Screening for pre-diabetes to predict future diabetes using various cut-off points for HbA1c and impaired fasting glucose: the Toranomon Hospital Health Management Center Study 4 (TOPICS 4)

Authors


: Hirohito Sone, Health Management Center, Toranomon Hospital, 2-2-2 Toranomon, Minato-ku, Tokyo 105-8470, Japan.
E-mail: hsone@md.tsukuba.ac.jp

Abstract

Diabet. Med. 29, e279-e285 (2012)

Abstract

Aim  To evaluate various screening criteria for pre-diabetes to identify which combination of impaired fasting glucose and elevated HbA1c values performs most effectively in predicting future diabetes in a large cohort of Japanese individuals.

Methods  The study included 4670 men and 1571 women without diabetes (diabetes: fasting plasma glucose ≥ 7.0 mmol/l, HbA1c≥ 48 mmol/mol (≥ 6.5%), or self-reported clinician-diagnosed diabetes). Pre-diabetes was diagnosed by a combination of impaired fasting glucose (fasting plasma glucose 5.6–6.9 mmol/l or 6.1–6.9 mmol/l) and elevated HbA1c [39–46 mmol/mol (5.7–6.4%) or 42–46 mmol/mol (6.0–6.4%)].

Results  During a 5-year follow-up, 338 incident cases of diabetes occurred. The combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and fasting plasma glucose 5.6–6.9 mmol/l yielded the highest sensitivity (86%) and generated a large population-attributable per cent risk (78%) for predicting development of diabetes. Among individuals classified as having pre-diabetes by any of the four combined criteria, 20.5–32.0% reverted to the normoglycaemic state as having neither elevated HbA1c nor impaired fasting glucose at the last follow-up examination. At 5.6 years after the baseline examination, however, pre-diabetic individuals who fulfilled both HbA1c 42–46 mmol/mol (6.0–6.4%) and fasting plasma glucose 6.1–6.9 mmol/l had a 100% cumulative risk of developing diabetes.

Conclusions  The combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and fasting plasma glucose 5.6–6.9 mmol/l would have the best performance in reducing the likelihood of missing future cases of diabetes. Identifying pre-diabetic individuals who strictly fulfil HbA1c 42–46 mmol/mol (6.0–6.4%) and fasting plasma glucose 6.1–6.9 mmol/l would predict definite progression to diabetes.

Introduction

Although early detection of pre-diabetic individuals greatly reduces the risk of progression to Type 2 diabetes [1], a recent report following the Diabetes Prevention Project [2] showed that only 4.8% of individuals with pre-diabetes reported a formal diagnosis from their physicians and that the majority of pre-diabetes individuals were undiagnosed. A meta-analysis of randomized controlled clinical trials showed that early interventions in pre-diabetic individuals successfully delayed or prevented progression to diabetes, although it remained unclear if early treatment of pre-diabetes affected major cardiovascular events [3]. Effective screening of pre-diabetic individuals has become a major priority to prevent a global diabetes epidemic, given the increasing prevalence of individuals with pre-diabetes and diabetes [4]. However, the complex history of assessment of non-diabetic hyperglycaemia or pre-diabetes makes it difficult to target persons most likely to progress to diabetes and thereby to efficiently enable early intervention. In 1997 and 1999, the American Diabetes Association [5] and the World Health Organization [6], respectively, defined the term ‘impaired fasting glucose (IFG)’ as an intermediate state wherein glucose levels do not meet diagnostic criteria for diabetes but are above normal. Impaired fasting glucose was considered present at fasting plasma glucose (FPG) values of 6.1–6.9 mmol/l (110–125 mg/dl). In 2003, the American Diabetes Association lowered the threshold to define IFG from FPG 6.1 mmol/l (110 mg/dl) to 5.6 mmol/l (100 mg/dl) so that the population at risk of developing diabetes with IFG would be more comparable with that with impaired glucose tolerance as determined by oral glucose tolerance test results [7]. However, the WHO did not adopt the change in definition of IFG. In addition to IFG, in 2009, an International Expert Committee proposed that HbA1c in a range of 42–46 mmol/mol (6.0–6.4%) presented a ‘very high risk’ of progression to diabetes [8]. More recently, in 2010, the American Diabetes Association introduced HbA1c 39–46 mmol/mol (5.7–6.4%) into diagnostic criteria for pre-diabetes to identify individuals at high risk for future diabetes [9]. Currently, a number of approaches exist to screen people at high risk using FPG and HbA1c with alternative cut-off points, such as FPG of 5.6–6.9 mmol/l or 6.1–6.9 mmol/l and HbA1c of 39–46 mmol/mol (5.7–6.4%) or 42–46 mmol/mol (6.0–6.4%). Cross-sectional studies reported that the prevalence of individuals classified as having pre-diabetes varied according to the glycaemic markers used for screening [10,11].

We recently reported that pre-diabetic individuals with FPG 5.6–6.9 mmol/l and HbA1c 39–46 mmol/mol (5.7–6.4%) had a markedly increased risk of diabetes when we diagnosed pre-diabetes according to the American Diabetes Association criteria [12]. Although evidence has shown that combining the two tests improved risk prediction [12–17], the performance of various screening criteria for pre-diabetes as a predictor of diabetes has not been evaluated using FPG of 5.6–6.9 mmol/l or 6.1–6.9 mmol/l and HbA1c of 39–46 mmol/mol (5.7–6.4%) or 42–46 mmol/mol (6.0–6.4%) as cut-off points. Furthermore, the potential impact of each screening strategy on the total population, including the increased risk revealed by screening results and the prevalence of individuals classified as having pre-diabetes, has not been clarified. In this study, we aimed to identify which combination of IFG and elevated HbA1c criteria using various cut-off points would perform most effectively in predicting future diabetes in a large cohort of Japanese individuals.

Patients and methods

Study population

The Toranomon Hospital Health Management Center Study included a cohort consisting mainly of apparently healthy Japanese government employees who underwent annual examinations for health screening in addition to some members of the general public. Details of the study have been described previously [12]. Routine health check-ups are the norm in Japan because the Japanese government and companies encourage people to have periodic health examinations [18]. Among 32 057 individuals who underwent the annual medical examinations, with the baseline examinations taking place from 1997 to 2003, we examined data on 6636 individuals who received annual examinations during a 4- or 5-year follow-up period after the baseline examination. Information on demographic characteristics, medical history and health-related habits was gathered at each annual examination using standard questionnaires. We excluded 310 individuals who had diabetes at the baseline examination (192 previously diagnosed and 118 undiagnosed) or with missing data on baseline characteristics (n = 89). Four individuals had both missing data and diabetes at baseline and therefore were counted in both categories. After this adjustment, 6241 individuals aged 24–82 years were eligible for analysis.

Diagnosis of diabetes and classification of pre-diabetes

Diagnosis of diabetes was made according to a FPG level ≥ 7.0 mmol/l, self-reported clinician-diagnosed diabetes or HbA1c≥ 48 mmol/mol (≥ 6.5%) [9]. Impaired fasting glucose was indicated by either FPG 5.6–6.9 mmol/l or FPG 6.1–6.9 mmol/l [5,6,9]. An elevated HbA1c level was indicated by HbA1c 39–46 mmol/mol (5.7–6.4%) or HbA1c 42–46 mmol/mol (6.0–6.4%) [8,9]. We combined the two tests to diagnose pre-diabetes. The definition of the normoglycaemic state varied according to which single criterion of FPG or HbA1c was used or whether a combination of the two was used for screening. Therefore, we defined ‘normoglycaemia’ as the absence of a positive test result by any screening criterion (as shown Table 2). For the four combinations, the risk of diabetes was also investigated when we considered discordant or concordant classifications (neither elevated HbA1c nor IFG, IFG alone, elevated HbA1c alone, or both elevated HbA1c and IFG).

Clinical measurements

Blood samples were collected after an overnight fast (12 h) and measured using an automatic clinical chemistry analyzer (LABOSPECT 008; Hitachi, Tokyo, Japan). Blood glucose, serum triglycerides, HDL cholesterol and uric acid concentrations were measured by enzymatic methods. Glycated haemoglobin was assessed by high-performance liquid chromat-ography (Tosoh, Tokyo, Japan). The value for HbA1c (%) was estimated as the National Glycohemoglobin Standardization Program (NGSP) equivalent value (%) which was calculated by the formula HbA1c (%) = HbA1c (Japan Diabetes Society) (%) + 0.4% [19]. The HbA1c measurements in International Federation of Clinical Chemistry (IFCC) units (mmol/mol) were also calculated.

Statistical analysis

To investigate definite progression to diabetes, the overall cumulative probability of developing diabetes was described by Kaplan–Meier analysis for individuals who were concordantly diagnosed as having pre-diabetes. Cox regression was used to estimate hazard ratios (HRs) and their 95% confidence intervals (CIs) for each baseline diagnosis of pre-diabetes with a group having neither elevated HbA1c nor IFG as the reference. Follow-up time for each participant was calculated from the date of the first examination to the date of confirmed diabetes or of the last follow-up examination. The property of each screening criterion for prediction of future diabetes was evaluated by sensitivity, specificity and the area under the receiver-operating characteristic curve (AUC). The AUC was calculated using age- and sex-adjusted HRs. We also calculated the population attributable risk per cent (PAR%) as follows: (Pe [HR-1])/(Pe [HR-1] + 1), where Pe = prevalence of pre-diabetes screened by each criterion in the population. The PAR% value indicates the potential impact of each screening criterion on the study population, including the effects of HRs and prevalence of individuals classified as having pre-diabetes [20–22]. All statistical analyses were performed by SPSS (version 16.0; SPSS, Chicago, IL, USA) and statistical significance was considered for P < 0.05.

The study protocol followed the Japanese Government’s Ethical Guidelines Regarding epidemiological studies in accordance with the Declaration of Helsinki and was reviewed by the Institutional Review Board at Toranomon Hospital.

Results

The mean (SD) age of the 6241 study participants was 49.9 years (8.7) and mean (SD) body mass index was 22.8 kg/m2 (2.8); the FPG and HbA1c levels at baseline were 5.3 mmol/l (0.5) and 5.3% (0.3), respectively (Table 1). Prevalence of individuals diagnosed as having pre-diabetes substantially differed according to each screening criterion at baseline; prevalence was 26.9% for FPG 5.6–6.9 mmol/l, 6.1% for FPG 6.1–6.9 mmol/l, 13.2% for HbA1c 39–46 mmol/mol (5.7–6.4%) and 3.3% for HbA1c 42–46 mmol/mol (6.0–6.4%) among the subject population (Table 2). When a combination of two tests was used, the prevalence of pre-diabetes increased to 33.5% for HbA1c 39–46 mmol/mol (5.7–6.4%) and/or FPG 5.6–6.9 mmol/l, 28.0% for HbA1c 42–46 mmol/mol (6.0–6.4%) and/or FPG 5.6–6.9 mmol/l, 16.7% for HbA1c 39–46 mmol/mol (5.7–6.4%) and/or FPG 6.1–6.9 mmol/l and 8.2% for HbA1c 42–46 mmol/mol (6.0–6.4%) and/or FPG 6.1–6.9 mmol/l. Even when alternative cut-off values were used, overlap was poor for the two tests (Fig1a).

Table 1.   Baseline characteristics of study participants (n = 6241)
Characteristic
  1. Data are mean ± SD, n (%) or median (25th, 75th).

Age (year)49.9 (8.7)
Males, n (%) 4670 (74.8)
Body mass index (kg/m2)22.8 (2.8)
Fasting plasma glucose (mmol/l) 5.3 (0.5)
HbA1c (mmol/mol)43 (4)
HbA1c (%) 5.3 (0.3)
Systolic blood pressure (mmHg)125 (16)
Diastolic blood pressure (mmHg)76 (11)
Triglycerides (mmol/l)1.07 (0.77, 1.57)
Total cholesterol (mmol/l)5.27 (0.84)
HDL cholesterol (mmol/l)1.40 (0.39)
Table 2.   Screening performance using impaired fasting glucose (IFG) and/or elevated HbA1c with alternative cut-off values for predicting the development of Type 2 diabetes
  n (% of total population)Incident cases/person-yearsIncidence rate (per 1000 person-years)HRAUCSensitivity (%)Specificity (%)PAR (%)*Reversion to normoglycaemia at follow up (%)
  1. FPG, fasting plasma glucose.

  2. Age- and sex-adjusted hazard ratios (HRs). The HRs and area under the curve (AUC) all indicated P < 0.001. Population attributable risk percent (PAR%): (Pe [HR–1])/(Pe [HR–1]+1). *Reversion to normoglycaemia was indicated by absence of a positive test result by any screening method at the last follow-up examination.

  3. Value for HbA1c (%) was estimated as the National Glycohemoglobin Standardization Program equivalent value (%) was calculated by the formula HbA1c (%) = HbA1c (Japan Diabetes Society) (%) + 0.4%.

Single test using either FPG or HbA1c
 FPG 5.6–6.9 mmol/l1680 (26.9%)262/756234.68.830.806787668383 (22.8)
 FPG 6.1–6.9 mmol/l380 (6.1%)155/1490104.014.10.789469644101 (26.6)
 HbA1c 39–46 mmol/mol (5.7–6.4%)822 (13.2%)184/360651.08.560.781548950263 (32.0)
 HbA1c 42–46 mmol/mol (6.0–6.4%)203 (3.3%)100/774129.215.10.73730983263 (31.0)
Combination of HbA1c and FPG
 HbA1c 39–46 mmol/mol (5.7–6.4%) and/or FPG5.6–6.9 mmol/l 2092 (33.5%)292/952730.611.70.811867078428 (20.5)
 HbA1c 42–46 mmol/mol (6.0–6.4%) and/or FPG 5.6–6.9 mmol/l1748 (28.0%)271/788334.49.770.811807571392 (22.4)
 HbA1c 39–46 mmol/mol (5.7–6.4%) and/or FPG 6.1–6.9 mmol/l 1043 (16.7%)238/456252.212.70.825708666293 (28.1)
 HbA1c 42–46 mmol/mol (6.0–6.4%) and/or FPG 6.1–6.9 mmol/l 511 (8.2%)194/206793.916.30.820579556143 (28.0)

Over a median of a 5.0-year follow up, we documented 338 incident cases of diabetes. Screening by a single criterion was specific for predicting the risk of diabetes, and the incidence rate (per 1000 person-years) was higher than screening by a combination of two tests (Table 2). Using the HbA1c 42–46 mmol/mol (6.0–6.4%) criterion, the incidence rate was 129.2/1000 person-years, age- and sex-adjusted HR was 15.1 (P < 0.001), and specificity was 98% for future diabetes. However, with that criterion, sensitivity was lowest at 30% and PAR% was also lowest at 32% among the single criteria examined. Lowering the threshold of FPG 6.1 mmol/l to 5.6 mmol/l increased the PAR% by 55%, although the HR was decreased by 60%. Lowering the threshold of HbA1c from 42 mmol/mol (6.0%) to 39 mmol/mol (5.7%) also increased the PAR% by 61% and decreased the HR by 76%.

However, the combination of two tests identified more incident cases of diabetes and substantially improved the sensitivity and PAR% for predicting development of diabetes. The combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 5.6–6.9 mmol/l generated the highest sensitivity, which was 86%, whereas specificity was lowest at 70%. When we used the combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 6.1–6.9 mmol/l, the AUC for predicting the incidence of diabetes had the highest value at 0.825 (95% CI 0.801–0.849). However, of individuals classified as having pre-diabetes by any of the criteria listed in Table 2, 20.5–32.0% had reverted to normoglycaemia (no positive screening result) at the last follow-up examination and none of these individuals subsequently developed diabetes.

When we investigated the overlap between the two tests, results of Kaplan–Meier analysis showed that fulfilling both criteria more definitely identified progression to diabetes (Fig. 1b). Fulfilling both the combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 5.6–6.9 mmol/l criteria (Fig. 1b, d) was predictive of an almost 50% cumulative risk of developing diabetes at 5.6 years after the baseline examination, fulfilling combinations of both the HbA1c 6.0–6.4% (42–46 mmol/mol) and FPG 5.6–6.9 mmol/l criteria or combinations of both the HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 6.1–6.9 mmol/l criteria was predictive of an approximate 80% cumulative risk. However, we found that strictly fulfilling the combination of both HbA1c 42–46 mmol/mol (6.0–6.4%) and FPG 6.1–6.9 mmol/l criteria (Fig. 1b, a) was predictive of a 100% cumulative incidence rate of diabetes. The 28% reversion rate to the normoglycaemic state (HbA1c < 42 mmol/mol (< 6.0%) and FPG < 6.1 mmol/l) at the final follow up, as shown in Table 2, was influenced by results of the discordant classification of pre-diabetes by either HbA1c or FPG at the baseline examination. Pre-diabetic individuals who fulfilled both HbA1c 42–46 mmol/mol (6.0–6.4%) and FPG 6.1–6.9 mmol/l criteria had a 50% cumulative incidence rate at 2.0 years after the baseline examination (Fig. 1b, a).

Figure 1.

 Prevalence of pre-diabetes identified by a combination of impaired fasting glucose and elevated HbA1c with various cut-off points (a) and subsequent risk of diabetes according to overlap between the two tests at baseline classification (b). Baseline classification of pre-diabetes was made using a combination of HbA1c 42–46 mmol/mol (6.0–6.4%) or HbA1c 39–46 mmol/mol (5.7–6.4%) and fasting plasma glucose (FPG) 6.1–6.9 mmol/l or 5.6–6.9 mmol/l. FPG, fasting plasma glucose.

Multivariate adjusted Cox analysis demonstrated that, compared with individuals with neither elevated HbA1c nor IFG at each screening, those with the combinations of (1) HbA1c 39–46 mmol/mol (5.7–6.4%)/FPG 5.6–6.9 mmol/l, (2) HbA1c 42–46 mmol/mol (6.0–6.4%)/FPG 5.6–6.9 mmol/l, (3) HbA1c 39–46 mmol/mol (5.7–6.4%)/FPG 6.1–6.9 mmol/l and (4) HbA1c 42–46 mmol/mol (6.0–6.4%)/FPG 6.1–6.9 mmol/l had HRs of (1) 32.5 (95% CI 23.0–45.8), (2) 53.7 (95% CI 38.4–75.1), (3) 37.9 (95% CI 28.1–51.1), and (4) 52.3 (95% CI 37.8–72.3), respectively (Table 3). As 66.5% of those with HbA1c 42–46 mmol/mol (6.0–6.4%) also had FPG 5.6–6.9 mmol/l (Fig. 1a), we examined whether HbA1c 42–46 mmol/mol (6.0–6.4%) alone without IFG was strongly predictive of risk for future diabetes. Multivariate-adjusted HR for HbA1c 42–46 mmol/mol (6.0–6.4%) alone without FPG 5.6–6.9 mmol/l or FPG 6.1–6.9 mmol/l was 7.42 (95% CI 3.67–15.0) or 11.4 (95% CI 7.93–16.5), respectively, with the risk shown to be only slightly higher compared with the presence of FPG 5.6–6.9 mmol/l alone or FPG 6.1–6.9 mmol/l alone without the presence of HbA1c 42–46 mmol/mol (6.0–6.4%) [HR 5.97 (95% CI 4.46–7.99) or HR 10.8 (95% CI 8.20–14.2), respectively].

Table 3.   Hazard ratios (HRs) of incidence of diabetes according to baseline diagnosis of pre-diabetes by a combination of impaired fasting glucose (IFG) and elevated HbA1c using a various cut-off points
  n Age- and sex- adjusted HR (95% CI)Multivariate adjusted HR (95% CI)
  1. FPG, fasting plasma glucose. Ref, reference group.

  2. Multivariate model was adjusted for age, sex, smoking habit (never/former/current), parental history of diabetes, BMI, systolic blood pressure, HDL cholesterol and log-transformed triglycerides.

HbA1c 39–46 mmol/mol (5.7–6.4%)/FPG 5.6–6.9 mmol/l
 HbA1c < 39 mmol/mol (< 5.7%) and FPG < 5.6 mmol/l 41491.00 (Ref)1.00 (Ref)
 FPG 5.6–6.9 mmol/l alone12706.86 (4.84–9.71)6.18 (4.34–8.80)
 HbA1c 39–46 mmol/mol (5.7–6.4%) alone4126.53 (4.10–10.4)6.05 (3.79–9.64)
 Both HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 5.6–6.9 mmol/l41038.6 (27.6–54.0)32.5 (23.0–45.8)
HbA1c 42–46 mmol/mol (6.0–6.4%)/FPG 5.6–6.9 mmol/l
 HbA1c < 42 mmol/mol (< 6.0%) and FPG < 5.6 mmol/l44931.00 (Ref)1.00 (Ref)
 FPG 5.6–6.9 mmol/l alone15456.79 (5.11–9.04)5.97 (4.46–7.99)
 HbA1c 42–46 mmol/mol (6.0–6.4%) alone688.25 (4.10–16.6)7.42 (3.67–15.0)
 Both HbA1c 42–46 mmol/mol (6.0–6.4%) and FPG 5.6–6.9 mmol/l13565.5 (47.2–90.8)53.7 (38.4–75.1)
HbA1c 39–46 mmol/mol (5.7–6.4%)/FPG 6.1–6.9 mmol/l
 HbA1c < 39 mmol/mol (< 5.7%) and FPG < 6.1 mmol/l51981.00 (Ref)1.00 (Ref)
 FPG 6.1–6.9 mmol/l alone22112.5 (8.93–17.4)11.4 (8.09–16.1)
 HbA1c 39–46 mmol/mol (5.7–6.4%) alone6636.75 (5.01–9.09)6.26 (4.63–8.45)
 Both HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 6.1–6.9 mmol/l15945.8 (34.5–60.8)37.9 (28.1–51.1)
HbA1c 42–46 mmol/mol (6.0–6.4%)/FPG 6.1–6.9 mmol/l
 HbA1c < 42 mmol/mol (< 6.0%) and FPG < 6.1 mmol/l57301.00 (Ref)1.00 (Ref)
 FPG 6.1–6.9 mmol/l alone30812.3 (9.48–16.1)10.8 (8.20–14.2)
 HbA1c 42–46 mmol/mol (6.0–6.4%) alone13112.5 (8.73–18.0)11.4 (7.93–16.5)
 Both HbA1c 42–46 mmol/mol (6.0–6.4%) and FPG 6.1–6.9 mmol/l7263.3 (46.4–86.5)52.3 (37.8–72.3)

Discussion

In this study, we showed that a combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 5.6–6.9 mmol/l criteria would perform best in reducing the likelihood of missing future cases of diabetes. Fulfilling both HbA1c 42–46 mmol/mol (6.0–6.4%) and FPG 6.1–6.9 mmol/l criteria was strongly predictive of definite progression to diabetes. In addition, our results demonstrated that combining the two tests improved the PAR% for predicting risk of diabetes. As there has been little clarification of the potential impact on a population of various screening strategies for pre-diabetes, our current results could contribute to an enhanced screening strategy for predicting the risk of diabetes.

An investigation of the economic justification of screening for pre-diabetes reported that to conduct screening appears to be cost saving compared with not performing screening [23]. However, the high cost of HbA1c testing makes it difficult to apply this in addition to plasma glucose for mass screening in some countries. Our results suggest that combining the values of HbA1c 39–46 mmol/mol (5.7–6.4%) and FPG 5.6–6.9 mmol/l simultaneously would substantially contribute to detecting as many individuals at risk as possible. If, on the other hand, specificity or certainty of progression to diabetes is paramount, screening by HbA1c 42–46 mmol/mol (6.0–6.4%) or the combination of HbA1c 42–46 mmol/mol (6.0–6.4%) and FPG 6.1–6.9 mmol/l could be considered appropriate. For efficient use of limited resources, which combination to apply would depend on the goal of each screening strategy.

The prevalence of pre-diabetes by changing cut-off values for each criterion in this study was comparable to that reported in a recent National Health and Nutrition Examination Survey study [11]. Among 3627 adults in the USA, the prevalence of individuals classified as having pre-diabetes was 26.2% for FPG 5.6–6.9 mmol/l, 14.2% for HbA1c 39–46 mmol/mol (5.7–6.4%) and 7.0% for FPG 6.1–6.9 mmol/l. When the two tests were combined, the prevalence of individuals classified as having pre-diabetes increased to 32.2% for HbA1c 39–46 mmol/mol (5.7–6.4%) and/or FPG 5.6–6.9 mmol/l and to 18.1% for HbA1c 39–46 mmol/mol (5.7–6.4%) and/or FPG 6.1–6.9 mmol/l. However, the difference in prevalence of pre-diabetes as a predictor of Type 2 diabetes was not investigated in that study. Two recent cohort studies among 12 375 adults in the USA followed for an average 4.4 years [24] and among 919 Caucasians followed for over 15 years [25] showed that HbA1c in the range 42–46 mmol/mol (6.0–6.4%) was highly predictive of future diabetes. In the present study, we also observed that screening for pre-diabetes using a single HbA1c 42–46 mmol/mol (6.0–6.4%) criterion was specific in predicting future diabetes and was associated with a high incidence rate or HR. As a possible reason, we found that two-thirds of individuals with HbA1c 42–46 mmol/mol (6.0–6.4%) also had FPG 5.6–6.9 mmol/l. Screening by HbA1c 42–46 mmol/mol (6.0–6.4%) would be potentially associated with an abnormality in maintaining normal glucose levels, whereas we observed that screening using a single criterion had a lesser impact on the total subject population.

In a study investigating the population-attributable risk determined from different definitions of pre-diabetes based on IFG or IGT, results suggested that lowering the diagnostic threshold of FPG from 6.1 mmol/l to 5.6 mmol/l would increase the population-attributable risk and that the combination of IFG and IGT would improve the value of the population-attributable risk in their subject population [26]. However, the impact of introducing HbA1c criteria was not addressed in that investigation. In the present study, we also found that lowering the value that defined IFG increased the PAR% by 55%, although the HR was decreased by 38%, and that introducing HbA1c values into IFG criteria improved the PAR%. Also, combining the two tests for screening of pre-diabetic individuals improved the potential impact on the whole population in this study. A recent meta-analysis demonstrated that although the progression to diabetes from the pre-diabetic state could be prevented through early interventions, this did not result in reductions in cardiovascular disease mortality [3]. As several screening tests are available and various thresholds are used, our findings would contribute to determining which combination of elevated IFG and elevated HbA1c values would most effectively predict future diabetes.

The strength of our study is the large sample size that included annual data on HbA1c and FPG levels. Prevention is a major priority to avert a global diabetes epidemic, especially among Asians with an increasing prevalence of pre-diabetes and diabetes [27], and who are reported to have a higher risk for diabetes compared with their white counterparts [28]. Several limitations must be considered. First, data on oral glucose tolerance testing were not available. As it was reported that performing an oral glucose tolerance test captured more individuals who progressed to diabetes compared with those identified by IFG [26], a study that includes this data is needed. It is crucial to consider, however, that in routine clinical practice, a common situation occurs in which evaluation of the prognosis of a patient must be made without oral glucose tolerance test results. Tests to determine FPG and HbA1c are routinely available, so our findings have practical implications for the clinical setting. As another limitation, we cannot deny the possibility of selection bias as our study included individuals who had taken health examinations and, therefore, these individuals were more health conscious than those in the general population. That only 19% (n = 6241/32 057) of the potential subject population was followed in the current study might lead to overestimation of the prevalence of pre-diabetes and associated risks for diabetes. Conversely, although the overall characteristics of the subject population were similar to those in general population-based studies in Japan [29,30], risk factors for diabetes such as BMI and blood pressure were relatively favourable in the mainly male (75%) individuals, which might lead to underestimation of incident cases. Further investigation in other cohorts of different ethnicities would be needed to confirm our findings.

In conclusion, the combination of HbA1c 39–46 mmol/mol (5.7–6.4%) and fasting plasma glucose 5.6–6.9 mmol/l would perform best in reducing the number of missed future cases of diabetes and would improve the population attributable risk Identifying pre-diabetic individuals who strictly fulfil both HbA1c 42–46 mmol/mol (6.0–6.4%) and fasting plasma glucose 6.1–6.9 mmol/l criteria should be predictive of definite progression to diabetes, which would effectively allow for early interventions to avoid development of diabetes.

Competing interests

Nothing to declare.

Acknowledgements

This work is supported in part by the Ministry of Health, Labour and Welfare, Japan. H.So. is recipient of a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science. The sponsor had no role in the design and conduct of the study. All authors sincerely thank the late Professor and Director Kinori Kosaka, who established the foundation for the TOPICS project and was always the foremost pillar of spiritual support for the TOPICS project. Thanks are also extended to Satomi Fukuya for her excellent secretarial assistance.

Ancillary