Receiver operating characteristic analysis of glycogenated nuclei in liver biopsy specimens: Quantitative evaluation of their relationship with diabetes and obesity

Glycogenated hepatocyte nuclei are a common finding in liver biopsy specimens from patients presenting with a variety of clinical disorders. Most commonly, glycogenated nuclei form part of a spectrum of pathological changes in the diabetic or obese patient. However, no previous investigation has assessed glycogenated hepatocyte nuclei in a quantitative manner. In this study we used receiver operating characteristic (ROC) analysis to assess quantitatively the strength of glycogenated nuclei as a marker for diabetes and/or obesity. The study population consisted of 102 liver biopsy specimens from the adult population: 20 from diabetic patients, 41 from obese patients, 10 from both obese and diabetic patients, and 31 from neither diabetic nor obese patients. The mean age was 48.5 years with a range of 18 to 80 years. We evaluated the extent of glycogenated nuclei by averaging the number present per high power field over a minimum of 10 high power fields, representing all zones of the liver. The extent of steatosis was graded on a numerical scale from 1 to 10. In ROC analysis perfect tests are associated with an area of 1.0 and completely random tests with an area of 0.5. Receiver operating characteristic analysis showed that glycogenated nuclei are a relatively good test for distinguishing diabetic from nondiabetic patients (area, 0.75), a poor test for distinguishing obese from nonobese patients (area, 0.60), and a fair test for either condition (area, 0.67). In addition, we observed that glycogenated nuclei were preferentially distributed in the periportal "zone 1" of the liver, showed no correlation with steatosis, and had no relation to patient age. For our patient study population the prevalence of diabetes mellitus was 12%. At a cutoff level of four glycogenated nuclei per 400× high power field, the specificity and sensitivity of glycogenated nuclei as a test for diabetes are 0.98 and 0.30, respectively, with a positive predictive value of 66%.