The association between hyperglycaemia and the risk of microvascular complications [1, 2], myocardial infarction [1, 2, 3], stroke [4], macrovascular mortality [5, 6, 7] and all-cause mortality [1, 7, 8, 9, 10, 11, 12] is well known in patients with Type 2 diabetes. A meta-regression analysis of 20 studies among non-diabetic men reported a graded relation between glucose concentration and cardiovascular disease (CVD) events, with blood glucose concentrations below the threshold for diabetes nevertheless associated with CVD events [13]. Several cohort studies [14, 15, 16, 17, 18, 19], including some on Asian populations, have indicated that people with pre-diabetic conditions have an increased risk of all-cause mortality and CVD morbidity and mortality.

The Diabetes Epidemiology: Collaborative Analysis of Diagnostic Criteria in Europe (DECODE) study has confirmed that people with IGT or IFG have a significantly increased risk of death from CVD and all causes, and that this risk ranks between those of normal and diabetic groups [20, 21]. The DECODE data also showed that the 2-h plasma glucose (2-h PG) concentration after a 75-g OGTT is an independent predictor of CVD mortality, while prediction of CVD death on the basis of fasting plasma glucose (FPG) is largely due to correlation with 2-h PG concentrations. Similar results for Asian populations are scarce, but much needed due to the rapidly increasing prevalence of diabetes in Asia [22, 23].

To assess the relationship between glucose concentrations and mortality, together with other established risk factors in Asian populations, we analysed the follow-up data from the Diabetes Epidemiology: Collaborative Analysis of Diagnostic Criteria in Asia (DECODA) Study [24]. The specific aims of this analysis were: (i) to assess to what degree FPG and 2-h PG predict all-cause and CVD mortality, taking account of other established CVD risk factors; and (ii) to identify glucose categories associated with excess all-cause and CVD mortality.

Subjects and methods

Participants

The study populations and methods used to recruit participants in the DECODA study have already been described [24, 25, 26]. Briefly, centres which had carried out studies on the prevalence of diabetes using a standard 75-g OGTT with FPG and 2-h PG in adult Asian populations were identified and invited to participate in the DECODA study. Data on FPG and 2-h PG concentrations and some other variables were transferred to the Diabetes and Genetic Epidemiology Unit of the National Public Health Institute in Helsinki, Finland, for pooled data analysis. This analysis included studies with prospective data on cause-specific mortality and all required confounding variables (height, weight, systolic and diastolic blood pressure, total serum cholesterol, smoking status).

A total of five centres contributed to the pooled analysis (Table 1). Of these three included Japanese populations: Funagata, Japan (Funagata Diabetes Study) [19], Seattle, USA (Japanese American Community Diabetes Study) [27, 28], Sao Paulo, Brazil (Japanese Brazilian Diabetes Study) [29]. Two included Asian Indian populations: Fiji and Mauritius [30, 31]. Of the 6817 subjects (3121 men and 3696 women) aged 30 to 89 years, 511 (7.5%) had previously been diagnosed with diabetes. The number of deaths from all causes was 381 among subjects without previously diagnosed diabetes and 107 among those previously diagnosed with diabetes. The corresponding numbers for CVD deaths were 182 and 58. The median duration of follow-up was 5.0 years. People without previously diagnosed diabetes contributed 37 026 person-years of follow-up, those previously diagnosed with diabetes contributed 3208 person-years. The follow-up ascertainment rate ranged from 92 to 100%.

Table 1 Surveys analysed for this DECODA study

Vital status and underlying cause of death were determined for each deceased participant from death certificates. This was done independently for each centre. Individuals who had emigrated and whose vital status could not be verified were treated as censored cases at the time of emigration or loss. The underlying cause of death was classified according to the International Classification of Diseases, 9th Revision, with codes 401 to 448 considered to be CVD [32].

Subjects clinically diagnosed as diabetes before the survey were classified as previously diagnosed cases. Subjects without previously diagnosed diabetes were placed in three glucose categories, firstly on the basis of 2-h PG criteria alone (11.1 mmol/l or more; 7.8–11.0 mmol/l; less than 7.8 mmol/l), and secondly on the basis of FPG concentrations alone (7.0 mmol/l or more; 6.1–6.9 mmol/l; and less than 6.1 mmol/l). Further classification of glucose category was done according to the 1999 World Health Organization (WHO) criteria [33], i.e. (i) normal: FPG less than 6.1 mmol/l and 2-h PG less than 7.8 mmol/l; (ii) impaired glucose regulation (IGR): FPG 6.1-6.9 mmol/l and/or 2-h PG 7.8-11.0 mmol/l; (iii) FPG diabetes: FPG 7.0 mmol/l or more; (iv) isolated post-challenge glucose diabetes: FPG less than 7.0 mmol/l and 2-h PG 11.1 mmol/l or more; (v) previously diagnosed diabetes.

All participants gave informed consent to participate in the study according to the Declaration of Helsinki or other ethical standards relevant to the time when data collection was conducted.

Statistical methods

The age-standardised prevalence of glucose intolerance (IGR, IFG and/or IGT, and diabetes) and the age-standardised CVD and all-cause mortality were calculated using 5-year age intervals with direct standardisation to the world standard population [34]. Analysis of variance was used to compare means of fasting glucose and/or 2-h PG concentrations at baseline among subjects without previously diagnosed diabetes between five cohorts. Adjusted means of age, BMI, blood pressure and total cholesterol were calculated using analysis of covariance, and compared with a given reference group by pair-wise comparison.

Hazard ratios for CVD and all-cause mortality were estimated in subjects without previously diagnosed diabetes using the multivariate Cox proportional hazards models for each of the five cohorts, adjusting for age, sex, BMI, systolic blood pressure, total cholesterol, plasma glucose (fasting or 2-h) and smoking status (coded as non-smokers, ex-smokers and current smokers). BMI was calculated as weight (kg) divided by height2 (m). In participants without previously diagnosed diabetes FPG and 2-h PG concentrations were natural-logarithmically transformed and used in the models as a continuous variable. The individual hazards ratios of each Cox model were pooled into an overall hazards ratio according to the Fleiss method [35]. A fixed rather than a random effects approach was chosen because the Q statistic for measuring study-to-study variation in effect size was not statistically different from zero [35]. Individual and the overall standardised hazards ratios, corresponding to a standard deviation increase of 1 in the natural-logarithmically transformed glucose concentrations, were reported with their 95% CI. To make results of FPG and 2-h PG compatible, hazard ratios for a standard deviation increase of 1 in natural-logarithmically transformed FPG and 2-h PG values were calculated and reported.

The relative risks for CVD and all-cause mortality, the exponential function of the beta coefficient and their 95% CIs were estimated for the three glucose categories according to FPG and 2-h PG as described above, using the multivariate Cox proportional hazards model adjusting for age, sex, cohort, BMI, systolic blood pressure, cholesterol and smoking status. A Wald test was done to examine the trend in the risk of events, by coding the three glucose categories linearly. A chi square log-likelihood ratio test was carried out to test the prediction of death by FPG, adjusted for 2-h PG, and the predictive value for 2-h PG, adjusted for FPG. A similar Cox model was used to estimate the relative risk of CVD and all-cause mortality for the five WHO glucose categories described above.

SPSS for Windows version 11.0 (Chicago, Ill., USA) was used for statistical analyses. The p values were based on two-sided tests and results were regarded as statistically significant at a p value of less than 0.05.

Results

Among subjects without previously diagnosed diabetes, the baseline mean level of 2-h PG was lower in the Funagata Study than in other studies (p<0.05), while the baseline mean level of FPG was higher in the Mauritius and the Fiji Study than in the Funagata Study (p<0.05,Table 2).

Table 2 Baseline characteristics of surveys analysed for this DECODA study

The age-standardised prevalence of all diabetes categories among subjects who where 40 to 74 years of age was lowest in the Funagata Study (6.6%), followed by the Sao Paulo (Japanese Brazilian) Study (18.7%), the Mauritius Study (22.3%), the Seattle (Japanese American) Study (22.9%) and the Fiji Study (30.8%) (Table 2). The lowest prevalence of IGR was found in the Funagata Study (10.4%), followed by the Sao Paulo (Japanese Brazilian) Study (18.0%), the Fiji Study (18.8%), the Mauritius Study (22.1%) and the Seattle (Japanese American) Study (32.9%).

Baseline mean values of BMI for each glucose category did not differ between the studies (Table 2). However, with the exception of total cholesterol in the Fiji Study, baseline mean values of CVD risk factors were lower in the Funagata Study than in other studies for each glucose category.

Age-standardised all-cause and CVD mortality among subjects of 40 to 74 years of age was lower in the three Japanese studies than in the two Asian Indian studies (Table 1). Age-standardised all-cause and CVD mortality rates ranged from 6.3 and 2.6 respectively per 1000 person-years in the Funagata Study to 28.0 and 13.2 respectively per 1000 person-years in the Fiji Study. In this age group the overall standardised mortality rate was 13.3 per 1000 person-years for all-causes and 6.8 per 1000 person-years for CVD.

There were no major cohort and ethnic variations in beta coefficients for prediction of death from FPG and 2-h PG, as the statistic Q values varied from 2.9 to 7.9, 4df, (all p values >0.05). There was no statistically significant evidence against pooling the data. Mortality from all causes and CVD increased, and significantly, with increasing 2-h PG and/or FPG, adjusting for age, sex, BMI, systolic blood pressure, total cholesterol and smoking status among subjects without previously diagnosed diabetes (Figs. 1, 2). The standardised hazards ratio corresponding to a standard deviation increase of 1 in log-transformed 2-h PG was 1.41 (95%CI: 1.28–1.54) for all-cause mortality and 1.49 (1.31–1.70) for CVD mortality. Corresponding standardised hazard ratios for a standard deviation increase of 1 in log-transformed FPG were 1.22 (1.12–1.33) and 1.32 (1.18–1.49). The risk prediction based on these models showed that the relative risk was 1.14 (1.05–1.25) for all-cause mortality and 1.24 (1.10–1.39) for CVD mortality when FPG increased from 7.0 to 8.0 mmol/l, corresponding to a 0.76 standard deviation increase in FPG. A similar 0.76 standard deviation increase in 2-h PG would correspond to an increase from 9 to 11.9 mmol/l, leading to a relative risk of 1.29 (1.18–1.41) and 1.35 (1.19–1.54) for all-cause mortality and CVD mortality respectively.

Fig. 1a, b
figure 1

Individual (◆ or ■) and overall (∣) hazard ratios and 95% CIs (lines) for all-cause mortality among subjects without previously diagnosed diabetes. Ratios are shown according to one standard deviation increase in natural log-transformed fasting (◆) (a) and 2-h (■) (b) plasma glucose concentrations in an OGTT, adjusting for age, sex, BMI, systolic blood pressure, total cholesterol and smoking status

Fig. 2a, b
figure 2

Individual (◆ or ■) and overall (∣) hazard ratios and 95% CIs (lines) for cardiovascular mortality among subjects without previously diagnosed diabetes. Ratios are shown according to one standard deviation increase in natural logtransformed fasting (◆) (a) and 2-h (■) (b) plasma glucose concentrations in an OGTT, adjusting for age, sex, BMI, systolic blood pressure, total cholesterol and smoking status

An increase in all-cause and CVD mortality coincided with an increasing deterioration of glucose metabolism from normal glucose tolerance to overt diabetes (Table 3). For the FPG criteria, the relative risk of death from CVD and all causes was statistically significant only in the diabetic glucose range (FPG ≥7.0 mmol/l) compared with normoglycaemic individuals (FPG <6.1 mmol/l) (p<0.001). For the 2-h PG criteria the relative risk of death from all causes was statistically significant for the IGT range (7.8–11.0 mmol/l) (p<0.05), while risk of CVD mortality only reached significance in the diabetic glucose range (2-h PG ≥11.1 mmol/l) (p<0.001). Compared with the nested models, the inclusion of FPG did not significantly improve the prediction of death obtained from the model based on 2-h PG criteria alone (χ2=0.43 for all-cause mortality and χ2=0.37 for CVD mortality, 2df, both p values >0.50). In contrast, the inclusion of the 2-h PG significantly improved the prediction of death obtained from the model based on FPG criteria alone (χ2=38.12 for all-cause mortality and χ2=24.08 for CVD mortality, 2df, both p values <0.0001).

Table 3 Multivariate hazard ratio (HR) with 95% CI of all-cause and cardiovascular mortality according to fasting and 2-h plasma glucose criteria

Cox proportional hazards analysis of the pooled data showed that participants with previously diagnosed diabetes had a higher risk of CVD and all-cause mortality than people in other glucose categories (Table 4). However the increased risk of death in previously diagnosed cases of diabetes was not statistically different from that in newly diagnosed cases. Diabetic subjects identified by 2-h PG alone (FPG <7.0 mmol/l and 2-h PG ≥11.1 mmol/l) had an equal risk of death from CVD and all causes to those screened by FPG (FPG ≥7.0 mmol/l). Subjects with IGR had a greater risk of death from CVD and all causes than those with normal glucose. However, the higher risk of death in cases of IGR was not statistically different from that in people with normal glucose concentrations.

Table 4 Glucose categories and multivariate relative risks of all-cause and cardiovascular disease mortality The DECODA Study

Discussion

Several previous cohort studies which have included Asian subjects have shown that hyperglycaemia increases the risk for CVD morbidity and mortality [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 30]. The excess risk starts at glucose concentrations well below the threshold for the diagnosis of diabetes by current diagnostic criteria [13, 14, 15, 16, 17, 18, 19, 20, 21]. Our study, which was based on five cohorts with populations of Asian origin and had a 5-year median follow-up, showed that post-challenge glucose was an independent predictor of CVD and all-cause mortality among people without previously diagnosed diabetes. The association was graded with higher 2-h PG predicting higher mortality. When the data were analysed according to glucose categories, it was found that with regard to the prediction of premature death, post-challenge glucose (2-h PG) was superior to fasting glucose in these Asian populations, as previously shown for European populations [20, 21].

In our study the effect of FPG and/or 2-h PG associated with increased risks of all-cause and CVD mortality was homogeneous across cohorts, although baseline mean values of CVD risk factors, the prevalence of diabetes and rates of CVD and all-cause mortality were different, being highest in Asian Indians, intermediate in Japanese immigrants and lowest in ethnic Japanese people living in Japan. Moreover, the magnitude of the increased risk in the overall cohort was similar for FPG and 2-h PG. It showed a graded relation with CVD and all-cause mortality, independent of known CVD risk factors: The relative risk increased 14% for all-cause mortality and 24% for CVD mortality when FPG increased from 7.0 to 8.0 mmol/l. A corresponding increase of 2-h PG from 9.0 to 11.9 mmol/l increased all-cause mortality by 29% and CVD mortality by 35%. Similar graded relations between glucose concentrations and CVD and all-cause mortality have been found in white populations [13, 36].

Over an observation period which was half as long as that in Europe [20, 21], subjects with IGT based on 2-h PG concentrations (7.8–11.0 mmol/l) had a significantly greater risk of all-cause mortality than people with normal 2-h PG concentrations (<7.8 mmol/l). However, subjects with IFG based on FPG concentrations (6.1–6.9 mmol/l) did not have a a greater risk of death than subjects with normal FPG (<6.1 mmol/l) in our study. It is possible that the excess risk of CVD mortality in subjects with IGT based on 2-h PG criteria (7.8–11.0 mmol/l) would have reached significance, if those with previous CVD events had been excluded from the analysis. In an 8-year follow-up of elderly non-diabetic Dutch people [16], the predictive ability of post-challenge glucose on CVD mortality dramatically improved and became statistically significant when subjects with previous CVD events were excluded. This could not be done in our study, because previous CVD events were defined by different criteria among the studies included in our analysis, so the validity of this information could not be assured.

Several recent reports from prospective epidemiological studies have suggested that CVD is more strongly associated with 2-h PG than with FPG [17, 19, 20, 21, 36, 37]. This stronger association with 2-h PG has also been reported for the degree of carotid atherosclerosis [38] and coronary atherosclerosis [39]. Although the mechanisms underlying these associations are not yet understood, it is noteworthy that the Insulin Resistance Atherosclerosis Study recently reported that higher acute-phase proteins, indicative of chronic sub-clinical inflammation, have a stronger relation to 2-h PG than to FPG [40] and insulin resistance [41]. Alternatively, increased 2-h PG concentrations could be a reflection of impaired early-phase glucose-stimulated insulin release, which could in turn increase oxidative stress [42], thus accounting for these findings.

Among people diagnosed with diabetes, either on the basis of FPG or 2-h PG concentrations, we showed that risk of death from CVD and all causes was higher than in people with the corresponding normal glucose categories. However, the excess risk observed in the diabetic FPG range disappeared after adjustment for 2-h PG. In contrast the prediction of death obtained with 2-h PG was not influenced by FPG. These findings underscore reports from Europe [20, 21] that diagnostic criteria based on post-challenge glucose are more effective in predicting premature death than criteria based on fasting glucose.

Isolated post-challenge hyperglycaemia is common in elderly women [37] and among non-obese European [20, 21, 43] and southern hemisphere populations [30]. In the studies cited, subjects with isolated post-challenge hyperglycaemia also had an excess risk of CVD mortality. We found that 43% of all newly diagnosed cases of diabetes were detected by post-challenge glucose alone and that these subjects had an equal risk of death from CVD and all causes as people diagnosed by fasting glucose. Thus using only fasting glucose to screen for Type 2 diabetes in Asians will fail to detect nearly half of the persons with previously undiagnosed diabetes and a very high risk of premature death.

We found that subjects with IFG based on FPG criteria (6.1–6.9 mmol/l) had higher age-standardised all-cause and CVD mortality than those with IGT based on 2-h PG criteria (7.8–11.0 mmol/l). However, the excess risk of all-cause and CVD mortality in people with IFG based on FPG criteria, compared with that in people with normal FPG based on FPG criteria (<6.1 mmol/l) was largely dependent on the 2-h PG concentration. In contrast, the excess risk of all-cause and CVD mortality in participants with IGT based on 2-h PG criteria, compared with people with normal 2-h PG concentrations based on 2-h PG criteria (<7.8 mmol/l) was not influenced by FPG concentrations. Thus among non-diabetic Asians IGT based on 2-h PG criteria seems to be more effective than IFG based on FPG criteria in predicting premature death. This observation is consistent with findings for European populations [20, 21]. The clinical implications, however, are less clear. Three different trials [44, 45, 46] have shown that non-diabetic persons with a 2-h PG concentration in the IGT range can prevent progression to diabetes, but none of these trials showed a reduction of CVD events. The recent STOP-NIDDM trial showed a 49% risk reduction in CVD events in IGT patients treated with acarbose [47]. The British Fasting Hyperglycaemia Study did not show any benefit on the development of diabetes with lifestyle change among subjects with IFG [48]. In relation to screening for diabetes, the epidemiological data in combination with the clinical trials suggest that 2-h PG has a stronger prognostic implication than FPG, although both are highly correlated. It is therefore important to also identify people with isolated high 2-h PG values.

In the DECODE study [20], the excess risk of death from all causes was similar between people with isolated IGT and people with isolated IFG. The majority of subjects with IGR also had isolated IGT [20, 21]. These results emphasise the importance of identifying non-diabetic subjects with IGT but normal fasting glucose. In our study, isolated IGT occurred in 71% of subjects with IGR. Because of the small number of deaths, we could not report separately the risk of death from all causes and CVD in the three IGR glucose categories.

In summary, post-challenge glucose was a graded and independent predictor of all-cause and CVD mortality among subjects without previously diagnosed diabetes in Asian (Japanese and Asian Indian) ethnic groups. While the OGTT is obviously needed to diagnose unrecognised diabetes and IGT, it is not ideal for screening purposes because it is time-consuming. Several risk scores [49, 50, 51, 52, 53] have been developed to identify individuals at high risk of having undiagnosed diabetes, but while the sensitivity and specificity of these risk scores may be stable between countries if applied to the same ethnic group, they are unlikely to perform equally well for different ethnic groups [54]. Against this background it is clear that the prevention of complications and premature death must start much earlier than the current cut-off points suggested for diabetes.