Endocrine Abstracts (2009) 19 OC14

Urinary steroid profiling as a biomarker tool for the detection of adrenal malignancy: results of the EURINE ACC Study

W Arlt1, S Hahner2, R Libe3, BA Hughes1, M Biehl4, H Stiekema4, P Schneider4, DJ Smith1, CHL Shackleton1, G Opocher5, J Bertherat3, B Allolio2, M Mannelli6, F Mantero5, M Fassnacht2, X Bertagna3 & PM Stewart1


1School of Clinical & Experimental Medicine, University of Birmingham, Birmingham, UK; 2Endocrine & Diabetes Unit, Department of Medicine 1, University of Wurzburg, Wurzburg, Germany; 3Department of Endocrinology, Faculty of Medicine, Institut Cochin, Rene Descartes University, Paris, France; 4Institute of Mathematics and Computing Science, University of Groningen, Groningen, The Netherlands; 5Division of Endocrinology, Department of Medical and Surgical Sciences, University of Padua, Padua, Italy; 6Department of Clinical Physiopathology, Division of Endocrinology, University of Florence, Florence, Italy.


Adrenal tumours have an incidence of 2–3% in the general population. Adrenocortical carcinoma (ACC), a highly malignant tumor with a poor prognosis, has an annual incidence of two per million but representation in pre-selected patient cohorts with adrenal masses undergoing surgery is up to 12%. Differentiating adrenocortical adenomas (ACA) from ACC represents a continuous challenge, with unfavourable sensitivities and specificities provided by tumour size, imaging criteria and even histology. In this study, we explored the value of steroid profiling by gas chromatography/mass spectrometry (GC/MS) in detecting adrenal malignancy. We collected 24-h urine samples from 287 adrenal tumour patients. For final analysis only samples from treatment-naïve patients with radiological evidence of tumour at time of urine collection were included (n=118; 83 ACA, 35 ACC); underlying diagnosis was ascertained by clinical assessment including follow-up and, where available, by histology. These data and those from 100 healthy controls were subjected to distance-based machine learning techniques, employing a prototype-based relevance learning approach to identify most discriminative steroids and to classify ACA versus ACC versus controls. ACC revealed a distinct steroid pattern with increased pregnenolone, 17OH-pregnenolone, progesterone and androgen metabolites in 86% while only 5 out of 83 ACA patients showed increases in any of those metabolites. Matrix relevance learning demonstrated androsterone and etiocholanolone as most differentiating between ACC and ACA while androsterone and DHEA were best to differentiate between ACA and controls. Preliminary results obtained from the machine learning analysis identified distinct benign and malignant steroid prototypes yielding an overall accuracy of 85% (ACA 96%, ACC 80%). In conclusion, urinary steroid profiling by GC/MS is a highly promising biomarker tool for differentiating benign from malignant adrenal tumours. Prospective studies will determine its predictive value in ACC and, importantly, in the differential diagnosis of adrenal incidentalomas.