capillary 5 4.8–11.1
(17) (50 g)
Balkau et al. 44–55 100 7,166 DM Hx 7,038 15.6 167 167 Blood 3 11.1
(21) (75 g)
Da Silva et al. 40–59 100 1,499 DM Hx 1,491 5 12 12 Capillary 5 6.8–17.8
(25) (100 g)
Grabauskas 45–59 100 2,455 DM Hx 2,413 10 114 114 Plasma 5 10–17.8
(33) (75 g)
Tuomilehto et al. 40 43 1,537 NA 1,537 5 51 51 Capillary 3 11.1
(22) (75 g)
1-h (dose)*
Stenhouse et al. 40–59 100 3,331 DM Hx 638 11 21 21 Plasma 5 6.8–23
(15) (50 g)
Donahue et al. 45–70 100 6,394 DM Hx 6,394 12 119 303 Serum 5 10.6–29.6
(18) (50 g)
Vaccaro et al. 34–65 100 873 DM Hx 873 19 144 144 Plasma 5 11.7 (mean)
(20) (50 g)
Reunanen et al. 40–59 100 3,351 DM Hx 3,267 4 64 64 Plasma 5 10.8–28.1
(27) (60–90 g)
Pyorala et al. 40–59 100 1,059 DM Hx 845 10 42 42 Blood 5 8.0–20.3
(16,65) (75–90 g)
Casual
Perry et al. (23) 40–59 100 7,735 DM Hx, 7,177 9.5 222 505 Serum 5 6.1
BCH
Hawthorne and 45–64 100 1,134 DM Hx 1,128 6 29 29 Whole 5 6.0–28.8
Gilmour (26) blood
Cruz-Vidal 45–64 100 8,248 DM meds 5,445 8.25 83 83 Whole 5 7.8–34.2
et al. (28) blood
Lund Haheim 40–49 100 16,172 DM Hx 16,021 18 80 80 Serum 5 6.2
et al. (32)
All studies 20–84 — 112,311† — 95,783† 12.4 3,074† 3,707† — — —
BCH, biochemical criteria for diabetes using the measured glucose level; DM Hx, history of diabetes; DM meds, on diabetes medications; NA, not available. *Dose
of oral glucose given; †does not include any study more than once.

Studies that were restricted to subjects excluded to allow exploration of the corre- diabetic levels). Studies that did not give
with a previous diagnosis of diabetes (based lation between glucose levels and the inci- quantile or interval definitions or number of
on history and/or the use of hypoglycemic dence of cardiovascular disease across the events and number of people at risk in each
agents) at the time of cohort inception were range of glucose values (from nondiabetic to quantile or interval were also excluded. In

234 DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999

 Coutinho and Associates

Table 2—Calculated regression coefficients and the reported effects of glucose and insulin on cardiovascular disease

Cardiovascular events Cardiovascular mortality Independent effects

Person- -coefficient -coefficient Glucose† Insulin
Reference years (95% CI) P value* (95% CI) P value* (P value) (P value)

Fasting
Scheidt-Nave et al. (19) 48,412 1.386 (0.646 to 2.126) 0.0003 1.386 (0.646 to 2.126) 0.0003 0.02 abcdef‡ NA
Shaten et al. (24) 133,287 0.400 ( 0.227 to 1.026) 0.22 0.400 ( 0.227 to 1.026) 0.22 0.01abcdf§ NA
Yarnell et al. (29) 17,396 1.320 (0.451 to 2.189) 0.0055 NA NA NSbcdef NA
Schroll and Hagerup (30) 3,830 0.003 ( 1.052 to 1.046) 0.995 0.003 ( 1.052 to 1.046) 0.995 NSbcde NS
Ohlson et al. (31) 14,144 0.339 ( 2.146 to 1.469) 0.71 NA NA NA NA
Pyorala et al. (16,65) 8,450 0.280 ( 0.204 to 0.764) 0.27 0.280 ( 0.204 to 0.764) 0.27 NSabcde 0.01
All 225,519 0.606 (0.113 to 1.099) 0.016 0.531 ( 0.006 to 1.067) 0.052 — —
2-h
Stamler et al. (14) 25,410 0.021 ( 0.126 to 0.168) 0.78 0.021 ( 0.126 to 0.168) 0.78 NSabcde NA
Pyorala et al. (16,65) 8,450 0.392 (0.067 to 0.716) 0.023 0.392 (0.067 to 0.716) 0.023 NSabcde 0.01
Fuller et al. (17) 271,590 0.792 (0.580 to 1.004) 0.0001 0.792 (0.580 to 1.004) 0.0001 NA NA
Balkau et al. (21) 109,792.8 1.111 (0.407 to 1.816) 0.005 1.111 (0.407 to 1.816) 0.005 NA 0.005
Da Silva et al. (25) 7,455 0.201 ( 0.520 to 0.922) 0.59 0.201 ( 0.520 to 0.922) 0.59 NSabcde NA
Grabauskas (33) 24,130 0.470 (0.232 to 0.707) 0.0002 0.470 (0.232 to 0.707) 0.0002 0.01abcdef NA
Tuomilehto et al. (22) 7,685 1.095 (0.362 to 1.829) 0.005 1.095 (0.362 to 1.829) 0.005 NA NA
All 454,512.8 0.531 (0.204 to 0.858) 0.0015 0.531 (0.204 to 0.858) 0.0015 — —
1-h
Stenhouse et al. (15) 7,018 0.102 ( 0.305 to 0.509) 0.63 0.102 ( 0.305 to 0.509) 0.63 NSabcde NA
Donahue et al. (18) 76,728 0.136 (0.057 to 0.215) 0.001 0.278 (0.159 to 0.397) 0.0001 0.001 abcdef NA
Vaccaro et al. (20) 16,587 0.501 (0.207 to 0.795) 0.001 0.501 (0.207 to 0.795) 0.001 0.05abcdef NA
Reunanen et al. (27) 13,068 0.114 ( 0.068 to 0.296) 0.23 0.114 ( 0.068 to 0.296) 0.23 NSabcde NA
Pyorala et al. (16,65) 8,450 0.499 (0.197 to 0.801) 0.002 0.499 (0.197 to 0.801) 0.002 NSabcde 0.01
All 121,851 0.242 (0.084 to 0.400) 0.0013 0.287 (0.140 to 0.433) 0.0001 — —
Casual
Perry et al. (23) 68,181.5 0.616 ( 0.148 to 1.380) 0.12 0.255 ( 1.417 to 0.907) 0.67 NA NA
Hawthorne and Gilmour (26) 6,768 0.073 ( 0.202 to 0.348) 0.61 0.073 ( 0.202 to 0.348) 0.61 NSabcde NA
Cruz-Vidal et al. (28) 44,921.3 0.248 (0.032 to 0.465) 0.046 0.248 (0.032 to 0.464) 0.046 NA NA
Lund Haheim et al. (32) 288,378 1.026 ( 0.935 to 2.978) 0.31 1.026 ( 0.935 to 2.978) 0.31 NSac NA
All 408,248.8 0.157 ( 0.092 to 0.405) 0.22 0.085 ( 0.172 to 0.341) 0.52 — —
*Goodness of fit; †P values are shown for studies that calculated the incremental effect of increasing glucose levels on the risk of cardiovascular or total mortality
after statistical adjustment for one or more cardiovascular risk factors including aage, bblood pressure (systolic or diastolic), cBMI or weight, dlipids, esmoking, or
f other risk factors. ‡The P value was significant in men across glucose ranges and in women for glucose values 6.1 mmol/l (110 mg/dl) only; §the P value was
significant for nonsmokers. NA, not available.

studies reported in more than one publica- disease for other cardiac risk factors and of interval were tabulated within each study.
tion, the most recent study that met the the relationship between baseline insulin The midpoint of each glucose quantile or
inclusion criteria was analyzed; when levels and cardiovascular disease (if meas- interval was used in subsequent analyses as
needed, data from the other publications ured) was also abstracted (Table 2). the baseline glucose level in that quantile; if
were used to complete the database. To assess the impact of both fasting and either the lowest or highest quantile or
The following data were extracted and postprandial glucose on the incidence of interval did not have lower or upper
tabulated from each paper (Table 1): num- cardiovascular events, all studies were clas- boundaries (e.g., subjects with glucose lev-
ber of subjects, subject sex and age distri- sified into four groups according to the els 4.2 mmol/l or 6.1 mmol/1), the
bution, duration of follow-up, method of method used for glucose measurement at boundary glucose value was used (e.g., 4.2
glucose measurement, number and defini- baseline: 1) fasting glucose level, 2) 2-h or 6.1 mmol/l). A relative risk model using
tion of glucose quantiles or intervals, num- post–oral glucose load glucose level, 3) 1-h a modified Poisson regression method was
ber of individuals at risk in each glucose post–oral glucose load glucose level, and 4) used to generate three mathematical models
interval, and number of fatal or first nonfa- casual (nonfasting) glucose level. (linear, quadratic, exponential) describing
tal cardiovascular events in each interval. the relationship between this baseline glu-
Information on whether or not the indi- Statistical analysis cose level and the relative risk of cardiovas-
vidual studies statistically adjusted esti- The number of first cardiovascular events cular events during follow-up for each
mates of the relationship between baseline and the number of person-years of follow- study. The relative risk was defined as the
glucose levels and incident cardiovascular up for each baseline glucose quantile or ratio of the cardiovascular event rate within

DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999 235

Relationship of glucose to cardiovascular risk

Figure 1—The observed crude relative risk for cardiovascular events according to the midpoint of each glucose interval for every included study (compared
with the lowest glucose interval) is shown. Results for studies with data that allowed calculation of the relative risk for comparable glucose values
oupedare gr
together. Four studies reported data that allowed calculation of the crude relative risks for midpoint glucose values between 4.4 and 4.7 mmol/l: studies of
fasting glucose A),
( studies of 2-h glucoseB),( studies of 1-h glucoseC),
( and studies of casual glucoseD).
(

each quantile or interval to the cardiovas- model that provided the best fit of the com- physically fit (34), 2) a second report of a
cular event rate at a blood glucose concen- bined data from studies in each glucose cat- previously included study (36), 3) nonre-
tration of 4.2 mmol/l (75 mg/dl), at which egory (fasting, 1-h, 2-h, and casual) was porting of glucose ranges within quantiles
the relative risk was set at ‘one’ before the then chosen for that category; statistical or intervals (37), 4) division of data into 3
papers were analyzed. The -coefficients of homogeneity was assessed according to glucose quantiles or intervals (38,39), 5)
each of these three models, weighted by the Cochran (13). nonreporting of number of individuals at
size of each study, were calculated for all This analysis was done for both car- risk in each glucose quantile or interval
studies in each glucose category (fasting, diovascular mortality and for the number (35,40,41), and 6) nonreporting of crude
2-h, 1-h, or casual). A general random of first cardiovascular events. For studies cardiovascular outcomes in each glucose
effects parametric approach (11) was used that only reported cardiovascular mortality, quantile or interval (42). The remaining 20
to combine the individual -coefficients death was considered to be the first event. studies are described in Tables 1 and 2.
into an overall - coefficient for each glu- Illustrative relative risks and confidence They comprised 95,783 persons and
cose category. Thus, for each glucose cate- intervals at any specified glucose level rep- 1,193,231 person-years of follow-up. Only
gory, three different mathematical models resent the range within which the com- four studies (18,23,29,31) reported inci-
were generated from the combined data bined curve would fall at that glucose level. dent cardiovascular events; 19 studies
describing the relationship between glucose reported cardiovascular mortality data. A
and cardiovascular disease. The degree to RESULTS — A total of 29 citations total of 3,707 events (3,074 cardiovascular
which each model described the combined (14–42) were initially selected; 9 were deaths) were recorded. The mean weighted
data was assessed by comparing the sum of excluded from the analysis (34–42). Rea- follow-up duration was 12.4 years (range
the deviances for studies in each category sons for exclusion included the following: 4–19 years), and 18 studies enrolled only
(using each model) with a 2 test (12). The 1) exclusion of subjects who were not middle-aged men.

236 DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999

 Coutinho and Associates

Because individual subject data were

not available for this analysis, the risk of a
cardiovascular event with increasing glu-
cose levels could not be adjusted for the
presence or absence of other cardiovascular
risk factors. Such an adjustment, however,
was originally reported in 14 of the 20
studies included. Of these studies, five
showed a significant effect of glucose even
after adjusting for one or more of age,
blood pressure, BMI, weight, lipid levels, or
smoking habits (Table 2).
Two of the three included studies
(16,21) that also reported insulin levels
reported a relationship with cardiovascular
events (Table 2). The crude glucose data
abstracted from both these studies were
also consistent with a positive relationship
between glucose and cardiovascular events
Figure 2—The -coefficients generated from the exponential model of the data of each study and (the
Table 2).
combined -coefficient for fasting and 2-h postprandial glucose values are displayed.
CONCLUSIONS — This systematic
overview and metaregression analysis
Of the included studies, six reported mg/dl), a fasting glucose of 6.1 mmol/l included published cohort studies of non-
fasting glucose levels, seven studies (110 mg/dl, the threshold value for the diabetic participants. It found a graded rela-
reported 2-h glucose levels, five studies classification of impaired fasting glucose tionship between the initial fasting and
reported 1-h glucose levels, and four stud- [10]) was associated with a relative risk of postprandial glucose level and the subse-
ies reported casual glucose levels (Table 1). cardiovascular events of 1.33 (95% CI quent 12-year occurrence of a cardiovas-
The glucose doses used in the glucose tol- 1.06–1.67); a 2-h glucose of 7.8 mmol/l cular event; this relationship was apparent
erance tests varied from 50 to 100 g. Most (140 mg/dl, the threshold value for for glucose levels that were below the dia-
studies reported events according to five impaired glucose tolerance) was associated betic threshold (Fig. 1). As fasting glucose
baseline glucose quantiles or intervals. The with a relative risk of cardiovascular events values have the least variability, estimates of
highest glucose interval for most studies of 1.58 (95% CI 1.19–2.10). the glucose–cardiovascular event relation-
included glucose values in the diabetic To determine if the observed glu- ship based on the fasting glucose studies
range. The observed relative risks for car- cose–cardiovascular disease relationship may be more precise than estimates based
diovascular events according to the mid- was due to the inclusion of subjects with
point of each glucose interval in the undiagnosed diabetes in the top quantile or
individual studies is graphically displayed interval, the analysis was repeated without
in Fig. 1. the data from this interval for those cita-
An analysis of the three different regres- tions reporting fasting and 2-h glucose
sion models (linear, quadratic, exponential) data, in which removal of the top interval
that were generated from the studies in removed most subjects with glucose levels
each of the four glucose categories revealed above the diabetic threshold but retained
that the exponential model provided the subjects with glucose levels at or below
best fit for the combined data from the this threshold. Therefore, only citations
studies in each of the four categories (P reporting fasting glucose data in which the
homogeneity 0.5), regardless of whether upper limit of the second highest interval
cardiovascular mortality or cardiovascular was between 6.1 and 7.8 mmol/l and cita-
events (including mortality) were analyzed. tions reporting 2-h glucose data in which
Therefore, this model was used for subse- the upper limit of the second highest inter-
quent analyses (Table 2). val was between 7.8 and 11.1 mmol/l were
Using this model, nine of the included included in this analysis. This analysis
studies (Table 2) showed a strong relation- showed that with removal of data from the
ship between glucose quantile and the risk top glucose interval, the exponential rela-
of cardiovascular events (P 0.01). The tionship was maintained, there was a sug-
combined data for each glucose category gestive trend (P = 0.056) between fasting Figure 3—The curves and 95% CIs (within
(Fig. 2) showed a consistent relationship (P glucose and cardiovascular events, and which the curve lies) generated from the model and
0.02). Using the modeled curves illus- there was a significant relationship between the combined -coeficient for fasting A),( and
trated in Fig. 3, compared with the refer- 2-h glucose (P = 0.00064) and cardiovas- 2-h postprandial B)( glucose values are displayed
ence fasting glucose of 4.2 mmol/l (75 cular events (Table 3). (relative risk = exp( (glucose – 4.2)/4.2).

DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999 237

Relationship of glucose to cardiovascular risk

Table 3—Calculated regression coefficients for the relationship between glucose and which (either with or without glucose) may
cardiovascular events after removal of the top glucose interval in selected studies be related to atherosclerosis. Possible can-
didates include hyperinsulinemia, hyper-
Cardiovascular events triglyceridemia, low HDL, visceral obesity,
and hypertension; all are factors that tend
Reference -coefficient (95% Cl) P value*
to be higher or more common in hypergly-
Fasting cemic patients compared with normo-
Scheidt-Nave et al. (19) 1.203 (0.328 to 2.079) 0.007 glycemic patients. Indeed, the clustering of
Shaten et al. (24) 0.083 ( 0.660 to 0.826) 0.827 glucose intolerance, hypertension, hyper-
Yarnell et al. (29) 1.346 (0.245 to 2.448) 0.023 triglyceridemia, visceral obesity, hyperinsu-
All 0.817 ( 0.020 to 1.655) 0.056 linemia, and insulin resistance in patients
2-h with cardiovascular disease is well known
Balkau et al. (21) 1.157 (0.029 to 2.284) 0.057 and has been labeled as the insulin resis-
Grabauskas et al. (33) 0.970 (0.251 to 1.688) 0.008 tance syndrome or Syndrome X (59).
Tuomilehto et al. (22) 1.128 ( 1.530 to 3.785) 0.432 Fourth, both hyperglycemia and cardiovas-
All 1.029 (0.438 to 1.620) 0.0006 cular disease may have common predis-
*Goodness of fit. posing factors (60,61). Possible candidates
include genetic factors, early nutritional
deficiency, low birth weight, and unidenti-
on combining studies that used different (48–50). Fourth, cross-sectional epidemio- fied environmental factors.
glucose loads and times of glucose sam- logical studies that included both diabetic The current analyses have several limita-
pling. Nevertheless, the fact that the analy- and nondiabetic patients also suggested a tions. First, the fact that the data from indi-
ses of studies, which measured fasting, 2-h progressive relationship between glucose vidual studies were modeled precludes
postprandial, 1-h postprandial, and casual levels and the prevalence of cardiovascular explicit exclusion of a threshold glucose
glucose levels, yielded consistent patterns disease (51). Fifth, this analysis is consis- value below which there is no graded rela-
suggests that these findings are robust. tent with a recent case-control study of tionship between glucose and cardiovascular
A number of factors support the con- subjects with an acute myocardial infarc- disease. Indeed, the finding that an expo-
clusion that an elevated glucose level below tion, in which a progressive relationship nential model best described the data is con-
the diabetic cutoff is associated with car- between glucose levels and the odds of a sistent with the possibility of both a
diovascular disease. First, the glucose cut- myocardial infarction (independent of lipid continuous relationship and a glycemic
offs that define diabetes (fasting and 2-h levels, smoking, abdominal obesity, and threshold for cardiovascular disease. Never-
post-glucose load values of 7.0 and 11.1 insulin levels) persisted even after exclud- theless, data from the individual studies (Fig.
mmol/l, respectively) were chosen to iden- ing patients with diabetes and impaired 1) support the conclusion that if there is a
tify people at risk for eye and kidney dis- glucose tolerance (52). threshold, it extends below the impaired glu-
eases (10,43–45), without regard to the A number of possibilities may explain cose tolerance threshold. This is supported
risk for cardiovascular disease. Thus there the relationship between glucose levels and by the observation of a clear significant rela-
is no a priori reason for this threshold to subsequent cardiovascular events. First, tionship between 2-h glucose and cardiovas-
have any special significance with respect glucose may itself be causally related to cular events (Table 3) in the subanalysis of
to the risk of cardiovascular disease. This is atherosclerosis through a number of mech- data from nondiabetic subjects. Second, it
also true for the glucose cutoff values for anisms including increased oxidative stress excluded several large studies that did not
impaired glucose tolerance (2-h post-glu- (53–55), nonenzymatic glycation of LDL, present data in a way that was amenable to
cose load value of 7.8–11.1 mmol/l), which other apolipoproteins (56) and clotting fac- this analysis (35,37–42). Nevertheless, the
were not originally defined on the basis of tors (fibrin and antithrombin-III) (57), and fact that the glucose–cardiovascular disease
identifying people at higher cardiovascular advanced glycation end-product formation relationship reported in these studies was
risk (8). Second, longitudinal studies in the vessel wall and matrix (57,58). Sec- similar to that observed in the present study
demonstrate that among patients with ond, minimally elevated glucose levels at suggests that exclusion of these studies did
established diabetes, the risk of cardiovas- baseline are likely to be a marker for the not introduce a material bias into our analy-
cular disease and mortality increases as the subsequent development of higher levels of sis. Third, the number of women included in
ambient glucose level increases (4–6). Sev- glucose, impaired glucose tolerance, or dia- the studies was small, thereby limiting the
eral prospective studies that included non- betes, and it is possible that these latter generalizability of the results. However, the
diabetic people, but that did not satisfy all conditions and not glucose elevation per se few studies, which did include women,
the criteria for inclusion in this metare- are related to cardiovascular disease. To the reported results that were consistent with
gression analysis, also showed a similar extent that higher baseline glucose levels the analysis as a whole. Indeed, recent epi-
relationship in the nondiabetic range are associated with even higher subsequent demiological data suggests that the relation-
(34,37,40–42,46,47). Third, recently pub- glucose levels and higher rates of diabetes, ship between glucose and cardiovascular risk
lished analyses of data from four of the the risk of cardiovascular disease with base- in women may be stronger than in men (62).
studies included in this summary also line elevated glucose levels would be mag- Fourth, different biochemical methods were
showed a progressive relationship between nified over a period of 10 years. Third, used to measure glucose levels in the various
glucose levels below the diabetic threshold glucose elevation may be confounded with studies (Table 1), and different algorithms
and subsequent cardiovascular events some other cardiovascular risk factor(s), were used for glucose tolerance testing such

238 DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999

 Coutinho and Associates

as the use of a 50- or 75-g glucose load. Such 3. Goldbourt U, Yaari S, Medalie JH: Factors 17. Fuller JH, Shipley MJ, Rose G, Jarrett RJ,
variability would tend to underestimate the predictive of long-term coronary heart dis- Keen H: Mortality from coronary heart dis-
true relationship between glucose levels and ease mortality among 10,059 male Israeli ease and stroke in relation to degree of gly-
subsequent cardiovascular events. Finally, civil servants and municipal employees: a cemia: the Whitehall Study. Br Med J
the original studies generally used a single 23-year mortality follow-up in the Israeli 287:867–870, 1983
Ischemic Heart Disease Study. Cardiology 18. Donahue RP, Abbott RD, Reed DM, Yano K:
measurement of glucose at baseline. Experi- Postchallenge glucose concentration and
82:100–121, 1993
ence with multiple measures of blood pres- 4. Moss SE, Klein R, Klein BEK, Meuer SM: coronary heart disease in men of Japanese
sure and cholesterol and the risk of The association of glycemia and cause-spe- ancestry: Honolulu Heart Program. Dia -
cardiovascular disease shows that single cific mortality in a diabetic population. Arch betes36:689–692, 1987
measures underestimate the relationship Intern Med154:2473–2479, 1994 19. Scheidt-Nave C, Barrett-Connor E, Wingard
between the risk factor and cardiovascular 5. Andersson DKG, Svardsudd K: Long-term DL, Cohn BA, Edelstein SL: Sex differences
disease compared with multiple measures. It glycemic control relates to mortality in type in fasting glycemia as a risk factor for
is possible therefore that the true relationship II diabetes. Diabetes Care 18:1534–1543, ischemic heart disease death. Am J Epidemiol
between glucose levels and subsequent car- 1995 133:565–576, 1991
diovascular events may be steeper than 6. Kuusisto J, Mykkanen L, Pyorala K, Laakso 20. Vaccaro O, Ruth KJ, Stamler J: Relationship
M: NIDDM and its metabolic control pre- of postload plasma glucose to mortality
demonstrated in the current analyses. with 19-year follow-up. Diabetes Care
dict coronary heart disease in elderly sub-
One important limitation relates to the jects. Diabetes43:960–967, 1994 15:1328–1334, 1992
fact that individual patient data from the 7. Malmberg K, Ryden L, Efendic S, Herlitz J, 21. Balkau B, Eschwege E, Papoz L, Richard JL,
original studies were not used for these Nicol P, Waldenstrom A, Wedel H, Welin L: Claude JR, Warnet JM, Ducimetiere P: Risk
analyses. Therefore, the observed relation- Randomized trial of insulin-glucose infu- factors for early death in non-insulin depen-
ship between glucose and cardiovascular sion followed by subcutaneous insulin dent diabetes and men with known glucose
events is unadjusted for other cardiovascu- treatment in diabetic patients with acute tolerance status. BMJ 307:295–299, 1993
lar risk factors such as age, weight, blood myocardial infarction (DIGAMI Study 22. Tuomilehto J, Schranz A, Aldana D,
pressure, lipids, smoking, and hyperinsu- Group): effects on mortality at 1 year. J Am Pitkaniemi J: The effect of diabetes and
linemia. Nevertheless, the included studies Coll Cardiol26:57–65, 1995 impaired glucose tolerance on mortality in
8. National Diabetes Data Group: Classifica- Malta. Diabet Med11:170–176, 1994
all enrolled patients within a similar age 23. Perry IJ, Wannamethee SG, Whincup PH,
tion and diagnosis of diabetes mellitus and
range (Table 1); many (but not all) showed other categories of glucose intolerance. Dia - Shaper AG: Asymptomatic hyperglycaemia
a significant effect of glucose after adjust- betes28:1039–1057, 1979 and major ischemic heart disease events
ment for these other factors (Table 2) and 9. Leahy JL, Bonner-Weir S, Weir GC: -cell in Britain. J Epidemiol Comm Health
insulin was not a consistent predictor of car- dysfunction induced by chronic hypergly- 48:538–542, 1994
diovascular disease in the individual studies cemia: current ideas on mechanism of 24. Shaten BJ, Kuller LH, Neaton JD, MRFIT
in which it was measured (Table 2) (63). impaired glucose-induced insulin secre- Research Group: Association between base-
Despite these limitations, these analy- tion. Diabetes Care15:442–454, 1992 line risk factors, cigarette smoking, and
ses and other studies provide support for 10. Expert Committee on the Diagnosis and CHD mortality after 10.5 years. Prev Med
the hypothesis that nondiabetic degrees of Classification of Diabetes Mellitus: Report 20:655–669, 1991
of the expert committee on the diagnosis 25. Da Silva A, Widmer LK, Ziegler HW, Nissen
fasting and postprandial hyperglycemia are C, Schweizer W: The Basle Longitudinal
and classification of diabetes mellitus. Dia -
associated with cardiovascular disease and betes Care20:1183–1197, 1997 Study: report on the relation of initial glu-
that dysglycemia is a cardiovascular risk 11. Whitehead A, Whitehead J: A general para- cose level to baseline ECG abnormalities,
factor (64). They highlight the importance metric approach to the meta-analysis of peripheral artery disease, and subsequent
of studying further the role of glucose and randomized clinical trials. Stat Med mortality. J Chronic Dis32:797–803, 1979
glucose lowering on the risk of cardiovas- 10:1665–1677, 1991 26. Hawthorne VM, Gilmour WH: Relation-
cular disease. 12. McCullagh P, Nelder JA: Generalized Linear ship of glucose to prevalence of ECG
Models. 2nd ed. London, Chapman and abnormalities at baseline and to 6-year
Hall, 1989 mortality in Scottish males aged 45–64 yr.
13. Cochran WG: The combination of esti- J Chron Dis32:787–796, 1979
Acknowledgments — M.C. was supported by 27. Reunanen A, Pyorala K, Aromaa A, Maatela
mates from different experiments. Biomet-
the Brazilian Conselho Nacional de Desenvolvi- J, Knekt P: Glucose tolerance and coronary
rics10:101–129, 1954
mento Cientifico e Tecnologico (CNPq), and heart disease in middle aged Finnish men:
14. Stamler R, Stamler J, Lindberg HA, Mar-
S.Y. is a Medical Research Council (Canada) Social Insurance Institution’s Coronary
quardt J, Berkson DM, Paul O, Lepper M,
Career Scientist. Heart Disease Study. J Chron Dis
Dyer A, Stevens E: Asymptomatic hyper-
glycemia and coronary heart disease in 32:747–758, 1979
middle-aged men in two employed popu- 28. Cruz-Vidal M, Garcia-Palmieri MR, Costas
References lations in Chicago. J Chronic Dis R, Sorlie PD, Havlik RJ: Abnormal blood
1. Kannel WB, McGee DL: Diabetes and car- 32:805–815, 1979 glucose and coronary heart disease: the
diovascular disease: the Framingham Study. 15. Stenhouse NS, Murphy BP, Welborn TA: Puerto Rico Heart Health Program. Dia -
JAMA241:2035–2038, 1979 Busselton Population Study: risk associated betes Care6:556–561, 1983
2. Stamler J, Vaccaro O, Neaton JD, Went- with asymptomatic hyperglycemia. J Chron 29. Yarnell JWG, Pickering JE, Elwood PC,
worth D: Diabetes, other risk factors, and Dis 32:693–698, 1979 Baker IA, Bainton D, Dawkins C, Phillips
12-year cardiovascular mortality for men 16. Pyorala K, Savolainen E, Lehtovirta E, Pun- DIW: Does nondiabetic hyperglycemia
screened in the Multiple Risk Factor Inter- sar S, Siltanen P: Glucose tolerance and predict future IHD? evidence from the
vention Trial. Diabetes Care 16:434–444, coronary heart disease: Helsinki Policemen Caerphilly and Speedwell studies. J Clin
1993 Study. J Chron Dis32:729–745, 1979 Epidemio4l 7:383–388, 1994

DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999 239

Relationship of glucose to cardiovascular risk

30. Schroll M, Hagerup L: Relationship of fast- group, systolic blood pressure, and blood Wilson PWF, Evans JC: Association of
ing blood glucose to prevalence of ECG glucose concentration in Trinidad, West HbA1c with prevalent cardiovascular dis-
abnormalities and 10 year risk of mortality Indies. Int J Epidemiol17:62–69, 1988 ease in the original cohort of the Framing-
from cardiovascular diseases in men born 41. Tomas-Abadal L, Varas-Lorenzo C, Bernades- ham heart study. Diabetes 41:202–208,
in 1914: from the Glostrup Population Bernat E, Balaguer-Vintro I: Coronary risk 1992
Studies. J Chron Dis32:699–707, 1979 factors and 20-year incidence of coronary 52. Pais P, Pogue J, Gerstein H, Zachariah E,
31. Ohlson LO, Svardsudd K, Welin L, Eriks- heart disease and mortality in a 20-year Savitha D, Jayprakash S, Nayak PR, Yusuf S:
son H, Wilhelmsen L, Tibblin G, Larsson B: Mediterranean industrial population: the Risk factors for acute myocardial infarction
Fasting blood glucose and risk of coronary Manresa study, Spain. Euro Heart J in Indians: a case-control study. Lancet
heart disease, stroke, and all-cause mortal- 15:1028–1036, 1994 348:358–363, 1996
ity: a 17-year follow-up study of men born 42. Jarrett RJ, McCartney P, Keen H: The Bedford 53. Baynes JW: Role of oxidative stress in devel-
in 1913. Diabetic Med3:33–37, 1986 survey: ten-year mortality rates in newly opment of complications of diabetes. Dia -
32. Lund Haheim L, Holme I, Hjermann I, diagnosed diabetics, borderline diabetics and betes40:405–412, 1991
Leren P: Nonfasting serum glucose and the normoglycemic controls and risk indices for 54. Lyons TJ: Glycation and oxidation: a role in
risk of fatal stroke in diabetic and nondia- coronary heart disease in borderline diabet- the pathogenesis of atherosclerosis. Am J
betic subjects: 18 year follow-up of the ics. Diabetologia22:79–84, 1982 Cardiol71:26B–31B, 1993
Oslo study. Stroke26:774–777, 1995 43. Harris MI: Undiagnosed NIDDM: clinical 55. Giugliano D, Ceriello A, Paolisso G: Oxida-
33. Grabauskas VJ: Glucose intolerance as and public health issues. Diabetes Care tive stress and diabetic complications. Dia -
contributor to noncommunicable disease 16:642–652, 1993 betes Care19:257–267, 1996
morbidity and mortality. Diabetes Care 44. Viberti G: A glycemic threshold for diabetic 56. Lyons TJ: Lipoprotein glycation and its meta-
11:253–257, 1988 complications? N Engl J Med332:1293–1294, bolic consequences. Diabetes41:67–73, 1992
34. Kohl HW, Gordon NF, Villegas JA, Blair 1995 57. Vlassara H, Bucala R, Striker L: Pathogenic
SN: Cardiorespiratory fitness, glycemic sta- 45. Krolewski AS, Laffel LMB, Krolewski M, effects of advanced glycosylation: bio-
tus, and mortality risk in men. Diabetes Quinn M, Warram JH: Glycosylated hemo- chemical, biologic, and clinical implica-
Care 15:184–192, 1992 globin and the risk of microalbuminuria in tions for diabetes and aging. Lab Invest
35. Wilson PWF, Cupples A, Kannel WB: Is patients with insulin-dependent diabetes 70:138–151, 1994
hyperglycemia associated with cardiovas- mellitus. N Engl J Med332:1251–1255, 58. Brownlee M: Glycation and diabetic com-
cular disease? the Framingham study. Am 1995 plications. Diabetes43:836–841, 1994
Heart J 121:586–590, 1991 46. Casiglia E, Pauletto P, Mazza A, Ginocchio G, 59. Laws A, Reaven GM: Insulin resistance and
36. Barrett-Connor E, Wingard DL, Criqui MH, di Menza G, Pavan L, Tramontin P, Capuani risk factors for coronary heart disease. Bail -
Suarez L: Is borderline fasting hypergly- M, Pessina AC: Impaired glucose tolerance lieres’ Clin Endocrinol Metab7:1063–1078,
cemia a risk factor for cardiovascular death? and its covariates among 2079 non-diabetic 1993
J Chron Dis37:773–779, 1984 elderly subjects. Acta Diabetol33:284–290, 60. Stern MP: Diabetes and cardiovascular dis-
37. Butler WJ, Ostrander LDJ, Carman WJ, 1996 ease: the “common soil” hypothesis. Dia -
Lamphiear DE: Mortality from coronary 47. Feskens EJM, Kromhout D: Glucose toler- betes44:369–374, 1995
heart disease in the Tecumseh study: long- ance and the risk of cardiovascular dis- 61. Jarrett RJ: The cardiovascular risk associ-
term effect of diabetes mellitus, glucose tol- eases: the Zutphen study. J Clin Epidemiol ated with impaired glucose tolerance. Dia -
erance and other risk factors. Am J 45:1327–1334, 1992 betic Med13 (Suppl. 2):S15–S19, 1996
Epidemio1l 21:541–547, 1985 48. Balkau B, Shipley M, Jarrett RJ, Pyorala K, 62. Park S, Barrett-Connor E, Wingard DL,
38. Pan W, Cedres LB, Liu K, Dyer A, Schoen- Pyorala M, Forhan A, Eschwege E: High Shan J, Edelstein SL: GHb is a better pre-
berger JA, Shekelle RB, Stamler R, Smith D, blood glucose concentration is a risk factor dictor of cardiovascular disease than fasting
Collette P, Stamler J: Relationship of clinical for mortality in middle-aged nondiabetic or postchallenge plasma glucose in women
diabetes and asymptomatic hyperglycemia men. Diabetes Care360:360–367, 1998 without diabetes: the Rancho Bernardo
to risk of coronary heart disease mortality 49. Orencia AJ, Daviglus ML, Dyer AR, Walsh Study. Diabetes Care19:450–456, 1996
in men and women. Am J Epidemiol M, Greenland P, Stamler J: One-hour post- 63. Wingard DL, Barrett-Connor E, Ferrara A:
123:504–516, 1986 load glucose and risks of fatal coronary Is insulin really a heart disease risk factor?
39. Janghorbani M, Jones RB, Harper Gilmour heart disease and stroke among nondiabetic Diabetes Care18:1299–1304, 1995
W, Hedley AJ, Zhianpour M: A prospective men and women: the Chicago Heart Asso- 64. Gerstein HC, Yusuf S: Dysglycaemia and
population based study of gender differen- ciation Detection Project in Industry (CHA) risk of cardiovascular disease. Lancet
tial in mortality from cardiovascular disease Study. J Clin Epidemiol50:1369–1376, 1997 347:949–950, 1996
and “all causes” in asymptomatic hyper- 50. Lowe LP, Liu K, Greenland P, Metzger BE, 65. Pyorala K, Savolainen E, Kaukola S, Haa-
glycemics. J Clin Epidemiol47:397–405, Dyer AR, Stamler J: Diabetes, asympto- pakoski J: Plasma insulin as coronary heart
1994 matic hyperglycemia and 22-year mortality disease risk factor: relationship to other
40. Miller GJ, Kirkwood BR, Beckles GLA, in black and white men: the Chicago Heart risk factors and predictive value during 9
Alexis SD, Carson DC, Byam NTA: Adult Association Detection Project in Industry 1/2-year follow-up of the Helsinki Police-
male all-cause, cardiovascular and cere- Study. Diabetes Care20:163–169, 1997 men Study population. Acta Med Scand
brovascular mortality in relation to ethnic 51. Singer DE, Nathan DM, Anderson KM, Suppl701:38–52, 1985

240 DIABETES CARE, VOLUME 22, NUMBER 2, FEBRUARY 1999