Abstract
413
Objectives: The value of FDG PET to predict the conversion from mild cognitive impairment (MCI) to Alzheimer’s dementia (AD) is currently under debate. Principal component analysis (PCA) of FDG PET data was proposed as an alternative approach for diagnosing neurodegenerative disorders (in particular, parkinsonism; e.g. Eckert et al. 2007, Tang et al. 2010, Niethammer et al. 2014). In this study, we used PCA to identify a metabolic covariance pattern that predicts the conversion from MCI to AD (so-called AD conversion-related pattern, ADCRP), and investigated the prognostic value of the resulting pattern expression score (PES).
Methods: FDG PET scans of 544 MCI patients obtained from the ADNI database (median follow-up based on the reversed Kaplan-Meier method: 48 months [95 % CI: 47 - 48 months]) were studied. We implemented voxel-based PCA (Spetsieris, 2015) and standard SPM (as reference) analyses to disclose cerebral metabolic patterns associated with conversion from MCI to AD in a training dataset (n=272). By Cox proportional hazards regression we examined the prognostic value of several candidate predictors (see below). Also, we constructed prognostic models with the identified significant predictors adjusted for age and sex with: i) clinical, ii) imaging, and iii) clinical and imaging variables in combination. For model validation, the results of each Cox regression model were independently applied to the test dataset (n = 272) by calculation of the prognostic index (PI, Royston, 2013) for each individual subject. The PI was used to stratify the dataset according to conversion risk (i.e., equally-sized groups of low, medium and high risk)
Results: In the training dataset, PCA revealed an ADCRP that involved several regions with relative hypometabolism (temporoparietal, frontal, posterior cingulate/precuneus) and regions with relative hypermetabolism (sensorimotor cortex, cerebellum). Among the predictor variables age, sex, MMSE score, APOE4 status, PES (ADCRP) and normalized FDG uptake (in significant clusters yielded by SPM), PES was the best independent predictor (Hazard Ratio = 2.29, 95 % C.I.: 1.70 - 3.09, p<0.001). Moreover, adding PES to the model including the clinical biomarkers significantly (p<0.001) increased its prognostic value (clinical variables alone: Harrell’s C = 0.72; combined model: Harrell’s C = 0.80). In the test dataset, the models showed similar results with the best prediction power given by the combined model (Harrell’s C = 0.79). Based on derived PI values, the best separation between conversion risk strata was achieved by combination of PES and clinical variables: low vs. medium risk: p<10-6, medium vs. high risk: p<10-3, low vs. high risk: p<10-13.
Conclusions: The PES of the ADCRP is a valid and observer-independent predictor of conversion from MCI to AD. Combining clinical variables and PES yielded a higher accuracy than each single tool in predicting conversion of MCI to AD, emphasizing the incremental utility of FDG PET. References: Eckert T, Van Laere K, Tang C, et al. Quantification of PD-related network expression with ECD SPECT. Eur J Nucl Med Mol Imaging. 2007 Apr; 34(4):496-501. Epub 2006 Nov 10. Tang C.C., Poston K.L. MD[asterisk], Eckert T. MD et al. Differential diagnosis of parkinsonism: a metabolic imaging study using pattern analysis, Lancet Neurol. 2010 Feb;9(2):149-58. doi:10.1016/S1474-4422(10)70002-8. Epub 2010 Jan 8. Niethammer M, Tang CC, Feigin A, et al. A disease-specific metabolic brain network associated with corticobasal degeneration. Brain 2014; 137: 3036-3046. Spetsieris P.G., Ko J.H., Tang C.C., et al. Metabolic resting-state brain networks in health and disease. Proceedings of the National Academy of Sciences of the United States of America. 2015; 112(8):2563-2568. doi:10.1073. Royston P. and Altman D. G., External validation of a Cox prognostic model: principles and methods. BMC Med Res Methodol. 2013 Mar 6; 13:33. doi: 10.1186/1471-2288-13-33.