Tau Positron Emission Tomography for Predicting Dementia in Individuals With Mild Cognitive Impairment

Abstract

Importance: An accurate prognosis is especially pertinent in mild cognitive impairment (MCI), when individuals experience considerable uncertainty about future progression.

Objective: To evaluate the prognostic value of tau positron emission tomography (PET) to predict clinical progression from MCI to dementia.

Design, setting, and participants: This was a multicenter cohort study with external validation and a mean (SD) follow-up of 2.0 (1.1) years. Data were collected from centers in South Korea, Sweden, the US, and Switzerland from June 2014 to January 2024. Participant data were retrospectively collected and inclusion criteria were a baseline clinical diagnosis of MCI; longitudinal clinical follow-up; a Mini-Mental State Examination (MMSE) score greater than 22; and available tau PET, amyloid-β (Aβ) PET, and magnetic resonance imaging (MRI) scan less than 1 year from diagnosis. A total of 448 eligible individuals with MCI were included (331 in the discovery cohort and 117 in the validation cohort). None of these participants were excluded over the course of the study.

Exposures: Tau PET, Aβ PET, and MRI.

Main outcomes and measures: Positive results on tau PET (temporal meta-region of interest), Aβ PET (global; expressed in the standardized metric Centiloids), and MRI (Alzheimer disease [AD] signature region) was assessed using quantitative thresholds and visual reads. Clinical progression from MCI to all-cause dementia (regardless of suspected etiology) or to AD dementia (AD as suspected etiology) served as the primary outcomes. The primary analyses were receiver operating characteristics.

Results: In the discovery cohort, the mean (SD) age was 70.9 (8.5) years, 191 (58%) were male, the mean (SD) MMSE score was 27.1 (1.9), and 110 individuals with MCI (33%) converted to dementia (71 to AD dementia). Only the model with tau PET predicted all-cause dementia (area under the receiver operating characteristic curve [AUC], 0.75; 95% CI, 0.70-0.80) better than a base model including age, sex, education, and MMSE score (AUC, 0.71; 95% CI, 0.65-0.77; P = .02), while the models assessing the other neuroimaging markers did not improve prediction. In the validation cohort, tau PET replicated in predicting all-cause dementia. Compared to the base model (AUC, 0.75; 95% CI, 0.69-0.82), prediction of AD dementia in the discovery cohort was significantly improved by including tau PET (AUC, 0.84; 95% CI, 0.79-0.89; P < .001), tau PET visual read (AUC, 0.83; 95% CI, 0.78-0.88; P = .001), and Aβ PET Centiloids (AUC, 0.83; 95% CI, 0.78-0.88; P = .03). In the validation cohort, only the tau PET and the tau PET visual reads replicated in predicting AD dementia.

Conclusions and relevance: In this study, tau-PET showed the best performance as a stand-alone marker to predict progression to dementia among individuals with MCI. This suggests that, for prognostic purposes in MCI, a tau PET scan may be the best currently available neuroimaging marker.

Figure 1.. Area Under the Receiver Operating Characteristic Curve (AUC) Analyses to Distinguish Stable Mild Cognitive Impairment (MCI) From Progression to Dementia in the Discovery Cohort

All models were corrected for age, sex, education, Mini-Mental State Examination score, cohort, and follow-up time. The scales for the AUC bar graphs are different between the analyses predicting all-cause dementia and Alzheimer disease (AD) dementia to best visualize the differences across neuroimaging measures. The error bars indicate the standard error. Aβ indicates amyloid β; MRI, magnetic resonance imaging; MTA, medial temporal atrophy; PET, positron emission tomography. ^aIndicates that the AUC is significantly higher than the base model. Differences were assessed using the DeLong method. ^bThe differences in the discovery cohort indicate that the result was replicated in the validation cohort.

Figure 2.. Validation of Area Under the Receiver Operating Characteristic Curve (AUC) Analyses to Distinguish Stable Mild Cognitive Impairment (MCI) From Progression to Dementia

All models were corrected for age, sex, education, Mini-Mental State Examination score, and follow-up time. The scales for the AUC bar graphs are different between the analyses predicting all-cause dementia and Alzheimer disease (AD) dementia to best visualize the differences across neuroimaging measures. The error bars indicate the standard error. Aβ indicates amyloid β; MRI, magnetic resonance imaging; MTA, medial temporal atrophy; PET, positron emission tomography. ^aIndicates that the AUC is significantly higher than the base model. Differences were assessed using the DeLong method. ^bThe differences in the discovery cohort indicate that the result was replicated in the validation cohort.

Figure 3.. Association Between Risk of Clinical Progression From Mild Cognitive Impairment (MCI) to Dementia and Neuroimaging Biomarker Positivity in the Discovery Cohort

Hazard ratios were obtained using Cox proportional hazard models using clinical follow-up as the time variable and controlling for cohort. The effects shown in orange were obtained from a model including only age, sex, education, and Mini-Mental State Examination (MMSE) score (ie, the base model). The effects shown in blue represent the hazard ratio from models including 1 neuroimaging measure added to the base model at a time. The Kaplan-Meier curves display hazard ratios associated with combinations of positive biomarkers, which were based on the quantitative markers. Groups: A−T−N− = 141, A+T−N− = 37, A−T+N- = 4, A−T−N+ = 58, A+T+N− = 71, A+T−N+ = 37, A−T+N+ = 5, A+T+N+ = 57. Curves were only displayed for groups of 5 or greater. Aβ indicates amyloid β; AD, Alzheimer disease; MRI, magnetic resonance imaging; MTA, medial temporal atrophy; PET, positron emission tomography. ^aDecreased risk associated with higher scores (continuous variables; eg, Mini-Mental State Examination score) and for biomarker positivity (categorical variables; eg, tau PET). ^bIncreased risk associated with higher scores (continuous variables; eg, Mini-Mental State Examination score) and for biomarker positivity (categorical variables; eg, tau PET).

Figure 4.. Validation of Cox Proportional Hazard Models

Hazard ratios were obtained using Cox proportional hazard models using clinical follow-up as the time variable and controlling for cohort. The effects shown in orange were obtained from a model including only age, sex, education, and Mini-Mental State Examination (MMSE) score (ie, the base model). The effects shown in blue represent the hazard ratio from models including 1 neuroimaging measure added to the base model at a time. The Kaplan-Meier curves display hazard ratios associated with combinations of positive biomarkers, which were based on the quantitative markers. Groups: A−T−N− = 43, A+T−N− = 29, A−T+N− = 0, A−T−N+ = 14, A+T+N− = 16, A+T−N+ = 8, A−T+N+ = 1, A+T+N+ = 6. Curves were only displayed for groups of 5 or greater. Aβ indicates amyloid β; AD, Alzheimer disease; MRI, magnetic resonance imaging; MTA, medial temporal atrophy; PET, positron emission tomography. ^aDecreased risk associated with higher scores (continuous variables; eg, MMSE score) and for biomarker positivity (categorical variables; eg, tau PET). ^bIncreased risk associated with higher scores (continuous variables; eg, MMSE score) and for biomarker positivity (categorical variables; eg, tau PET).