Diagnostic Strategies to Detect Glucose Intolerance in a Multiethnic Population

Between December 1996 and October 1998, individuals of South Asian, Chinese, and European origin were randomly recruited from three cities in Canada to undergo a cardiovascular health assessment (19). All participants were required to have fasted for at least 8 h. All nondiabetic participants had fasting glucose and HbA_1c measurements taken, and the glucose measurement was repeated 2 h after ingestion of 75 g oral glucose. Immediately after being drawn, all samples were placed on ice and centrifuged and aliquots prepared within 1 h of collection. All serum and plasma aliquots were transferred on dry ice to the core laboratory in Hamilton, Canada, for central core laboratory analysis using standard methodology. Aliquots were frozen at −70°C, and all analyses were conducted within 3 days after arrival at the core lab. Blood for glucose measurement was gathered in serum separator tubes and measured using enzymatic methods with a hexokinase reference. The stated precision was a coefficient of variation (CV) <1% and observed precision was a CV <3.4%. HbA_1c was collected in EDTA tubes and analyzed using a 765 Glycomat machine, which uses low-pressure cation exchange chromatography in conjunction with gradient elution to separate human hemoglobin subtypes (20,21). This method is associated with a CV <4%, and at the time of study set-up, this method was not certified by the NGSP. The 1998 WHO diagnostic criteria was used as the “gold standard” and participants were classified as 1) normal: fasting glucose <7.0 mmol/l and a 2-h glucose <7.8 mmol/l; 2) IGT: fasting glucose <7.0 mmol/l and a 2-h glucose (post-75 g glucose load) 7.8–11.0 mmol/l; or 3) diabetic: a fasting glucose ≥7.0 or a 2-h (post-75 g glucose load) glucose ≥11.1 mmol/l. The 1997 ADA criteria were also applied to classify individuals as 1) normal: fasting glucose <6.1 mmol/l; 2) impaired fasting glucose: fasting glucose 6.1–6.9 mmol/l; or 3) diabetic: fasting glucose ≥7.0 mmol/l.

To determine the optimal fasting plasma glucose value for the diagnosis of diabetes, we calculated receiver operator characteristic (ROC) curves by plotting the sensitivity (true positive rate) of a range of fasting plasma glucose values versus the false positive rate (1 specificity). The ROC curve is a graphic means of assessing the ability of a screening test to discriminate between healthy and diseased people and allows precise determination of an optimal cut-point, which maximizes the true-positive rate and minimizes the false-positive rate (22). For each cut-point, positive and negative likelihood ratios were calculated. The positive likelihood is the ratio of true-positive rate to false-positive rate, and the negative likelihood ratio is the ratio of true-negative rate to false-negative rate. Unlike the positive and negative predictive values, likelihood ratios are not altered by changes in the prevalence of the target disorder, in this case diabetes, which occurs when the screening test is applied to heterogeneous populations (e.g., geographically or ethnically distinct groups). Likelihood ratios can also be used together with pretest probability estimates to generate the post-test probability of disease, thereby indicating by how much a given diagnostic test result can raise or lower the pretest probability of a disease. Likelihood ratios >1 increase the probability that the target disorder is present, and likelihood ratios <1 decrease the probability that the target disorder is present (23).

Of the 985 participants who were recruited, 37 (3.8%) had established diabetes at entry and were excluded, and of the remaining 948 people, 936 had both fasting and 2-h plasma glucose samples measured. Of these, 315 were South Asian, 307 were Chinese, and 314 were European. Using the ADA criteria, 91.7% of subjects were classified as normal, 5.2% had impaired fasting glucose, and 3.1% had diabetes. Using the WHO criteria, only 78.4% were classified as normal, 15.2% had IGT, and 6.4% had diabetes (Table 1).

A substantial proportion (14.5%) of the individuals who were classified as “normal” by the ADA criteria had IGT, and 2.1% had diabetes. Conversely, of those classified as normal by the WHO criteria, only 2.5% had impaired fasting glucose. The sensitivity of the ADA criteria to diagnose diabetes using the WHO criteria as the gold standard was 48.3% (95% CI 35.7–61.0) overall and was 31.6% (10.7–52.5) among Europeans, 28.6% (4.9–52.2) among Chinese, and 70.4% (53.1–87.6) among South Asians.

By constructing an ROC curve, the optimum cut-point using fasting plasma glucose to diagnose diabetes was 5.7 mmol/l. This was associated with a sensitivity and specificity of 83.3 and 88.0%, respectively, a positive LR of 7.0 (95% CI: 5.5–8.4), and a negative LR of 0.2 (0.1–0.3) (Table 2, Fig. 1). The diagnostic ability varied substantially between ethnic groups, although using a 5.7 mmol/l cut-point significantly improved the sensitivity and slightly lowered the specificity of the fasting glucose to diagnose diabetes (Table 2). Using an ROC curve, the optimal cut-off for HbA_1c of 5.9% was associated with a specificity and sensitivity of 75.0 and 79.1%, respectively, for diabetes overall, with substantial variation between ethnic groups. (Table 2, Fig. 2).

Using ROC curve analysis, the cut-points for fasting glucose or HbA_1c to diagnose diabetes in each ethnic group were determined, and substantial variation was observed (Table 3)

To improve the diagnostic precision for diabetes, a combination of the optimal fasting glucose and HbA_1c values was identified. The paired dual optimal cut-point for fasting glucose and HbA_1c was obtained by multiple ROC analysis (24). The combined optimal cut-off occurs when a subject simultaneously has a fasting glucose ≥5.7 mmol/l and an HbA_1c ≥5.9%, identical to the pairing of the single optimal cut-offs obtained for each marker independently. The highest probability of diabetes was found among people who had a fasting glucose ≥5.7 mmol/l and an HbA_1c ≥5.9%. This was associated with a sensitivity of 71.7%, a specificity of 95.0% overall, a positive LR of 14.3 (95% CI 9.6–19.0), and a negative LR of 0.3 (0.2–0.4). Using this cut-point, the sensitivity of this method was highest among South Asians and lowest among European-origin Canadians (Fig. 3, Table 4).

A similar procedure was used to identify the optimal cut-point for the determination of IGT using both fasting glucose and HbA_1c independently and in combination. However, cut-points were not easy to identify, and a fasting plasma glucose cut-point ≥5.3 mmol/l was associated with a false-positive rate of 32.4% (95% CI 29.1–35.7%) (Fig. 4). A nearly linear ROC curve for IGT was observed. This indicates that for every increase in test sensitivity there is a nearly equal increase in false-positive rate (or a proportional decrease in specificity). For HbA_1c, a cut-point ≥5.6% was also associated with a high false-positive rate of 42% (38.5–45.4%) (Fig. 5).

The combination of fasting plasma glucose and HbA_1c did not improve the precision of classifying people with IGT (Fig. 6).

Applying the dual cut-point of fasting glucose ≥5.7 mmol/l and HbA_1c ≥5.9% in our study cohort, 9.8% (95% CI 8.7–9.9) of people would be classified as diabetic and would not require an oral glucose tolerance test (OGTT). HbA_1c was found to correlate with stages of glucose tolerance as defined by the WHO. Subjects classified as normal had a mean HbA_1c of 5.4%, as IGT had a mean HbA_1c of 5.7%, and as type 2 diabetic had a mean HbA_1c of 7.0%, and these differences were significant at P < 0.001 (Table 5).

The ADA cut-point of a fasting plasma glucose ≥7.0 mmol/l to diagnose diabetes is not sensitive and results in an unacceptably high number of false negatives. Therefore, relying on a fasting glucose alone to exclude diabetes will falsely reassure approximately one-fifth of the general population that they do not have diabetes or IGT, when they in fact suffer an important abnormality in their glucose metabolism. This is consistent with the observations of other comparisons of the ADA to WHO criteria. For example, the 10 most recent studies we identified in which the ADA criteria were compared with other criteria, the sensitivity ranged from as low as 31% in a an obese French population to as high as 79% in Canarian Caucasians (13,14,25–32) (Table 6). We identified two cut-off values (fasting plasma glucose ≥5.7 mmol/l and an HbA_1c ≥5.9%) that minimize the number of false positives and false negatives. The corresponding likelihood ratios for these cut-points may be used together with the patient’s pretest probability of diabetes to generate a post-test probability of diabetes. The advantage of using likelihood ratios is their stability in the face of changing disease prevalence, which is important to consider given the tremendous variation in the prevalence of diabetes observed among people of various ethnic origin. Fasting glucose and HbA_1c measurements in combination and the accompanying likelihood ratios improve the diagnostic classification of people as being at low risk for having diabetes versus moderate to high risk for having diabetes and therefore allow for the selective use of the OGTT. Unfortunately, neither fasting glucose nor HbA_1c alone or in combination can reliably classify which people have IGT, and if classification of this state is desired for clinical or epidemiological purposes, an OGTT test remains necessary.

When comparing ethnic groups, the sensitivity of the ADA criteria versus the WHO standard to diagnose type 2 diabetes was highest among people of South Asian origin and lowest among people of European origin. The difference in the ROC curve cut-point between South Asians and Europeans is especially large for HbA_1c (6.2 vs. 5.6%, Table 3), which is consistent with the large body of evidence that demonstrates that people of South Asian origin have a greater burden of glucose intolerance than Europeans (33,34). Given the differences in the sensitivity and specificity (and hence positive and negative predictive values) of the fasting glucose and HbA_1c to diagnose diabetes between ethnic groups, using likelihood ratios can help clinicians more reliably estimate the post-test probability of diabetes. The pretest estimation of diabetes should be influenced by knowledge that diabetes is more prevalent in certain ethnic groups. In our study population the odds ratios (ORs) for diabetes included South Asian ethnicity OR 1.87 (95% CI 1.3–2.6), age per year 1.05 (1.0–1.1), BMI ≥30 2.75 (1.73–4.35), abdominal obesity (WHR ≥0.85) 3.17 (1.9–5.2), HDL cholesterol (<0.9 mmol/l) 2.58 (1.54–4.32), and elevated triglycerides (>2.26 mmol/l) 4.15 (3.0–5.8) (34).

Based on our results and the results of other investigators (13,14), using a fasting glucose measurement alone inconsistently identifies people who have an abnormal 2-h glucose concentration after an oral glucose challenge. On the other hand, HbA_1c, which can be taken in the non-fasting state, as we have shown, correlates with the stages of glucose tolerance as defined by WHO criteria. Therefore, the HbA_1c is a useful and potentially more feasible method of classifying individuals with glucose intolerance than the OGTT in epidemiological studies.

The principal message from this report is that the potential gain in specificity using only a fasting glucose measurement to diagnose diabetes, while eliminating false positives, is associated with a large number of “false-negative” results. Furthermore, our study population included a large representation of people of Chinese and South Asian origin, people who have repeatedly been observed to develop glucose intolerance upon adoption of Western lifestyles (35,36). Therefore, the application of a uniform set of fasting glucose cut-points is associated with a large variation in the confidence with which we can exclude the presence of diabetes in these groups. Therefore, using a lower fasting plasma glucose cut-off together with the HbA_1c decreases the chance of falsely excluding people who have diabetes and leads to a more selective use of the OGTT, a test that is associated with unnecessary cost and inconvenience for patients.

In conclusion, the sensitivity of the ADA criteria to diagnose diabetes is low, and there is substantial variation between ethnic groups. However, use of fasting glucose and HbA_1c together, compared with using fasting glucose alone, reduces the number of people who are falsely classified as having normal glucose tolerance. To identify people with IGT, the OGTT remains necessary.