Skip navigation

Diagnosing diabetes

Meta-analysis with single patient data is rare, but meta-analysis of a diagnostic test with single patient data is probably unique. It is even more remarkable for diabetes, where in the USA at least half the diabetics are undiagnosed and where only 20% of diagnosed diabetics are diagnosed according to the acknowledged World Health Organisation criteria of fasting plasma glucose and oral glucose tolerance test on at least two occasions. Some brave hearts from Los Angeles have performed a superb analysis to determine that a single haemoglobin A 1c test (HbA 1c ) can diagnose diabetes with confidence. Bandolier can only bring you a flavour of the enormous amount of work that went into this analysis.

Glycosylated haemoglobin

Glycosylated haemoglobins are normal haemoglobin to which a glucose molecule becomes added in a non-enzymatic manner. The percentage haemoglobin that is glycated is directly proportional to the time that red blood cells have been exposed to glucose, and to glucose concentrations. Measurement of the glycated haemoglobin fraction gives an integrated picture of the average blood glucose concentration during the half life of the cells - that is over the last 60 days. HbA 1c is usually given as a percentage of the total haemoglobin.

Data search

The authors searched for papers which had data on both an oral glucose tolerance test and glycosylated haemoglobin. They then contacted authors to try to obtain individual patient data. Out of 34 possible studies, 31 investigators responded, and 18 were able to provide data. Fasting plasma glucose concentrations, two-hour post dextrose glucose concentrations and glycosylated haemoglobin levels were available from 11,276 individuals (83% of all subjects in the literature). Analyses were restricted to 10 studies with HbA 1c information on 8,255 individuals.

WHO criteria

The WHO criteria for interpretation of glucose tolerance tests after 75 g oral dextrose is:

Normal: Fasting plasma glucose <6.4 mmol/L and
2 hour plasma glucose <7.8 mmol/L

Diabetes: Fasting plasma glucose >=7.8 mmol/L or
2 hour plasma glucose >=11.1 mmol/L

Impaired glucose tolerance (IGT): any other condition

HbA 1c

HBA 1c levels in individuals with normal glucose tolerance tests was a mean of 5.3% with a standard deviation of 0.5%. Using the usual practice of describing a normal range as +/- 2 standard deviations, the normal range was 4.3% to 6.3%.

However, with some beautiful but complicated modelling the authors were able to define a level of 7.0% as a cutoff for diagnosing diabetes. In their model they identified subpopulations representing normal subjects and diabetics. The 7.0% cutoff gave a sensitivity of 99.6% - that is a HbA 1c of 7.0% correctly identified 996 out of 1000 patients who had diabetes in their model. The same cutoff had a specificity of 99.9% - that is a HbA 1c of 7.0% correctly identified 999 out of 1000 patients who were normal in their model.

When applied to the actual oral glucose tolerance results, of those subjects with an HbA 1c level of at least 7.0%, 89% had diabetes, 7% had impaired glucose tolerance, and 4% were normal.


The authors give good arguments why HbA 1c levels should be used for diagnosing diabetes, including the difficulty of performing glucose tolerance tests reliably, the lack of reproducibility of the oral glucose tolerance test, and the clinical practice of using HbA 1c as a surrogate test for glucose lowering, with the aim of keeping the HbA 1c levels of diabetic patients below 7.0%. Moreover, an HbA 1c level of below 7.0% would generally be treated by diet and exercise, regardless of the diagnosis of impaired glucose tolerance or diabetes defined by an oral glucose tolerance. Perhaps we will learn soon that diabetes can be diagnosed with a fasting plasma glucose plus a single HbA 1c level.


  1. 1 AL Peters, MB Davidson, DL Schriger, V Hasselblad. A clinical approach for the diagnosis of diabetes mellitus: an analysis using glycosylated hemoglobin levels. Journal of the American Medical Association 1996 276: 1246-52

previous or next story in this issue