Effects of Age and Sex on Estimated Diabetes Prevalence Using Different Diagnostic Criteria: The Tromsø OGTT Study

HbA1c 6.5% has recently been recommended as an alternative diagnostic criterion for diabetes. The aims of the study were to evaluate the effects of age, sex, and other factors on prevalence of diabetes and to compare risk profiles of subjects with diabetes when defined by HbA1c and glucose criteria. Subjects were recruited among participants in the longitudinal population-based Tromsø Study. HbA1c, fasting plasma glucose, and 2-hour plasma glucose were measured in 3,476 subjects. In total, 294 subjects met one or more of the diagnostic criteria for diabetes; 95 met the HbA1c criterion only, 130 met the glucose criteria only, and 69 met both. Among subjects with diabetes detected by glucose criteria (regardless of HbA1c), isolated raised 2-hour plasma glucose was more common in subjects aged ≥ 60 years as compared to younger subjects and in elderly women as compared to elderly men. Subjects with diabetes detected by glucose criteria only had worse cardiometabolic risk profiles than those detected by HbA1c only. In conclusion, the current HbA1c and glucose criteria defined different subjects with diabetes with only modest overlap. Among a substantial proportion of elderly subjects, and especially elderly women, the 2-hour plasma glucose was the only abnormal value.


Introduction
Criteria for the diagnosis of diabetes are based on measurements of fasting plasma glucose (FPG), 2-hour plasma glucose (2hPG), or haemoglobin A 1c (HbA 1c ). Single raised values with symptoms or raised values on two occasions of any one of these tests, or a combination of these tests can be used for diagnosis of diabetes [1,2]. e most commonly used test is the FPG as it is simple and inexpensive. e 2hPG is measured in combination with FPG in the oral glucose tolerance test (OGTT), where plasma glucose is measured in the morning aer an overnight fast and 2 hours aer oral ingestion of 75 g glucose. HbA 1c was recently introduced as a diagnostic test for diabetes. Compared to glucose measurements, HbA 1c has better sample stability, lower within-person variation and is independent of acute factors such as illness, recent food ingestion, stress, or exercise [3].
Diagnostic levels of FPG, 2hPG, and HbA 1c are based on thresholds for increased risk of micro-and macrovascular disease, in particular retinopathy [1,4]. In the DETECT-2 study, sensitivity and speci�city for prediction of prevalent retinopathy were almost equal when comparing FPG, 2hPG and HbA 1c [5]. Several recent studies have shown that both the prevalence of diabetes and the subjects diagnosed with diabetes vary when different diagnostic criteria for diabetes are applied [6][7][8][9][10][11]. According to current guidelines, clinicians can choose freely among FPG, OGTT, and HbA 1c when testing a patient for diabetes [1,2]. As HbA 1c and glucose criteria have been shown to identify different subjects with diabetes with relatively modest overlap, the choice of test may affect the test outcome [6,7,9]. is is important both at the individual level, where correct diagnosis, treatment, and prevention of later complications are in focus, and at the population level where early identi�cation of the "correct" individuals at risk of developing complications is important for cost-effective utilisation of resources. Furthermore, race, age, and sex have been reported to affect the outcome of diabetes testing with different diagnostic criteria [6-8, 12, 13]. is could have implications for the preferred choice of test in subgroups of patients. In Tromsø we have recently performed a large health survey where we measured HbA 1c , FPG, and 2hPG in 3,476 subjects without previously diagnosed diabetes. ese data enabled us to study the effect of age, sex, and other factors on diabetes de�ned by different diagnostic criteria and to compare cardiometabolic risk pro�les of subjects with diabetes de�ned by different criteria.

Subjects.
Subjects were recruited from the sixth survey of the longitudinal population-based Tromsø Study performed by the University of Tromsø from October 2007 to December 2008, where HbA 1c was measured in 12,769 participants. All subjects without self-reported diabetes and with HbA 1c in the range 5.8-6.9% and a random sample of approximately 200 subjects with HbA 1c 5.3% and 5.4% and 100 subjects with HbA 1c 5.5%, 5.6%, and 5.7%, respectively, were invited to participate in the Tromsø OGTT Study. Race was not registered, but practically all subjects were Caucasian.

Measurements.
Waist and hip circumference, height, weight, and blood pressure were measured, body mass index (BMI) was de�ned, and physical activity score (PAS) was calculated as previously described [14]. HbA 1c was determined by high performance liquid chromatography (HPLC) using an automated analyser (Variant II, Bio-Rad Laboratories Inc., Hercules, CA, USA). e reference interval was 4.3-6.1%. is analysis has been certi�ed by the National Glycohemoglobin Standardization Program (NGSP) as having documented traceability to the Diabetes Control and Complication Trial (DCCT) reference method [15]. Haemoglobin (Hb) was measured by photometry using an automated analyser (reference intervals 11.5-16.0 g/dL for women and 13.0-17.0 g/dL for men). Plasma glucose, serum insulin, and serum C-peptide were measured and analysed as previously described [14]. Serum triglyceride (TG) was analysed with an enzymatic colorimetric assay using an automated clinical chemistry analyser (reference interval 0.5-2.6 mmol/L). Estimates of insulin sensitivity in the fasting state were calculated using homeostasis model assessment (HOMA-IR) and the Quantitative Insulin-Sensitivity Check Index (QUICKI) [16,17], and insulin sensitivity including the 2-hour values for glucose and insulin with the insulin sensitivity index (ISI 0.120 ) according to the formula by Gutt [18].
OGTTs were performed from February 2008 until August 2010 as previously described [14]. All OGTTs were performed in the morning aer an overnight fast. To minimize time between OGTT and HbA 1c , the latter was measured simultaneously with the OGTT from September 2008 onwards. HbA 1c from the Tromsø Study 2007-2008 was used for the 932 participants who completed OGTT before September 2008. Mean change in HbA 1c for the 2,544 subjects who measured HbA 1c on both occasions was −0.03 ± 0.3%. For the purpose of this study, we chose to classify subjects with a single value of FPG ≥ 7.0 mmol/L, 2hPG ≥ 11.1 mmol/L, and/or HbA 1c ≥ 6.5% as having diabetes, even though subjects were asymptomatic. Subjects with diabetes were subdivided into diabetes detected by HbA 1c only, by OGTT (raised FPG and/or 2hPG) only and by both. Furthermore, subjects with diabetes detected by OGTT (regardless of HbA 1c ) were subdivided into diabetes detected by FPG (regardless of 2hPG) and by isolated raised 2hPG.

Statistics.
Normal distribution was evaluated by visual inspection of histograms and determination of skewness and kurtosis, and aer natural log transformation of TG, PAS, QUICKI, HOMA-IR, and ISI 0.120 , all variables except the PAS (where several subjects had "0" values) were considered normally distributed. Ln values were used when these variables were dependent variables. Pearson Chi-square test was used for subgroup analysis in Table 2. Comparisons between groups were performed with logistic regression for categorical variables and univariate analysis of variance with Bonferroni post hoc adjustment or Mann Whitney U test for continuous variables in Table 3. Venn diagrams were constructed to illustrate overlap between diagnostic criteria and scatterplots to illustrate the distribution of FPG and 2hPG values in relation to HbA 1c . Unless otherwise stated, data are expressed as mean ± SD for normally distributed values and as median (5, 95 percentile) for non-normally distributed values. All tests were two-sided, and value < 0.05 was considered statistically signi�cant. e Statistical Package for Social Sciences version 17.0 was used for all statistical analyses (SPSS Inc., Chicago, IL, USA).

Results
Among the 4,393 subjects who were invited, 3,520 attended and 3,476 completed the OGTT. e number of subjects planned to participate, invited to OGTT, and attended at different HbA 1c levels, as measured in the Tromsø Study 2007-2008, is presented in Table 1. In total, 294 (8.5%) subjects met one or more of the diagnostic criteria for diabetes. Mean age was 61 years and 49.5% were women.

�re�a�ence �� �iabetes �e�ne� b� �i�erent �ia�n�stic
Criteria. Among those who completed OGTT, 164 (4.7%) met the HbA 1c criterion, 119 (3.4%) met the FPG criterion, and 126 (3.6%) met the 2hPG criterion. In total 199 (5.7%) met the OGTT (FPG and/or 2hPG) criteria. As presented in Table 2, 95 (32.3%) of those with diabetes met the HbA 1c criterion only, 130 (44.2%) met the OGTT criteria only, and 69 (23.5%) met both criteria. e overlap between subjects with diabetes de�ned by HbA 1c and OGTT varied between 10-35% in different subgroups. HbA 1c alone detected more subjects with diabetes as compared to OGTT alone in those with BMI < 25 kg/m 2 , TG < 1.2 mmol/�, and high PAS, but there were no signi�cant differences in subgroup analysis of age and sex ( Table 2). Among those with diabetes detected by OGTT (regardless of HbA 1c ), isolated raised 2hPG was more common in subjects aged ≥ 60 years and women ( Table 2). is effect of age and sex was not due to differences in BMI. Strati�cation for age showed that the sex difference was signi�cant only in those aged ≥ 60 years, where 58% of women and 36% of men had isolated raised 2hPG ( < 1). Mean age and BMI did not differ signi�cantly between men and women. Furthermore, the sex difference was signi�cant only in the two lower BMI groups ( < ) and in the lowest PAS tertile ( < ). e distribution of subjects with diabetes detected by HbA 1c only, OGTT only, and both, as well as by OGTT components (FPG and isolated raised 2hPG) is illustrated strati�ed for age and sex in Figure 1. e overlap between subjects with diabetes de�ned by HbA 1c and OGTT was relatively consistent, but prevalence of isolated raised 2hPG was higher in subjects aged ≥ 60 years as compared to younger subjects, and in elderly women as compared to elderly men. In subjects aged ≥ 60 years the distribution of 2hPG values in relation to HbA 1c values was more scattered as compared to younger subjects (Figure 2), illustrating that for many subjects in this age group an HbA 1c value < 6.5% did not exclude a 2hPG value above the cut off point for diabetes. Table 3, subjects with diabetes detected by HbA 1c only had lower TG, lower systolic blood pressure, higher insulin sensitivity and were less insulin resistant and more physically active as compared to subjects with diabetes detected by OGTT only. Among subjects with diabetes detected by OGTT (regardless of HbA 1c ), those with raised FPG differed from those with isolated raised 2hPG by being younger, predominantly men and more insulin resistant (Table 3).

�.�. �re�a�ence o� Dia�etes De�ne� �� Di�erent Diagnostic
Criteria. In our population, we found prevalence of diabetes detected by OGTT only to be higher than prevalence of diabetes detected by HbA 1c only. e present study also con-�rmed results from recent studies showing that HbA 1c and OGTT de�ne different subjects with diabetes with relatively modest overlap, which in our study was only 23.5% [6,7,9]. Prevalence of diabetes de�ned by HbA 1c and OGTT, and overlap between these, differs in previous studies, probably due to differences in age, race, and sex composition of the populations and/or lack of standardisation of HbA 1c and glucose measurements [7,8,12,13,19].
Race, age, and sex have been reported to affect the outcome of diabetes testing with different diagnostic criteria [6-8, 12, 13]. Our study population did not allow us to study the effect of race as practically all subjects were Caucasian. When comparing subjects aged ≥ 60 years with younger subjects, we found no difference in prevalence of diabetes detected by HbA 1c only and OGTT only. Among those with diabetes detected by OGTT (regardless of HbA 1c ), prevalence of isolated raised 2hPG was higher in older (≥60 years) as compared to younger subjects. Furthermore, we found that among subjects aged ≥ 60 years, having a 2hPG in the diabetic range but a nondiabetic HbA 1c value was more common as compared to younger subjects. Similarly, in the Finnish population-based cross sectional FIN-D2D study including 2,826 men and women aged 45-74 years, any given HbA 1c value was found to imply a much higher 2hPG and slightly lower FPG in elderly as compared to middle aged subjects [13]. e 2hPG is known to increase more with age than FPG [20,21]. Possible explanations for the increased prevalence of isolated raised 2hPG among elderly subjects could be reduced basal insulin secretion [22], delayed insulin response aer oral glucose intake [21], physical inactivity, and/or weight gain [23].
In our data, there was no sex difference in diabetes detected by HbA 1c only and OGTT only. However, we found     that among those with diabetes detected by OGTT (regardless of HbA 1c ), isolated raised 2hPG was more common in elderly women as compared to elderly men, a difference that could not be explained by differences in age or BMI. Similarly, the FIN-D2D study reported that HbA 1c tends to miss more elderly diabetic people and especially women [13]. Previous studies have suggested that differences in FPG and HbA 1c levels are likely to re�ect sex-speci�c differences in glucose regulation as they, unlike differences in 2hPG, remained aer adjusting for height and body composition [24,25]. We also found that HbA 1c alone detected more subjects with diabetes as compared to OGTT alone in subjects with BMI < 25 kg/m 2 as compared to those with higher BMI. In a recently published paper, we reported that a particular HbA 1c value implied relatively higher 2hPG and FPG in subjects with high BMI compared to subjects with lower BMI [14].
As very few reports have addressed this issue, it remains uncertain whether BMI has an effect on diagnosis of diabetes by different criteria.

�.�. C�aracteristics of ����ects �it� Dia�etes De�ne� �� Dif� ferent Diagnostic
Criteria. In our population, subjects with diabetes detected by OGTT only had a worse cardiometabolic risk pro�le than those detected by HbA 1c only . Previous studies have shown con�icting results� some have found the worst risk pro�les in subjects with diabetes de�ned by OGTT [6,8,9], some in subjects with diabetes de�ned by HbA 1c [26], and some have found the two groups to have equally unfavourable risk pro�les [8,10]. In the international A1C-Derived Average Glucose study including 427 subjects with diabetes, HbA 1c , FPG, and 2hPG were all associated with CVD risk factors, but the strongest association was seen with HbA 1c [27]. We did not have data to evaluate the risk of diabetes complications in the different groups. Although both HbA 1c and 2hPG have been shown to be independent risk factors for cardiovascular morbidity and mortality, the added prognostic information may be marginal as compared to standard nonglycaemic risk factors [28][29][30].
In a prospective study based on the Norwegian populationbased longitudinal HUNT study, the risk of macrovascular complications in subjects with relatively low HbA 1c values was found to be mainly related to conventional risk factors [31]. e strength of our study is that OGTT was performed in a large number of subjects recruited from a population representative of the general population in our area. e main shortcomings of our study are that only subjects with HbA 1c in the range of 5.3-6.9% were invited to participate and that subjects included at an early stage of the study did not have HbA 1c measured simultaneously with the OGTT, but were included in the analysis with the HbA 1c value measured in the Tromsø Study 2007-2008. We chose to include these subjects in the analysis as we found that change in HbA 1c from the Tromsø Study to the OGTT visit was negligible for those who had HbA 1c measured at both occasions. Furthermore, in the absence of clear symptoms, diagnosis of diabetes requires raised values of HbA 1c , FPG, or 2hPG on two occasions. For practical reasons, we did not repeat either HbA 1c , or the OGTTs, but chose to classify subjects with a single raised value of HbA 1c , FPG, or 2hPG as having diabetes. As FPG, and especially 2hPG, are known to have high within-person variation, repeating the OGTTs to con�rm the diagnosis would probably have reduced the number of subjects with diabetes detected by OGTT [32]. HbA 1c is known to be affected by anaemia. Hb was measured in the Tromsø Study 2007-2008, but not simultaneously as OGTT. However, anaemia is not a source of error when analysing HbA 1c with the HPLC method used in our study as the analysis is not performed if there are too few or too many erythrocytes in the sample. Haemolytic anaemia could result in falsely low HbA 1c , but the condition is rare in our population and is not likely to affect the results. Other shortcomings are that we did not have information about retinopathy or other end organ diseases, and that we did not differentiate between type 1 and type 2 diabetes. However, as subjects in our study did not have previously diagnosed diabetes and age ranged from 30-87 years, most diabetes cases were likely to be type 2 diabetes. e cross-sectional study design is a major limitation when evaluating the impact of using different diagnostic criteria for diabetes. Prospective studies are needed to clarify which test detects the population with the highest risk of disease progression and complications of diabetes.

Conclusions
e current HbA 1c and glucose criteria for diabetes de�ned different subjects with only modest overlap. Among those with diabetes detected by OGTT (regardless of HbA 1c ), isolated raised 2-hour plasma glucose was more common in subjects aged ≥ 60 years as compared to younger subjects, and in elderly women as compared to elderly men. As race, age, sex, and possibly BMI seem to affect HbA 1c , FPG, and 2hPG and the relationship between these, creating an algorithm for choice of diagnostic test in different subgroups is a possibility and may be bene�cial. If the aim is to detect as many patients with diabetes as possible, our data suggest that OGTT would be preferable for those aged ≥ 60 years, and especially women, while HbA 1c would be preferable for the younger and those with low BMI. However, in order to decide which diagnostic test should be preferred, and whether race, age, sex, and/or BMI speci�c guidelines should be considered, prospective studies with micro-and macrovascular endpoints are needed.