Multisite study of the relationships between antemortem [11C]PIB-PET Centiloid values and postmortem measures of Alzheimer's disease neuropathology

Abstract

Introduction: We sought to establish the relationships between standard postmortem measures of AD neuropathology and antemortem [¹¹C]PIB-positron emission tomography ([¹¹C]PIB-PET) analyzed with the Centiloid (CL) method, a standardized scale for Aβ-PET quantification.

Methods: Four centers contributed 179 participants encompassing a broad range of clinical diagnoses, PET data, and autopsy findings.

Results: CL values increased with each CERAD neuritic plaque score increment (median -3 CL for no plaques and 92 CL for frequent plaques) and nonlinearly with Thal Aβ phases (increases were detected starting at phase 2) with overlap between scores/phases. PET-pathology associations were comparable across sites and unchanged when restricting the analyses to the 56 patients who died within 2 years of PET. A threshold of 12.2 CL detected CERAD moderate-to-frequent neuritic plaques (area under the curve = 0.910, sensitivity = 89.2%, specificity = 86.4%), whereas 24.4 CL identified intermediate-to-high AD neuropathological changes (area under the curve = 0.894, sensitivity = 84.1%, specificity = 87.9%).

Discussion: Our study demonstrated the robustness of a multisite Centiloid [¹¹C]PIB-PET study and established a range of pathology-based CL thresholds.

Keywords: Alzheimer's disease neuropathologic changes; CERAD; Centiloid; Harmonization; Neuropathology; Pittsburgh compound-B; Positron emission tomography; Thal; Threshold; β-amyloid.

Figure 1.. Overview of the present study.

A) PET data from the original article (34 young controls (YC) and 45 patients with AD) were downloaded from the GAAIN website and processed at UCSF following all the Centiloid guidelines (the * indicates that values were calculated on site based on the images downloaded from the website, in compliance with Centiloid nomenclature). PET were warped to template using SPM8, transformed into Standardized Uptake Value Ratios (SUVR) maps using the whole cerebellum as a reference region (shown in green), and average values were extracted from the cortical volume of interest (VOI, shown in purple). The average SUVR of the YC/AD groups were used to define the 0/100 CL points. For all 79 scans, CL values were calculated and compared to the CL values specified in the original paper (scatter plot) to validate the Centiloid method implementation on our site based on the CL quality control (QC) requirements (i.e. regression slope between 0.98 and 1.02, intercept between −2 and 2, R² above 0.98). B) Flow diagram of the original data used in the present study showing an overview of the main analyses. The histogram shows the distribution of PIB-PET CL values in the whole group (n=179, spanning from −26 to 169) and the scatter plot illustrates the distribution of CL values (individual data points, medians, quartiles) according to primary clinical diagnosis at the time of PET. Pairwise group comparisons were conducted using Mann-Whitney tests and corresponding p-values are shown on the plot (for the sake of clarity, we only indicate the results of comparison between contiguous groups, but all the other p’s < 0.05).

Figure 2.. Relationships between CL values and post mortem measures of Aβ pathology (A) and levels of AD neuropathologic changes (B).

Bar graphs indicate the distribution of the neuropathologic measures in the whole group, A) CERAD, n=179 and Thal, n=162; B) ADNC, n=167. The bubble plot illustrates the relationships between the two measures of Aβ pathology (correlation between ordinal scales is shown using gamma coefficient and 95%CI). Scatter plots illustrate the distribution of CL values according to neuropathologic measures, showing individual data points (blue: none-to-sparse CERAD, red: moderate-to-frequent CERAD), medians (the actual values are also specified), and quartiles. Pairwise group comparisons were conducted using Mann-Whitney tests and corresponding p-values are shown on the plot (we only indicate the results of comparison between contiguous groups, but all the other p’s < 0.05). Spearman’s ρ correlation coefficients [95%CI] are indicated. Dotted lines illustrate the thresholds identified by the ROC analyses (see text): 12.2 to detect moderate-tofrequent CERAD scores and 24.4 to detect intermediate-to-high ADNC levels.

Figure 3.. Complementary analyses

A) Robust associations between CL values and CERAD scores across centers. B) Main analyses (similar to Figures 2A) restricted to patients with PET to death interval ≤ 2 years (n=63 from all centers). C) Relationships between visual reads of the SUVR scans (by three independent raters blind to clinical information and CL values) and CL values in the subsample from UCSF/UCD (n=73). D) Relationships between CL values and Thal phase in individuals with maximal (frequent) CERAD scores (n=76 from all centers, including n=22 who died within 2 years of PET). Scatter plots show individual data points (blue: none-to-sparse CERAD, red: moderate-to-frequent CERAD), medians (if n ≥ 4, the actual values are also specified), and quartiles (if n ≥ 8). Spearman’s ρ correlation coefficients [95%CI] are indicated. Dotted lines illustrate the thresholds identified by the ROC analyses (see text): 12.2 to detect moderate-to-frequent CERAD scores and 24.4 to detect intermediate-to-high ADNC levels.