Comparison of Group-Level and Individualized Brain Regions for Measuring Change in Longitudinal Tau Positron Emission Tomography in Alzheimer Disease

Abstract

Importance: Longitudinal tau positron emission tomography (PET) is a relevant outcome in clinical trials evaluating disease-modifying therapies in Alzheimer disease (AD). A key unanswered question is whether the use of participant-specific (individualized) regions of interest (ROIs) is superior to conventional approaches where the same ROI (group-level) is used for each participant.

Objective: To compare group- and participant-level ROIs in participants at different stages of the AD clinical continuum in terms of annual percentage change in tau-PET standardized uptake value ratio (SUVR) and sample size requirements.

Design, setting, and participants: This was a longitudinal cohort study with consecutive participant enrollment between September 18, 2017, and November 15, 2021. Included in the analysis were participants with mild cognitive impairment and AD dementia from the prospective and longitudinal Swedish Biomarkers For Identifying Neurodegenerative Disorders Early and Reliably 2 (BioFINDER-2) study; in addition, a validation sample (the AVID 05e, Expedition-3, Alzheimer's Disease Neuroimaging Initiative [ADNI], and BioFINDER-1 study cohorts) was also included.

Exposures: Tau PET (BioFINDER-2, [18F]RO948; validation sample, [18F]flortaucipir), 7 group-level (5 data-driven stages, meta-temporal, whole brain), and 5 individualized ROIs.

Main outcomes and measures: Annual percentage change in tau-PET SUVR across ROIs. Sample size requirements in simulated clinical trials using tau PET as an outcome were also calculated.

Results: A total of 215 participants (mean [SD] age, 71.4 (7.5) years; 111 male [51.6%]) from the BioFINDER-2 study were included in this analysis: 97 amyloid-β (Aβ)-positive cognitively unimpaired (CU) individuals, 77 with Aβ-positive mild cognitive impairment (MCI), and 41 with AD dementia. In the validation sample were 137 Aβ-positive CU participants, 144 with Aβ-positive MCI, and 125 with AD dementia. Mean (SD) follow-up time was 1.8 (0.3) years. Using group-level ROIs, the largest annual percentage increase in tau-PET SUVR in Aβ-positive CU individuals was seen in a composite ROI combining the entorhinal cortex, hippocampus, and amygdala (4.29%; 95% CI, 3.42%-5.16%). In individuals with Aβ-positive MCI, the greatest change was seen in the temporal cortical regions (5.82%; 95% CI, 4.67%-6.97%), whereas in those with AD dementia, the greatest change was seen in the parietal regions (5.22%; 95% CI, 3.95%-6.49%). Significantly higher estimates of annual percentage change were found using several of the participant-specific ROIs. Importantly, the simplest participant-specific approach, where change in tau PET was calculated in an ROI that best matched the participant's data-driven disease stage, performed best in all 3 subgroups. For the power analysis, sample size reductions for the participant-specific ROIs ranged from 15.94% (95% CI, 8.14%-23.74%) to 72.10% (95% CI, 67.10%-77.20%) compared with the best-performing group-level ROIs. Findings were replicated using [18F]flortaucipir.

Conclusions and relevance: Finding suggest that certain individualized ROIs carry an advantage over group-level ROIs for assessing longitudinal tau changes and increase the power to detect treatment effects in AD clinical trials using longitudinal tau PET as an outcome.

Figure 1.. Group-Level Regions of Interest (ROIs)

Group-level ROIs include previously published data-driven stages for [¹⁸F]RO948 (A), temporal and whole-brain meta-ROIs (B), and Braak stages (C).

Figure 2.. Approaches to Derive Individualized Regions of Interest (ROIs)

A, In the quartile 1 approach, we extracted gaussian mixture modeling (GMM)–based probabilities of being tau positive from all brain regions in the Schaefer brain atlas and defined the top 10% of ROIs as tau epicenters (ie, brain regions in which tau emerges first). After calculating the connectivity-based distance of each ROI to the participant-specific epicenters, ROIs were grouped into nonoverlapping quartiles on the basis of their connectivity-based distance to the epicenter (ie, quartile 1 [Q1] is closest to the epicenter). B, In the probability-based approach, GMM-based probabilities of being tau positive were first extracted from the FreeSurfer (Laboratory for Computational Neuroimaging) atlas and change in tau–positron emission tomography (PET) standardized uptake value ratio (SUVR) calculated across different probability intervals. In order to limit the number of intervals, 0.2 to 0.4 (ie, 0.2, 0.3, 0.4) and 0.7 to 1 (ie, 0.7, 0.8 0.9, 1) were used as lower and upper start values, respectively. The interval that resulted in the highest annual percentage change in tau-PET SUVR across participants was then selected for use. The approach is illustrated for amyloid-β (Aβ)–positive cognitively unimpaired (CU) individuals. C, In the overlap index approach, change in tau PET was determined using the mask resulting from the spatial overlap of the baseline and follow-up scans. D, In the highest tau-PET–positive data-driven stage (DDS) approach, cutoffs were applied to baseline SUVR data in group-level data-driven ROIs, with change calculated using the highest (ie, latest in the tau progression cascade) ROI that was tau positive.

Figure 3.. Subtype and Stage Inference Regions of Interest

The 4-tau–positron emission tomography (PET) subtypes are shown. A, Subtype 1, limbic-predominant phenotype. B, Subtype 2, medial temporal lobe (MTL) sparing. C, Subtype 3, posterior occipitotemporal phenotype. D, Subtype 4, lateral temporal. A representation of different stages (5, 10, 15, 20) for subtype 1 is displayed.

Figure 4.. Annual Percentage Change in Tau–Positron Emission Tomography (PET) Standardized Uptake Value Ratio (SUVR) and Sample Size Reductions in a Theoretical Clinical Trial Using Tau-PET as an End Point Using Different Regions of Interest (ROIs)

A, Mean percentage change in [¹⁸F]RO948 SURV per year along with 95% CIs for group level and individualized ROIs. P values are shown groupwise for comparisons between individualized ROIs and the best-performing group-level ROI. B, Mean percentage sample size reductions along with 95% CIs are shown for group-level and individualized ROIs compared with the temporal meta-ROI and best-performing data-driven ROIs (ie, data-driven stage I in amyloid β (Aβ)–positive cognitively unimpaired [CU]; data-driven stage II in Aβ-positive mild cognitive impairment [MCI]; data-driven stage III in Alzheimer disease [AD] dementia). P values are shown for each bar for comparisons against 0. Comparisons in sample reductions between temporal and the best-performing data-driven stage were only significant in Aβ-positive CU individuals (ie, sample size reductions were significantly greater in Aβ-positive CU individuals using data driven stage I, compared with the temporal meta-ROI). ^aP <.001. ^bP <.01.