Current issues of ACP Journal Club are published in Annals of Internal Medicine


Meta-analysis: Glycosylated hemoglobin levels are useful for diagnosing diabetes

ACP J Club. 1997 Mar-Apr;126:46. doi:10.7326/ACPJC-1997-126-2-046

Source Citation

Peters AL, Davidson MB, Schriger DL, Hasselblad V, for the Meta-analysis Research Group on the Diagnosis of Diabetes Using Glycated Hemoglobin Levels. A clinical approach for the diagnosis of diabetes mellitus. An analysis using glycosylated hemoglobin levels. JAMA. 1996 Oct 16;276:1246-52.



To determine the usefulness of glycosylated hemoglobin (HbAlc) levels for diagnosing diabetes mellitus compared with oral glucose tolerance test (OGTT) results.

Data sources

Studies were identified in MEDLINE (1966 to June 1994), bibliographies of relevant studies, and personal reprint files.

Study selection

Studies were selected if HbAlc levels were measured and OGTTs were done in the same patient. Studies were excluded if they included patients who had conditions that would alter glucose tolerance or if authors did not supply the relevant data.

Data extraction

Glycosylation measurement method, glucose load, population type (patients with a previous positive screening test, self-referrals from the general population, referrals from high-risk populations, and patient populations purposefully enriched with persons known to have diabetes), and the prevalence of diabetes mellitus. National Diabetes Data Group criteria were used to define normal glucose tolerance, and World Health Organization (WHO) criteria were used to define impaired glucose tolerance and diabetes. After being transformed to allow for differences in blood samples (whole blood or plasma, or venous or capillary), type of glucose measurement, and glucose load, patient data were pooled across studies.

Main results

18 studies (11 276 patients) met the criteria. Through use of the OGTT results as a diagnostic standard, the sensitivity, specificity, and positive predictive value were calculated for HbAlc levels that were 2, 3, and 4 standard deviations (SDs) above the mean. At 4 SDs, the sensitivity, specificity, and positive predictive value were 36%, 100%, and 97%, respectively. {The likelihood ratio of a positive test (LR+) approached infinity, and the likelihood ratio of a negative test (LR-) was 0.64.}* Because of lack of agreement between HbAlc levels and OGTT results, a model using 3 subpopulations was developed (subpopulation 1, persons without diabetes; subpopulation 2, undetermined; and subpopulation 3, persons with diabetes). Sensitivity was defined as the proportion of subpopulation 3 with an HbA1c value above various HbA1c cutpoints, and specificity was defined as the proportion of subpopulation 1 with an HbAlc value below these cutpoints. When an HbAlc value of 7% was used as the cutpoint (a value above which pharmacologic intervention may be indicated), the sensitivity was 99.6%, the specificity was 99.9%, and 30% {LR+ 996 and LR- 0.004}* of subpopulation 2 were determined to have diabetes. Applying the HbA1c cutpoint of ≥ 7% to the OGTT results, 89% of patients had diabetes, 7% had impaired glucose tolerance, and 4% had normal glucose tolerance, according to the modified WHO criteria.


The glycosylated hemoglobin level is useful for identifying patients with diabetes who require pharmacologic intervention.

Source of funding: In part, American Diabetes Association.

For article reprint: Not available.

*Numbers calculated from data in article.


Severe diabetes mellitus with symptomatic hyperglycemia is readily diagnosed. Distinguishing milder forms from the normal range of plasma glucose is more challenging. Current criteria rely on the OGTT, which is inconvenient and has poorly reproducible results. This meta-analysis by Peters and colleagues assesses the diagnostic characteristics of HbA1c and shows that an elevated level (> 7%) is highly specific for diabetes. As a result, this test seldom yields false-positive results and minimizes the problems of incorrect labeling of patients and inappropriate therapy. The test is not very sensitive—only about 40% of cases of diabetes diagnosed by the OGTT alone have HbA1c values > 7%. Is this low sensitivity for the mildest forms of diabetes a critical drawback to HbA1c as a diagnostic test?

Current standards specify treatment to lower HbA1c levels to < 7%, which delays or prevents development of chronic diabetes complications (1). Failure to diagnose the disease when the basal value falls below this level does not deprive patients of beneficial pharmacologic therapy to lower their glucose levels because diet and exercise would be used as long as HbA1c levels remain < 7%. The other argument that is sometimes made for diagnosing the mildest forms of diabetes is that it encourages treatment of other risk factors for coronary disease. However, smoking cessation and treatment of hypertension and obesity are indicated regardless of the presence of diabetes. Only the decision to treat hypercholesterolemia in an otherwise low-risk patient might be influenced by the presence of diabetes; and in practice, few such patients are evaluated with an OGTT. HbA1c is, therefore, quite sensitive for detecting diabetes in which diagnosis has clear implications for pharmacologic therapy.

Because HbA1c is the most convenient diagnostic test for diabetes, the demonstration of acceptable test characteristics affirms its already widespread use for this purpose. Further studies will help to define the optimal diagnostic cutpoint.

William E. Clutter, MD
Washington University School of MedicineSt. Louis, Missouri, USA


1. Standards of medical care for patients with diabetes mellitus. American Diabetes Association. Diabetes Care. 1994;17:616-23.