Value of hemoglobin A1c in diagnosing diabetes mellitus within a chronic disease management system illustrated by the receiver operating characteristic curve.

OBJECTIVE: To develop a receiver operating characteristic (ROC) curve of glycosylated hemoglobin (HbA1c) for diagnosing diabetes mellitus within a chronic disease management system.
METHODS: A case-control study including medical records from January 1, 1997, to December 31, 2005, was conducted at the Sioux Falls Veterans Affairs Medical Center. Medical records for the case group (patients with diabetes) were selected based on 1 of 3 criteria: International Classification of Diseases, Ninth Revision, Clinical Modification or Current Procedural Terminology codes specific for type 1 and type 2 diabetes; patients' use of medications (oral hypoglycemic agents, antidiabetes agents, or insulin); or results from random blood or plasma glucose tests (at least 2 measurements of blood glucose > or = 200 mg/dL). Records for the control group were selected based on patients having HbA1c measured, but not meeting the above diagnostic criteria for diabetes during the study period. Records for cases and controls were randomly frequency-matched, one-to-one. The control group was randomly divided into 5 sets of an equal number of records. Five sets of an equal number of cases were then randomly selected from the total number of cases. Each test data set included 1 case group and 1 control group, resulting in 5 independent data sets.
RESULTS: In total, 5040 patient records met the case definition in the diabetes registry. Records of 15 patients who were prescribed metformin only, but did not meet any other case criteria, were reviewed and excluded after determining the patients were not diabetic. The control group consisted of 5 sets of 616 records each (totaling 3080 records), and the case group consisted of 5 sets of 616 records each (totaling 3080 records). Thus, each of the 5 independent data sets of 1 case group and 1 control group contained 1232 records. The case group was predominantly composed of white men (mean age, 69 years; mean body mass index, 31 kg/m2). Demographic data were similar for control patients. The ROC curve revealed that a HbA1c > or = 6.3% (mean + 1 SD) offered the most accurate cutoff value for diagnosing type 2 diabetes mellitus, with the following statistical values: C statistic, 0.78; sensitivity, 70%; specificity, 85%; and positive likelihood ratio, 4.6 (95% confidence interval, 4.2-5.0).
CONCLUSION: A HbA1c value > or = 6.3% may be a useful benchmark for diagnosing diabetes mellitus within a chronic disease management system and may be a useful tool for monitoring high-risk populations.