Prediction of Longitudinal Cognitive Decline in Preclinical Alzheimer Disease Using Plasma Biomarkers

Abstract

Importance: Alzheimer disease (AD) pathology starts with a prolonged phase of β-amyloid (Aβ) accumulation without symptoms. The duration of this phase differs greatly among individuals. While this disease phase has high relevance for clinical trial designs, it is currently unclear how to best predict the onset of clinical progression.

Objective: To evaluate combinations of different plasma biomarkers for predicting cognitive decline in Aβ-positive cognitively unimpaired (CU) individuals.

Design, setting, and participants: This prospective population-based prognostic study evaluated data from 2 prospective longitudinal cohort studies (the Swedish BioFINDER-1 and the Wisconsin Registry for Alzheimer Prevention [WRAP]), with data collected from February 8, 2010, to October 21, 2020, for the BioFINDER-1 cohort and from August 11, 2011, to June 27, 2021, for the WRAP cohort. Participants were CU individuals recruited from memory clinics who had brain Aβ pathology defined by cerebrospinal fluid (CSF) Aβ42/40 in the BioFINDER-1 study and by Pittsburgh Compound B (PiB) positron emission tomography (PET) in the WRAP study. A total of 564 eligible Aβ-positive and Aβ-negative CU participants with available relevant data from the BioFINDER-1 and WRAP cohorts were included in the study; of those, 171 Aβ-positive participants were included in the main analyses.

Exposures: Baseline P-tau181, P-tau217, P-tau231, glial fibrillary filament protein, and neurofilament light measured in plasma; CSF biomarkers in the BioFINDER-1 cohort, and PiB PET uptake in the WRAP cohort.

Main outcomes and measures: The primary outcome was longitudinal measures of cognition (using the Mini-Mental State Examination [MMSE] and the modified Preclinical Alzheimer Cognitive Composite [mPACC]) over a median of 6 years (range, 2-10 years). The secondary outcome was conversion to AD dementia. Baseline biomarkers were used in linear regression models to predict rates of longitudinal cognitive change (calculated separately). Models were adjusted for age, sex, years of education, apolipoprotein E ε4 allele status, and baseline cognition. Multivariable models were compared based on model R2 coefficients and corrected Akaike information criterion.

Results: Among 171 Aβ-positive CU participants included in the main analyses, 119 (mean [SD] age, 73.0 [5.4] years; 60.5% female) were from the BioFINDER-1 study, and 52 (mean [SD] age, 64.4 [4.6] years; 65.4% female) were from the WRAP study. In the BioFINDER-1 cohort, plasma P-tau217 was the best marker to predict cognitive decline in the mPACC (model R2 = 0.41) and the MMSE (model R2 = 0.34) and was superior to the covariates-only models (mPACC: R2 = 0.23; MMSE: R2 = 0.04; P < .001 for both comparisons). Results were validated in the WRAP cohort; for example, plasma P-tau217 was associated with mPACC slopes (R2 = 0.13 vs 0.01 in the covariates-only model; P = .01) and MMSE slopes (R2 = 0.29 vs 0.24 in the covariates-only model; P = .046). Sparse models were identified with plasma P-tau217 as a predictor of cognitive decline. Power calculations for enrichment in hypothetical clinical trials revealed large relative reductions in sample sizes when using plasma P-tau217 to enrich for CU individuals likely to experience cognitive decline over time.

Conclusions and relevance: In this study, plasma P-tau217 predicted cognitive decline in patients with preclinical AD. These findings suggest that plasma P-tau217 may be used as a complement to CSF or PET for participant selection in clinical trials of novel disease-modifying treatments.

Figure 1.. Plasma P-tau217 and Longitudinal Cognition in β-Amyloid (Aβ)–Positive Cognitively Unimpaired Individuals

Longitudinal cognition among all Aβ-positive cognitively unimpaired individuals in the BioFINDER-1 cohort, by plasma P-tau217 at baseline. All plasma P-tau217 data (including quartile limits) were log₁₀ transformed and standardized as z scores compared with the Aβ-negative reference population, with 0 representing the mean in the reference population and 1 representing 1 SD higher than the mean in the reference population. P-tau217 quartile limits were −1.753 to 0.384 for quartile 1, greater than 0.384 to 1.307 for quartile 2, greater than 1.307 to 2.571 for quartile 3, and greater than 2.571 to 5.425 for quartile 4. In panels A and B, participant-specific slopes were derived from participant-specific linear regression models. Each box shows the IQR, with the median represented by the horizonal line. The upper whisker extends from the limit of the IQR to the largest value, no further than 1.5 times IQR from the IQR limit. The lower whisker extends from the limit of the IQR to the smallest value, at most 1.5 times IQR of the IQR limit. In panels C and D, trajectories were derived from linear mixed effects models with baseline plasma P-tau217 by time as a predictor, adjusted for age at baseline (ie, age at first blood sampling and cognitive testing), sex, apolipoprotein ε4 allele status, and years of education (together with the interaction terms between time and these covariates). The models included random intercepts and slopes. The effects package for R software, version 4.0.2, was used to generate estimated values, with covariates at their mean levels. Time was capped at 6 years due to sparse data available for longer follow-up times. MMSE indicates Mini-Mental State Examination; and mPACC, modified Preclinical Alzheimer Cognitive Composite.

Figure 2.. Plasma P-tau217 and Conversion to Alzheimer Disease Dementia

Kaplan-Meier survival curves among β-amyloid (Aβ)–positive cognitively unimpaired individuals in the BioFINDER-1 cohort. The data set is the same as shown in Figure 1 over a longer follow-up period (ie, not capped at 6 years). Vertical tick marks on lines indicate times at which the patient was censored. All plasma P-tau217 data (including quartile limits) were log₁₀ transformed and standardized as z scores compared with the Aβ-negative reference population, with 0 representing the mean in the reference population and 1 representing 1 SD higher than the mean in the reference population. P-tau217 quartile limits were −1.753 to 0.384 for quartile 1, greater than 0.384 to 1.307 for quartile 2, greater than 1.307 to 2.571 for quartile 3, and greater than 2.571 to 5.425 for quartile 4. The P value was derived from a log rank test for comparison between the 4 P-tau217 quartiles.

Figure 3.. Simulated Clinical Trials Using Plasma P-tau217 for Inclusion

Relative sample sizes for hypothetical clinical trials involving β-amyloid (Aβ)–positive cognitively unimpaired individuals based on the BioFINDER-1 cohort. The simulations had statistical power of 80% at α = .05 using the lmmpower function in the longpower package for R software, version 4.0.2, and assuming a 30% treatment effect for slopes, 1:1 allocation of treatment, total trial length of 48 months, and outcome measures every 12 months. Relative sample sizes are across 500 bootstrap iterations of hypothetical trials. All available longitudinal cognitive data were used for these models. The reference model (without enrichment and with 100% inclusion) was P-tau217 Q1 through Q4. In the 3 enrichment models (Q2-Q4, Q3-Q4, and Q4), only participants in higher quartiles of plasma P-tau217 were included. All plasma P-tau217 data (including quartile limits) were log₁₀ transformed and standardized as z scores compared with the Aβ-negative reference population, with 0 representing the mean in the reference population and 1 representing 1 SD higher than the mean in the reference population. P-tau217 quartile limits were −1.753 to 0.384 for quartile 1, greater than 0.384 to 1.307 for quartile 2, greater than 1.307 to 2.571 for quartile 3, and greater than 2.571 to 5.425 for quartile 4. MMSE indicates Mini-Mental State Examination; and mPACC, modified Preclinical Alzheimer Cognitive Composite.