The logistic regression and ROC analysis of group-based screening for predicting diabetes incidence in four years

Kobe J Med Sci. 2006;52(6):171-80.

Abstract

In diabetes screening with hemoglobin A1c in lieu of plasma glucose, the optimum cut-off point for predicting the incidence of diabetes mellitus in the four-year period was examined. In addition, considerations were given on items in the screening and questionnaire aside from hemoglobin A1c, which would be useful in predicting diabetes aside from hemoglobin A1c. The optimum cut-off point of hemoglobin A1c to predict diabetes, based on receiver operating characteristic curve, was 5.3 percent (sensitivity, 84.2%; specificity, 92.1%). Based on the logistic regression analysis, useful items (other than hemoglobin A1c) were alanine aminotransferase and gamma-glutamyl transpeptidase. A combined application of hemoglobin A1c with alanine aminotransferase and gamma-glutamyl transpeptidase for predicting the incidence of diabetes in the four-year period resulted in the sensitivity of 86.8% and the specificity of 96.3%. When the combined application was compared with the sole use of hemoglobin A1c at 5.3%, the combined use was superior to the latter in terms of both sensitivity and specificity, resulting in the reduction of false positives by more than 50%.

MeSH terms

  • Adult
  • Alanine Transaminase / blood
  • Cholesterol, HDL / blood
  • Cohort Studies
  • Diabetes Mellitus / epidemiology*
  • Female
  • Glycated Hemoglobin A
  • Hemoglobins / analysis
  • Humans
  • Incidence
  • Logistic Models
  • Male
  • Middle Aged
  • ROC Curve*
  • Retrospective Studies
  • Sensitivity and Specificity
  • gamma-Glutamyltransferase / blood

Substances

  • Cholesterol, HDL
  • Glycated Hemoglobin A
  • Hemoglobins
  • hemoglobin A1c protein, human
  • gamma-Glutamyltransferase
  • Alanine Transaminase