Introduction

Diabetes mellitus (DM) is a significant global health issue that results in a considerable health and economic burden1. The current standard method of DM diagnosis involves an oral glucose tolerance test (OGTT) and measurement of HbA1c2. DM is diagnosed when any one of the following criteria is met: fasting plasma glucose ≥ 126 mg/dL, 2-h plasma glucose (OGTT-2 h PG) ≥ 200 mg/dL during OGTT, HbA1c ≥ 6.5%, or random plasma glucose ≥ 200 mg/dL with classical symptoms of hyperglycemia. However, OGTT can be cumbersome in some clinical settings and is frequently omitted during the diagnostic process of DM3.

In a real-world observational study conducted in Europe, patients were screened for diabetes following an event of coronary artery disease4. A total of 22.2% of subjects did not receive OGTT during the DM screening. Among those diagnosed with DM using OGTT and HbA1c, around 20% of them would not have been detected if OGTT was not performed. Similarly, several studies report a high percentage (13.74 to 44%) of patients who did not perform OGTT during the screening of DM5,6.

The diagnosis of DM without OGTT leads to underdiagnosis of glycemic status and may result in poorer health consequences. In the literature, a high OGTT-2 h PG is associated with a poorer cardiovascular risk factor profile, larger carotid intima-media thickness, and a higher Framingham risk score for the risk of coronary arterial disease7. Besides, OGTT-2 h PG is a prognostic indicator for subjects with acute coronary syndrome8 and myocardial infarction9. A high OGTT-2 h PG is also a well-known risk factor for diabetic microvascular complications, including diabetic retinopathy and nephropathy10. Post-prandial hyperglycemia was associated with cancer incidence and mortality in the literature11. Furthermore, glucose variability (GV) is known to be associated with higher risk of stroke, myocardial infarction and all-cause mortality12, and it has also emerged as a recognized risk factor for diabetic complications13. The information of GV could be missed if OGTT is not performed. A recent Mendelian randomization study also suggests that FPG, OGTT-2 h PG and HbA1c had different effects on various health outcomes14. Specifically, only OGTT-2 h PG, but not FPG or HbA1c, was associated with small vessel stroke and aortic aneurysm. This emphasizes the importance of OGTT. However, it remains unclear whether misdiagnosis of glycemic status due to not performing OGTT is associated with higher mortality or not.

Therefore, we conducted a prospective cohort study where we collected OGTT and HbA1c results at enrollment and followed these subjects for an average of 10 years. The aim of this study is to elucidate the impact of underdiagnosed glycemic status on all-cause mortality. We also identified target populations that were more likely to be underdiagnosed and developed screening strategies to improve diagnostic performance while minimizing the need for OGTT.

Methods

Study populations

The participants of this research were prospectively enrolled in the Taiwan Lifestyle Study, a community-based cohort study15,16,17. We recruited individuals aged 18 or above who resided in Yun-Lin County, Taiwan between 2006 and 2019. The subjects were enrolled on a volunteer basis via recruitment posters on the bulletin board at the Yun-Lin branch of National Taiwan University. Those who agreed to participate and signed written informed consent were recruited consecutively. For every study participant, we collected their demographic data, personal and family histories, blood pressure (BP), body mass index (BMI) and waist circumference (WC) measurements, as well as laboratory tests, including oral glucose tolerance tests and hemoglobin A1c (HbA1c), by study nurses and physicians. Participants were informed about and consented to the linkage of their collected data with national databases, such as the National Health Insurance Research Database (NHIRD), as stated in the written informed consent. The study protocol was approved by the institutional review board of National Taiwan University Hospital (NTUH IRB numbers: 9461701008, 200705039R, and 201412122RINC) and was conducted in accordance with the Declaration of Helsinki.

Diabetes mellitus definition and insulin resistance measurements

DM was defined according to the American Diabetes Association (ADA) recommendations using OGTT and HbA1c. DM was diagnosed by endocrinologists if any of the following criteria were met: fasting plasma glucose (FPG) ≥ 126 mg/dL, OGTT 2-h plasma glucose (OGTT-2 h PG) ≥ 200 mg/dL, or HbA1c ≥ 6.5%. There were 13 participants who did not complete the OGTT and HbA1c tests due to personal reasons (intolerable to OGTT in most subjects). Since we needed the complete data of OGTT and HbA1c to categorize the study subjects, these 13 subjects were excluded. Besides, since the primary objective of this study was to examine the significance of OGTT in individuals without diabetes at baseline and investigate whether undetected diabetes, resulting from the absence of OGTT, contributes to increased mortality rates, subjects with pre-existing diabetes, including those who received anti-diabetic medications (N = 41), were excluded. We also excluded 176 subjects diagnosed with DM by fasting plasma glucose ≥ 126 mg/dL and HbA1c ≥ 6.5%, as they would receive anti-diabetic medications after the diagnosis of DM, which may change the risk of further complications and mortality. Participants who were newly diagnosed with diabetes based on OGTT and/or HbA1c were referred to outpatient clinics for further evaluation and treatment in accordance with standard clinical care. Detailed study flowchart was illustrated in Fig. 1. Insulin resistance was estimated using the homeostatic model assessment (HOMA2-IR) based on plasma fasting glucose and insulin levels18.

Fig. 1
figure 1

Study flowchart. DM, diabetes mellitus; FPG, fasting plasma glucose; HbA1c, hemoglobin A1c; NHIRD, National Health Insurance Research Database; OGTT, oral glucose tolerance test.

Full size image

Categorization of correctly or incorrectly classified glycemic status

Based on OGTT and HbA1c results, subjects were classified as normoglycemia, prediabetes, or DM. Among these groups, participants were categorized into the “misclassified glycemic status by FPG and HbA1c” group (the exposed group) if both FPG and HbA1c failed to identify a glycemic abnormality that was detected by the OGTT-2 h PG. For example, participants classified as normoglycemic or prediabetic by FPG and HbA1c alone, but meeting the criteria for prediabetes or diabetes based on OGTT 2 h-PG, were considered misclassified. In contrast, participants whose glycemic status was accurately identified using FPG and HbA1c, regardless of OGTT results, were assigned to the “correctly classified glycemic status” group (the reference group). In other words, if the glycemic classification based on FPG and HbA1c alone was consistent with the classification based on the combination of FPG, HbA1c, and OGTT 2 h-PG, the participant was considered “correctly classified”. If the glycemic classification based on FPG and HbA1c alone differed from the classification based on the combination of FPG, HbA1c, and OGTT 2 h-PG, the participant was considered “misclassified”.

Detailed categorization of these groups is provided in Supplementary Table 1.

Metabolic syndrome definition

Metabolic Syndrome was defined according to the National Cholesterol Education Program (NCEP) Adult Treatment Panel III (ATP III) criteria, which included central obesity (≥ 90 cm or 80 cm in men or women respectively in Han Chinese populations), raised triglycerides (≥ 150 mg/dL), reduced HDL-cholesterol (< 40 mg/dL or 50 mg/dL in men or women, respectively), raised blood pressure (≥ 130/85 mmHg), and raised fasting plasma glucose (≥ 100 mg/dL)19. Metabolic Syndrome was diagnosed if 3 out of 5 components were present.

Assays for biochemical parameters

Plasma glucose and lipid profiles were measured using an automatic analyzer (Toshiba TBA 120FR, Toshiba Medical Systems Co., Ltd., Tokyo, Japan). Plasma insulin was tested using an automatic analyzer with microparticle enzyme immunoassay (Abbott AxSYM system; Abbott Laboratories, Abbott Park, IL). HbA1c was examined using an automatic analyzer (HLC-723 G7 HPLC system, Tosoh Corporation, Tokyo, Japan), which was certified by the National Glycohemoglobin Standardization Program (NGSP) and standardized to the Diabetes Control and Complications Trial (DCCT) reference assay.

Mortality record and data linkage to the NHIRD

The NHIRD was used to define mortality. The NHIRD was initiated in 1995 and utilized until present, which contains registered data and reimbursement claims of the National Health Insurance (NHI) program in Taiwan. The cause of death data (CDD) is one of the datasets in NHIRD, which contains all the registered death records in Taiwanese populations. Causes of death are coded using the International Classification of Diseases-9 or -10 (ICD-9 or ICD-10). The NHIRD contains only registered medical records, such as information on diagnoses, treatments, costs, and causes of death. We integrated the comprehensive demographic data, personal history, and laboratory data collected from the Taiwan Lifestyle Study at enrollment and linked this dataset with the NHIRD by anonymized identifiers to obtain the survival status of all the participants enrolled in the study until 30/08/2019. All enrolled subjects, including those classified as having normoglycemia, prediabetes, or DM, were followed for mortality outcomes. The follow-up duration was calculated from the time of enrollment until either the occurrence of death or the end of the follow-up period, which was 30/08/2019. The dataset of NHIRD was first accessed on 07/09/2020.

Development of screening strategies for diabetes mellitus in subjects aged over 60 years

It is widely recognized that advanced age is associated with a higher prevalence of metabolic syndrome20, and we did observe a significantly higher presence of metabolic syndrome components in individuals aged over 60 in the present study (Supplementary Table 4). Consequently, we developed three novel screening strategies for diabetes based on the presence of components of metabolic syndrome, in order to minimize the necessity of performing the OGTT while having a good diagnostic efficacy in individuals aged 60 and above.

In strategy 1, OGTT is recommended for individuals aged over 60 who exhibit any component of metabolic syndrome. In strategy 2, OGTT is suggested for individuals aged 60 and above who have either elevated blood pressure or increased waist circumference, which are the two components of metabolic syndrome that does not require blood tests. Therefore, this approach provides a more practical alternative for clinical use. Lastly, strategy 3 entails recommending an OGTT for individuals aged 60 and above who had metabolic syndrome, i.e. had three or more components of metabolic syndrome.

Statistical analyses

Categorical variables were presented as numbers (percentages), while continuous variables with normal distribution were shown as means ± standard deviation (SD). Continuous variables without normal distribution were reported as medians (inter-quartile ranges) and logarithmic transformation was performed for statistical analysis. Statistical differences were analyzed using t-tests or chi-squared tests. Cox proportional hazards models were used to estimate hazard ratios (HRs) for all-cause mortality. Significant differences between subjects with correct and incorrect glycemic status diagnoses, and between survivors and non-survivors, were considered potential confounders. Additionally, variables associated with mortality in the literature, including age21, gender21, BMI22, WC23, smoking24, alcohol25, hypertension26, HOMA2%-IR27, TG28, LDL-C29, HDL-C30, and hsCRP31, were considered confounders. Model 1 included only age. Model 2 used stepwise model selection for confounders, with or without age adjustment. Model 3 included all confounders, with or without age adjustment. Since age was a significant confounder influencing the relationship between misclassified glycemic status and mortality, a subgroup analysis was conducted for participants below and above 60 years old. The 60-year age threshold was based on WHO32 and United Nations33 definitions. Screening strategy performance was analyzed using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and numbers needed to screen (NNS). A two-tailed p-value below 0.05 was considered significant. Statistical analyses were performed using Stata/SE 14.0 for Windows (StataCorp LP, College Station, TX).

Results

We analyzed 1935 subjects over a median follow-up period of 10.2 years (interquartile range 8.7–11.7 years). Based on OGTT and HbA1c results, subjects were classified as normoglycemia, prediabetes, or DM (Supplementary Table 1). Of these groups, 156 subjects were incorrectly classified into normoglycemia (N = 101) or prediabetes (N = 55) if the diagnosis was made only by FPG and HbA1c, and these subjects were categorized as the “misclassified glycemic status by FPG and HbA1c” group (the exposed group). On the other hand, 1779 subjects whose glycemic status was correctly determined using only fasting plasma glucose and HbA1c were utilized as the reference group, the “glycemic status correctly classified” group. We compared baseline characteristics between these two groups (Table 1). The misclassified group had a significantly higher percentage of all-cause mortality compared to those correctly classified (8.3% vs. 3.6%, p = 0.004). Additionally, subjects with misclassified glycemic status had a shorter follow-up duration, were older, had a larger BMI and waist circumference, higher blood pressure, worse glycemic profiles including fasting and OGTT-2 h PG, higher fasting insulin, more insulin resistance, higher TG, higher hsCRP, and were more likely to have metabolic syndrome. Furthermore, comparison of subjects who died or survived throughout the follow-up period was presented in Supplementary Table 2. The subjects who died during follow-up exhibited a greater likelihood of misclassified glycemic status, a higher percentage of having metabolic syndrome, advanced age, a higher proportion of male gender, increased waist circumferences, elevated proportion of smoking, higher blood pressure levels, worse OGTT-2 h PG and HbA1c values, as well as elevated levels of hsCRP.

Table 1 Baseline characteristics by subgroups with correct or incorrect classification of the glycemic status.
Full size table

We analyzed hazard ratios (HRs) for all-cause mortality in those with misclassified glycemic status compared to those without (Supplementary Table 3). The unadjusted HR was 2.537 (95% confidence interval (CI) 1.397–4.607, p = 0.002), but after adjusting for age (Model 1), the HR decreased dramatically to 1.685 and was no longer statistically significant (95% CI 0.926–3.070, p = 0.088). In model 2, the HR of misdiagnosis of glycemic status for mortality was significantly higher after adjusted for gender, BMI, and WC (HR 2.597; 95% CI 1.421–4.747; p = 0.002). Similarly, in model 3, the HR for mortality remained statistically higher which involved multiple variables for adjustment including gender, BMI, WC, smoking, alcohol, hypertension, HOMA2%-IR, TG, LDL-C, HDL-C, and hsCRP (HR 2.451; 95% CI 1.301–4.613; p = 0.005). When age was included in Model 2 or Model 3, the HR for mortality reduced but was still in favor of increased risk based on the 95% CI, although the p value became non-significant (HR of misclassified glycemic status, in model 2 + age: 1.770 (95% CI 0.960–3.262, p = 0.067, in model 3 + age: 1.707, 95% CI 0.897–3.250, p = 0.104). These findings indicate that age is an important confounding factor in the relationship between misclassified glycemic status due to the lack of OGTT and mortality.

Therefore, we conducted a subgroup analysis by age and found that individuals over 60 were more likely to be misclassified in this age group and linked to a significantly increased risk of mortality (Table 2). The hazard ratios for mortality associated with misclassified glycemic status, stratified by age group, were illustrated in Fig. 2. In the ≥ 60 age group, those with misclassified glycemic status had higher hazard ratio of mortality (crude HR 2.082, 95% CI 1.098–3.946, p = 0.025). Multivariable adjustment analysis demonstrated that the HRs for mortality associated with misclassified glycemic status were significantly elevated in model 1, model 2, model 2 + age, model 3, and model 3 + age (HRs of 1.937, 2.505, 2.081, 2.268 and 2.043, respectively, all with p < 0.05). These findings suggest a positive association between misclassified glycemic status and increased risk of mortality in subjects with age over 60. Those aged more than 60 also had larger BMI, higher blood pressure, glucose levels, insulin resistance, triglyceride level, and hsCRP (Supplementary Table 4).

Table 2 Subgroup analysis in different age groups. Hazard ratios (HR) (95% confidence interval (CI)) for mortality are shown in the two age groups.
Full size table
Fig. 2
figure 2

Age-stratified hazard ratios for mortality from crude (unadjusted) and adjusted models. Forest plots present hazard ratios (HRs) and 95% confidence intervals (CIs) for all-cause mortality among individuals aged < 60 years (Panel A) and ≥ 60 years (Panel B). Model 1: adjusted for age. Model 2: adjusted for variables which were statistically significant in stepwise model selection, including gender, BMI and WC. Model 2 + age: adjusted for age, gender, BMI and WC. Model 3: adjusted for all potential confounders other than age, including gender, BMI, WC, smoking, alcohol, hypertension, HOMA2%-IR, TG, LDL-C, HDL-C and hsCRP. Model 3 + age: adjusted for age, gender, BMI, WC, smoking, alcohol, hypertension, HOMA2%-IR, TG, LDL-C, HDL-C and hsCRP.

Full size image

To develop an effective screening strategy for correctly diagnosing DM and minimizing the need for OGTT, we developed three screening strategies based on age and the presence of metabolic syndrome (Fig. 3). Strategy 1 recommended OGTT and HbA1c tests for individuals aged over 60 who met any one component of metabolic syndrome. In strategy 2, individuals aged over 60 who met the blood pressure or waist circumference criteria of metabolic syndrome should undergo OGTT and HbA1c tests. Strategy 3 recommended OGTT and HbA1c tests for individuals aged over 60 who had metabolic syndrome (three or more components). Using only FPG and HbA1c for DM diagnosis had low sensitivity (74.0%). Strategies 1, 2, and 3 had higher sensitivity (89.9%, 88.7%, and 80.7%, respectively) and required fewer OGTTs to diagnose one individual with DM (4.0, 3.8, and 2.3, respectively) (Table 3).

Fig. 3
figure 3

Different screening strategies of diabetes in all study populations. (A) Strategy 1: use OGTT and HbA1c if age ≥ 60 years plus any one components of metabolic syndrome, while others using only FPG and HbA1c to diagnose DM. (B) Strategy 2: use OGTT and HbA1c if age ≥ 60 years plus either the presence of blood pressure or waist circumference component of metabolic syndrome, while others using only FPG and HbA1c to diagnose DM. (C) Strategy 3: age ≥ 60 years plus any three or more components of metabolic syndrome, while others using only FPG and HbA1c to diagnose DM. BP, blood pressure; DM, diabetes mellitus; FPG, fasting plasma glucose; MS, metabolic syndrome, OGTT, oral glucose tolerance test; WC, waist circumference.

Full size image
Table 3 Performance of different screening strategies for diabetes in all study subjects.
Full size table

Discussion

In the present study, we showed that misclassification of glycemic status due to omission of OGTT was associated with higher mortality. Age played an important role and was associated with the percentage of incorrect diagnosis and mortality. Subjects aged over 60 had the highest percentage of incorrect diagnosis and mortality. In this age group, the presence of components of metabolic syndrome were significantly different between those with and without correct diagnosis of glycemic status. Therefore, three screening strategies to decide the need of OGTT were proposed based on the presence of components of metabolic syndrome. In strategy 1 and 2, OGTT and HbA1c were performed in those who met any components of metabolic syndrome (strategy 1) or those who met either blood pressure or waist circumference criterion (strategy 2), 19.4% and 17.6% of subjects should receive OGTT and HbA1c, and the sensitivity was 89.9% and 88.7%, respectively. In strategy 3, OGTT and HbA1c were done in those with 3 or more components of metabolic syndrome, which resulted in fewer OGTT needed (5.9%), but the sensitivity was lower (80.7%).

In the present study, we found that subjects whose glycemic status were misclassified by FPG and HbA1c alone had a higher mortality. This finding is supported by previous reports showing the prognostic value of post-load hyperglycemia during OGTT. A large cohort study in China disclosed that post-load hyperglycemia (OGTT-2 h PG ≥ 200 mg/dL) was more closely associated with incident cardiovascular events, cancers, and mortalities, compared with fasting hyperglycemia (FPG ≥ 126 mg/dL) or high HbA1c (≥ 6.5%)34. Two-hour post-load plasma glucose was used to improve the prognostic prediction in subjects with coronary artery disease or acute coronary syndrome in several studies8,35,36,37. Every one mmol/L (18 mg/dL) increase in OGTT-2 h PG, the risk of major cardiovascular events (MACE) increased by 8–11%8. In fact, the current diabetic diagnostic threshold of OGTT-2 h PG (≥ 200 mg/dL) adopted by the ADA was amended and established based on several studies regarding glycemic cut-off of two-hour post-load glucose and its relationship with risks of diabetic microvascular complications10.

The importance of OGTT-2 h PG during the screening of DM cannot be overemphasized. In our previous report, we screened subjects without a history of diabetes by OGTT and HbA1c38. Among the 8.3% of subjects with newly diagnosed DM, 52% of them had isolated post-load hyperglycemia. In other words, over half of them would not have been found to have diabetes if OGTT was not performed. Another observational study in subjects with coronary artery disease also revealed that 18.83% participants with newly diagnosed diabetes were found by OGTT-2 h PG, but not by FPG or HbA1c4. In the National Health and Nutrition Examination Surveys (NHANES), the overall prevalence of undiagnosed DM was 2.9% by FPG and HbA1c criteria39. This number increased 1.72-fold to 5% if FPG, OGTT-2 h PG and HbA1c were used to screen DM. In a recent Mendelian randomization study, it was found that FPG, OGTT-2 h PG, and HbA1c had varying effects on different atherosclerotic and thrombotic outcomes14. These findings suggest that these glycemic traits are distinct from each other, and OGTT-2 h PG may play a unique role in contributing to health outcomes, highlighting its significance and emphasizing that it should not be overlooked. Several studies have demonstrated that isolated post-load hyperglycemia is associated with significant insulin resistance, even when fasting glucose levels remain within normal or prediabetic ranges40. This supports our finding that elevated HOMA2%-IR in the misclassified group may be driven by disproportionately high fasting insulin levels in the context of early compensatory mechanisms. In the present study, 8.3% of participants would have been misclassified in terms of glycemic status if OGTT had not been performed. This misclassification may have led to delayed diagnosis and treatment, potentially contributing to the increased mortality observed in this group. These findings underscore the importance of including OGTT in glycemic screening strategies, particularly in high-risk populations. Since OGTT is time-consuming and may be cumbersome clinically, it is important to identify subjects who would benefit most if OGTT is performed.

Age was identified as an important variable affecting the accuracy to correctly classify one’s glycemic status by FPG and HbA1c only in this study. Nearly one-sixth (16.6%) of subjects had incorrect diagnosis of diabetes in elderly subjects aged over 60. Our findings were consistent with the results in the literature. Since 2010, the ADA recommended HbA1c to be one of the diagnostic criteria of glycemic status41. However, HbA1c has several limitations and can be incorrect in certain conditions, such as conditions that alter RBC’s life span10. One study in France reported a total of 18% of undiagnosed prediabetes and diabetes if only FPG and HbA1c were used in overweight or obese patients. Age and waist circumference were independently associated with high OGTT-2 h PG. In addition, with every 10-year increase in age, the risk of glycemic status misclassification due to the absence of OGTT-2 h PG increased by 28 to 78%42. Another study in the U.S. showed that aging was associated with increased HbA1c levels and decreased diagnostic specificity of HbA1c43. One recent study investigated the misclassification of prediabetes and diabetes in the National Health and Nutrition Examination Surveys (NHANES) from 2005 to 2016 using hemoglobin glycation index (HGI), the difference between one’s observed and predicted HbA1c. A total of 15% subjects showed clinically significant glycemic mismatches, which potentially lead to HbA1c-related misdiagnosis and inappropriate management, especially in older adults or non-Hispanic Black individuals44. Taken together, all these findings suggest that HbA1c and FPG alone are not sufficient to correctly diagnose glycemic status in older adults, and OGTT is needed. As compared with older adults, younger individuals are less likely to be underdiagnosed with diabetes (6.3% vs. 14.6% in our study) and tend to have a more favorable health status even if their diabetes remains undiagnosed. These factors may contribute to the lack of significance in mortality observed in our study. However, early identification of dysglycemia in younger population remains clinically important, as long-term complications may still develop without timely intervention.

Increasing the diagnostic accuracy while avoidance of unnecessary OGTT remains a challenge in the screening of diabetes. Several risk score-based strategies were proposed, such as the Finnish Diabetes Risk Score (FINDRISC)45, but the results were not good enough. In our previous report, we used FPG and A1c to determine the need of OGTT46,47. OGTTs were performed if FPG was between 100 and 125 mg/dL or HbA1c was between 6.1 and 6.4%. The sensitivity and specificity for this strategy were 85.2 and 100%, and the numbers of OGTT required were decreased by 81.8%. In the present study, we found that age was an important determinant for the correctness of the diagnosis of glycemic status. Therefore, we developed three screening strategies based on age and the presence of metabolic syndrome. Strategy 1 or strategy 2 showed high sensitivity (88.7 to 89.9%), reduced needs of OGTT (17.6 to 19.4%), and reasonable numbers needed to screen (NNS) of 3.8 to 4.0. In other words, one case of diabetes could be detected in every 3 to 4 OGTTs performed. Strategy 2 may be more feasible than strategy 1, since only blood pressure and waist circumference are needed, and blood tests are not required. For the same reason, strategy 1 needs two visits to complete the screening, whereas strategy 2 can be completed in a single visit. In strategy 3, the need of OGTT and NNS decreased to 5.9% and 2.3, respectively. However, the sensitivity also reduced to 80.7%. Since this strategy also requires two visits to be completed, it may be useful only when the need of OGTT should be minimized. Considering the potential complications and increased mortality resulted from underdiagnosed diabetes, the balance between OGTT percentage and sensitivity, and clinical feasibility, strategy 2 is more preferred to screen DM in elder subjects.

To our best knowledge, this is the first study unveiling that subjects with incorrect diagnosis of glycemic status due to lack of OGTT were associated with a higher all-cause mortality, and the consequences were more profound in subjects aged more than 60. In addition, age-specific screening strategy for diabetes was proposed for the first time in response to the fact that the percentage of underdiagnosis by merely FPG and HbA1c increased by age. Furthermore, components of metabolic syndrome were incorporated in designing the screening strategy of DM, and three strategies for screening of diabetes were proposed. One could balance the accuracy to diagnose DM and the percentage of OGTT required to choose the best screening strategy in different clinical settings and needs.

There were some limitations of this study. First, all participants in this study were Taiwanese, so the findings should be tested in other ethnic groups. Second, the recruitment of participants was based on a voluntary basis rather than random sampling, which may limit the generalizability of the results. Participants who volunteered for the study might have had a higher level of health awareness and healthier lifestyle compared to the general population. Third, only 77 out of 1935 participants (4.0%) died during the follow-up period, with 43.1% due to cancer, 12.8% due to cardiovascular disease, and 7.4% due to cerebrovascular disease. The relatively low mortality rate may limit the statistical power, especially the impact of glycemic misclassification on specific causes of death. Moreover, we were not able to obtain all comorbidities such as dementia or chronic obstructive pulmonary disease (COPD) for adjustment in every subject enrolled. Therefore, there may be some residual confounding effects from these unmeasured factors. However, it is worth noting that these differences in prevalence, health awareness and the presence of comorbidities did not impact the relationship between misclassified glycemic status and mortality, nor did they affect the sensitivity and specificity of the screening strategies. Besides, only baseline data were used. Therefore, changes of clinical factors were not considered in the analyses and in the development of screening strategy. However, using data in the baseline only has a better feasibility than using data both in the baseline and during follow-up. Last but not least, the high within-person variability of blood glucose may limit the diagnostic accuracy of the OGTT48. In addition, the one-hour OGTT has been proposed by the International Diabetes Federation (IDF) to diagnose intermediate hyperglycemia (IH) and diabetes, using OGTT-1 h PG cutoffs of 155 mg/dL and 209 mg/dL, respectively49. This approach may enable earlier detection of dysglycemia and facilitate timely intervention, owing to the relative simplicity and greater acceptability of the OGTT-1 h PG measurement, although it was not analyzed or compared in the current study.

In conclusion, we have demonstrated that diagnosis of diabetes without OGTT leads to underdiagnosed diabetes, which is associated with a higher mortality, especially in elderly adults aged over 60 years. We developed three screening strategies based on age and components of metabolic syndrome to determine if OGTT is required. These strategies could improve the accuracy on the diagnosis of DM and reduce the need for OGTT. Since subjects of diabetes identified by the strategies will be treated, it may reduce the incidence of complication by underdiagnosis by FPG and HbA1c only and improve their prognosis.