Blood Glucose and Risk of Cardiovascular Disease in the Asia Pacific Region

APCSC is an individual participant data overview (meta-analysis) of prospective cohort studies. The methods have been reported in detail elsewhere (18). In brief, prospective cohort studies were eligible for inclusion if they contained study populations from the Asia Pacific region and had at least 5,000 person-years of follow-up recorded or planned. At a minimum, they had to have recorded data on date of birth or age, sex, blood pressure at baseline, and date or age at death. Additional data sought included date of baseline survey, glucose, fasting status, blood cholesterol, height, weight, smoking, and any data on repeat measures of risk factors. Outcome data included nonfatal stroke and ischemic heart disease (IHD) and cause of death (18).

All analyses are restricted to participants aged ≥20 years who had blood glucose and fasting status recorded at baseline. Analyses reported here excluded individuals with a baseline glucose value of ≥20 mmol/l. Data on fasting and nonfasting glucose were analyzed separately. Stratified Cox proportional-hazards analyses (19) were used to regress time until first event against baseline glucose with individual participant data collected on all cohorts. All analyses were stratified by sex and cohort to control for confounding and reduce statistical heterogeneity. Age was treated as an external time-dependent covariate (20) to assess change in hazards as an individual’s age increases. These “age at risk” analyses account for the fact that cohorts had different start and follow-up times.

Age-specific analyses included age at risk categories of <60, 60–69, and ≥70 years (analyses were also conducted by sex) and region (Asia vs. Australasia and comparison of regions within Asia). Where possible, analyses were undertaken for total (fatal and nonfatal) and fatal (death occurring within 28 days of event) CVD outcomes. Further sensitivity analyses included restricting analyses to those participants who had no diagnosis of diabetes at baseline and/or adjusting for other potential confounders, such as systolic blood pressure (mmHg), serum total cholesterol (mmol/l), smoking (current vs. not current), and BMI (kg/m²), for those cohorts who had recorded these variables at baseline. Insufficient data were available for ethnic-specific analysis. Effect modification was assessed with the use of statistical interaction terms for sex and region in the Cox model.

A single baseline measure is subject to random fluctuations of fasting glucose, due partly to the measurement process and partly to any real but temporary deviations at the baseline visit from the usual blood glucose level (21). The distribution of single baseline glucose measures is therefore wider than the distribution of true usual glucose values due to regression to the mean. The term for the resulting underestimation of an exposure with an end point that occurs with one-off baseline measures is “regression dilution bias” (21). Usual glucose values are estimates that have been corrected for regression dilution bias and are therefore closer to true glucose values. These usual estimates were made using repeated glucose measurement data from three cohort studies. A mixed-model approach was used to calculate the regression dilution ratio (λ) that took into account the varying time intervals between the glucose measurements. Several other methods of correction (22), including the method of MacMahon et al. (21), have been used in previous analyses of APCSC data, and each yielded very similar regression dilution ratios (11). To adjust for regression dilution bias, the regression coefficient (β*) was calculated by multiplying the uncorrected regression coefficient (β) by λ⁻¹ (where λ⁻¹ = 1.6). A similar approach was used to adjust the standard errors.

The hazard ratios and corresponding 95% CIs were estimated for a 1-mmol/l reduction in usual glucose. In the nonparametric analyses, participants were divided into quartiles according to baseline glucose (<5, 5–5.9, 6–6.9, and ≥7 mmol/l), and the hazard ratios were plotted against “usual” glucose rather than baseline glucose. The 95% CIs for each exposure group were estimated by treating the hazard ratios as “floating variances.” This approach does not affect the hazard ratios but enables the comparison between pairs of exposure groups rather than the comparison of a single exposure group with an arbitrary reference group (23,24).

Baseline glucose data were available in 17 of the 43 cohort studies (Table 1). Of these cohorts, fasting glucose measurements were recorded in 13 studies, including 237,468 participants with 1,194,320 person-years of follow-up, and nonfasting glucose measurements were recorded in 6 studies, including 27,996 participants with 185,324 person-years of follow-up (the Melbourne and Akabane cohorts included a mixture of fasting and nonfasting glucose data). During a mean follow-up of 5 years, there were 1,661 stroke and 816 IHD events in the fasting glucose studies, and during a mean follow-up of 6.5 years, there were 144 stroke and 240 IHD events in the nonfasting glucose studies.

Levels of other risk factors (with the exception of smoking prevalence) tended to be lower in the lowest two quartiles (<5 and 5.0–5.9 mmol/l) compared with the top two glucose quartiles (6.0–6.9 and ≥7mmol/l). For example, in those studies with fasting glucose data, the mean systolic blood pressure levels were 120–127 vs. 135 mmHg, mean total cholesterol was 4.9–5.2 vs. 5.4–5.5 mmol/l, and mean BMI was 22.8–24.1 vs. 25.0–25.5 kg/m².

There was a positive log-linear association between usual fasting glucose and the risk of total stroke and IHD (Fig. 1). For both end points, this association was continuous down to at least 4.9 mmol/l, with no evidence of a threshold level. Overall, a 1 mmol/l lower usual fasting glucose level was associated with a 21% (95% CI 18–24%) lower risk of total stroke and a 23% (19–27%) lower risk of total IHD. Associations were of similar magnitude for both stroke and IHD mortality (fatal events) and incidence (total events).

There was a strong continuous relationship between usual fasting glucose and the risk of CVD death (Fig. 1); a 1 mmol/l lower usual fasting glucose level was associated with a 19% (15–22%) lower risk of CVD death.

Figure 2 illustrates that while the 95% CIs became wider, the strength of the associations for usual fasting glucose and either CVD end point were not substantially changed when adjusting for confounders, excluding individuals with a diagnosis of diabetes or excluding those with a diagnosis of diabetes and adjusting for confounders. Overall, the associations for usual nonfasting glucose were weaker and less precise than those for usual fasting glucose. This was particularly evident for stroke.

There was a suggestion of a slightly stronger association between usual fasting glucose for both total stroke and IHD in the youngest age-group, but the 95% CIs overlapped with those of older age-groups. There was no significant difference between the associations for males and females (e.g., a 1 mmol/l lower usual fasting glucose was associated with 21% [14–29%] and 23% [11–34%] lower risk of total IHD in males and females, respectively). The hazard ratio for stroke in Australasia appeared stronger than in Asia; however, the CIs of the hazard ratios overlapped, and Australasia estimates were only based on two Australian cohorts. Sensitivity analyses that excluded those with diabetes (at baseline) and adjusted for potential confounders diminished the regional differences. No significant differences were found between regions within Asia. Limited data were available by stroke subtype, but analyses indicated that a 1 mmol/l lower usual fasting glucose level was associated with a 20% (0–36%) lower risk of total hemorrhagic stroke (excluding subarachnoid hemorrhage) and a 17% (−1 to 32%) lower risk of total ischemic stroke.

Nine cohort studies included data on fasting glucose and diagnosis of diabetes at baseline. Figure 3 demonstrates that the continuous log-linear association was maintained, regardless of diabetes status at baseline. Overall, a 1 mmol/l lower usual fasting glucose was associated with a 20% (13–25%) reduction in total stroke risk and a 23% (18–29%) reduction in total IHD risk.

This analysis of cohorts from the Asia Pacific region clearly indicates a positive continuous association between usual blood glucose and CVD risk. This association extends down to about 4.9 mmol/l, well below the usual fasting glucose cutoff levels for diagnosis of diabetes and impaired glucose tolerance. The positive association was maintained when individuals with diabetes at baseline were excluded and with adjustment for potential confounders. The associations were very similar for males and females and across age subgroups. There was no strong evidence of regional differences, but small numbers limited subgroup analysis. The overall associations for usual nonfasting glucose were weaker than those for usual fasting glucose; however, nonfasting measures are less robust because they are affected by the postprandial state.

This study has several advantages over previous studies, such as the substantial size of the database, the availability of individual participant data across a range of glucose levels, and the ability to correct for regression dilution bias, exclude those with diabetes at baseline, and control for potential confounders. Unlike most other analyses, it also enabled the shape and strength of the dose-response relationship to be assessed in non-Western cohorts.

No data were available for oral glucose tolerance tests (OGTTs). Previous comparisons have concluded that 2-h post-load glucose from an OGTT is a superior test to fasting glucose for diabetes (25,26) and predicting cardiovascular and/or all-cause mortality (27,28). However, the OGTT is more difficult and expensive to perform than simpler fasting blood glucose measures (8,29). Fasting blood glucose measures are still useful (29) and are likely to be more reliable than the OGTT across several studies (8).

Many studies and overviews dating back to the 1970s and 1980s investigated whether a threshold relationship existed between glycemia and IHD risk (30) and investigated the effect of adjusting for classic risk factors (31). The overall results of the early studies were inconclusive and conflicting (30,31), and it was unclear whether glucose was a risk factor in the absence of diabetes (32,33). Subsequent analyses have found a positive association between the highest and lowest glucose centiles and CVD. The shape of the association has often been reported as nonlinear (e.g., J-shaped) (34–40). However, the categorization of glucose and small study size substantially limited the ability to explore the shape of the dose-response relationship across the whole range of glucose levels.

A recent overview of 20 studies concluded that there was a significant exponential association between glucose and CVD in nondiabetic individuals that extended below the usual “diabetic threshold” (8). Another prospective study demonstrated that HbA_1c was continuously related to subsequent CVD and all-cause mortality through the whole population distribution, with lowest rates in individuals with HbA_1c concentrations <5% (∼4.5–5.0 mmol/l fasting glucose) (41).

Randomized controlled trials indicate to what extent the “epidemiologically expected” reductions in disease are realized. Trials have examined the effect of blood glucose lowering in type 2 diabetes (42–48), but several have had small sample size, less glycemic control than planned, and relatively short follow-up periods (49,50). Overall, the trials did demonstrate decreased microvascular complications and a nonsignificant trend toward decreasing macrovascular complications. The modest size of the effect on macrovascular outcomes is consistent with the predicted effects from this analysis and the relatively small change in achieved glucose differences. Further analysis showed a continuous association between glycemia and the risk of macrovascular complications below diabetic thresholds (45,49,50).

There are several implications for clinical and public health practice from this study. The continuous association suggests that a wider group of individuals may benefit from glucose lowering, not just those with diabetes or impaired glucose tolerance, but further trial evidence would confirm this. Risk prediction tools could be improved by including fasting glucose as a continuous risk factor, rather than diagnosis of diabetes. Finally, and perhaps most importantly, there are potential benefits from population-wide lowering of determinants of CVD risk. The most appropriate approach to cardiovascular prevention is often a coordinated effort to lower the risk profiles of entire populations coupled with a targeted approach to those at highest absolute risk of CVD. Previous analyses have demonstrated continuous associations between risk factors such as blood pressure, cholesterol, and BMI and CVD risk (11–15). Results from the current study support the view that this is also the case for blood glucose, and the blood glucose levels may be an important risk factor in the Asia Pacific region in individuals with and without diagnosed diabetes.

C.M.M. Lawes, V. Parag, D.A. Bennett, I. Suh, T.H. Lam, G. Whitlock, F. Barzi, W.H. Pan, and A. Rodgers.

V. Parag, D.A. Bennett, R.B. Lin, F. Barzi, and M. Woodward.

D.F. Gu, T.H. Lam, C.M.M. Lawes, S. Macmahon, W.H. Pan, A. Rodgers, I. Suh, H. Ueshima, M. Woodward.

Aito Town: A. Okayama, H. Ueshima, H. Maegawa; Akabane: N. Aoki, M. Nakamura, N. Kubo, T. Yamada; Anzhen02: Z.S. Wu; Anzhen: C.H. Yao, Z.S. Wu; Australian Longitudinal Study of Ageing: G. Andrews; Australian National Heart Foundation: T.A. Welborn; Beijing Ageing: Z. Tang; Beijing Steelworkers: L.S. Liu, J.X. Xie; Blood Donors’ Health: R. Norton, S. Ameratunga, S. MacMahon, G. Whitlock; Busselton: M.W. Knuiman; Canberra-Queanbeyan: H. Christensen; Capital Iron and Steel Company: X.G. Wu; CISCH: J. Zhou, X.H. Yu; Civil Service Workers: A. Tamakoshi; CVDFACTS: W.H. Pan; East Beijing: Z.L. Wu, L.Q. Chen, G.L. Shan; Electricity Generating Authority of Thailand (EGAT): P. Sritara; Fangshan: D.F. Gu, X.F. Duan; Fletcher Challenge: S. MacMahon, R. Norton, G. Whitlock, R. Jackson; Guangzhou: Y.H. Li; Guangzhou Occupational: T.H. Lam, C.Q. Jiang; Hisayama: M. Fujishima, Y. Kiyohara, H. Iwamoto; Hong Kong: J. Woo, S.C. Ho; Huashan: Z. Hong, M.S. Huang, B. Zhou; Kinmen: J.L. Fuh; Konan: H. Ueshima, Y. Kita, S.R. Choudhury; KMIC: I. Suh, S.H. Jee, I.S. Kim; Melbourne: G.G. Giles; Miyama: T. Hashimoto, K. Sakata; Newcastle: A. Dobson; Ohasama: Y. Imai, T. Ohkubo, A. Hozawa; Perth: K. Jamrozik, M. Hobbs, R. Broadhurst; Saitama: K. Nakachi; Seven Cities: X.H. Fang, S.C. Li, Q.D. Yang; Shanghai Factory Workers: Z.M. Chen; Shibata: H. Tanaka; Shigaraki Town: Y. Kita, A. Nozaki, H. Ueshima; Shirakawa: H. Horibe, Y. Matsutani, M. Kagaya; Singapore Heart: K. Hughes, J. Lee; Singapore NHS92: D. Heng, S.K. Chew; Six Cohorts: B.F. Zhou, H.Y. Zhang; Tanno/Soubetsu: K. Shimamoto, S. Saitoh; Tianjin: Z.Z. Li, H.Y. Zhang; Western Australia AAA Screenees: P. Norman, K. Jamrozik; Xi’an: Y. He, T.H. Lam; Yunnan: S.X. Yao. (Note: The italicized studies provided data used in this article.)

This project was supported by a grant from the Health Research Council of New Zealand, the National Institute on Aging Grant PO1-AG17625, the National Health and Medical Research Council of Australia, and an unrestricted educational grant from Pfizer. The sponsors had no influence on design, analysis, or interpretation of results.