Assessing multiple-group diagnostic problems with multi-dimensional receiver operating characteristic surfaces: application to proton MR Spectroscopy (MRS) in HIV-related neurological injury.

TitleAssessing multiple-group diagnostic problems with multi-dimensional receiver operating characteristic surfaces: application to proton MR Spectroscopy (MRS) in HIV-related neurological injury.
Publication TypeJournal Article
Year of Publication2008
AuthorsYiannoutsos CT, Nakas CT, Navia BA
Corporate Authorsproton MRS Consortium
JournalNeuroimage
Volume40
Issue1
Pagination248-55
Date Published2008 Mar 1
ISSN1053-8119
KeywordsAdult, Aging, AIDS Dementia Complex, Algorithms, Aspartic Acid, Basal Ganglia, Brain, Brain Chemistry, Choline, Creatinine, Female, HIV Infections, Humans, Image Processing, Computer-Assisted, Inositol, Magnetic Resonance Imaging, Male, Middle Aged, Predictive Value of Tests, ROC Curve
Abstract

We present the multi-dimensional Receiver Operating Characteristic (ROC) surface, a plot of the true classification rates of tests based on levels of biological markers, for multi-group discrimination, as an extension of the ROC curve, commonly used in two-group diagnostic testing. The volume under this surface (VUS) is a global accuracy measure of a test to classify subjects in multiple groups and useful to detect trends in marker measurements. We used three-dimensional ROC surfaces, and associated VUS, to discriminate between HIV-negative (NEG), HIV-positive neurologically asymptomatic (NAS) subjects and patients with AIDS demential complex (ADC), using brain metabolites measured by proton MRS. These were ratios of markers of inflammation, Choline (Cho) and myoinositol (MI), and brain injury, N-acetyl aspartate (NAA), divided by Creatine (Cr), measured in the basal ganglia and the frontal white matter. Statistically significant trends were observed in the three groups with respect to MI/Cr (VUS=0.43; 95% confidence interval (CI) 0.33-0.53), Cho/Cr (0.36; 0.27-0.45) in the basal ganglia and NAA/Cr in the frontal white matter (FWM) (0.29; 0.20-0.38), suggesting a continuum of injury during the neurologically asymptomatic stage of HIV infection, particularly with respect to brain inflammation. Adjusting for age increased the combined classification accuracy of age and NAA/Cr (p=0.053). Pairwise comparisons suggested that neuronal damage associated with NAA/Cr decreases was mainly observed in individuals with ADC, raising issues of synergism between HIV infection and age and possible acceleration of neurological deterioration in an aging HIV-positive population. The three-dimensional ROC surface and its associated VUS are useful for assessing marker accuracy, detecting data trends and offering insight in disease processes affecting multiple groups.

DOI10.1016/j.neuroimage.2007.09.056
Alternate JournalNeuroimage
PubMed ID18191586
PubMed Central IDPMC3774155
Grant ListNS 36524 / NS / NINDS NIH HHS / United States
R01 NS036524-03 / NS / NINDS NIH HHS / United States
R01 NS036524-04 / NS / NINDS NIH HHS / United States
R01 NS036524-05 / NS / NINDS NIH HHS / United States
R01 NS036524-07 / NS / NINDS NIH HHS / United States
R01 NS036524-08 / NS / NINDS NIH HHS / United States