A generalizable data-driven model of atrophy heterogeneity and progression in memory clinic settings

Abstract

Memory clinic patients are a heterogeneous population representing various aetiologies of pathological ageing. It is not known whether divergent spatiotemporal progression patterns of brain atrophy, as previously described in Alzheimer's disease patients, are prevalent and clinically meaningful in this group of older adults. To uncover distinct atrophy subtypes, we applied the Subtype and Stage Inference (SuStaIn) algorithm to baseline structural MRI data from 813 participants enrolled in the DELCODE cohort (mean ± standard deviation, age = 70.67 ± 6.07 years, 52% females). Participants were cognitively unimpaired (n = 285) or fulfilled diagnostic criteria for subjective cognitive decline (n = 342), mild cognitive impairment (n = 118) or dementia of the Alzheimer's type (n = 68). Atrophy subtypes were compared in baseline demographics, fluid Alzheimer's disease biomarker levels, the Preclinical Alzheimer Cognitive Composite (PACC-5) as well as episodic memory and executive functioning. PACC-5 trajectories over up to 240 weeks were examined. To test whether baseline atrophy subtype and stage predicted clinical trajectories before manifest cognitive impairment, we analysed PACC-5 trajectories and mild cognitive impairment conversion rates of cognitively unimpaired participants and those with subjective cognitive decline. Limbic-predominant and hippocampal-sparing atrophy subtypes were identified. Limbic-predominant atrophy initially affected the medial temporal lobes, followed by further temporal regions and, finally, the remaining cortical regions. At baseline, this subtype was related to older age, more pathological Alzheimer's disease biomarker levels, APOE ε4 carriership and an amnestic cognitive impairment. Hippocampal-sparing atrophy initially occurred outside the temporal lobe, with the medial temporal lobe spared up to advanced atrophy stages. This atrophy pattern also affected individuals with positive Alzheimer's disease biomarkers and was associated with more generalized cognitive impairment. Limbic-predominant atrophy, in all participants and in only unimpaired participants, was linked to more negative longitudinal PACC-5 slopes than observed in participants without or with hippocampal-sparing atrophy and increased the risk of mild cognitive impairment conversion. SuStaIn modelling was repeated in a sample from the Swedish BioFINDER-2 cohort. Highly similar atrophy progression patterns and associated cognitive profiles were identified. Cross-cohort model generalizability, at both the subject and the group level, was excellent, indicating reliable performance in previously unseen data. The proposed model is a promising tool for capturing heterogeneity among older adults at early at-risk states for Alzheimer's disease in applied settings. The implementation of atrophy subtype- and stage-specific end points might increase the statistical power of pharmacological trials targeting early Alzheimer's disease.

Keywords: Alzheimer’s disease; disease heterogeneity; episodic memory; executive function; structural MRI.

Figure 1

Two atrophy subtypes were identified in the DELCODE sample. (A) The progression of atrophy across SuStaIn stages. Atrophy is measured in z-scores that were scaled and centred to Aβ-negative cognitively unimpaired (CU) participants. The displayed values were obtained by (C) modelling each region of interest-based atrophy marker across SuStaIn stages using monotone regression splines. The open diamond denotes the knot position at SuStaIn stage = 5. (B) Distributions of diagnostic groups across the atrophy-negative group and the two atrophy subtypes. (D) Distributions of SuStaIn stages for each subtype and diagnostic group. Data-points are jittered on the x-axis. ^aPercentages refer to the proportion among all participants assigned an atrophy subtype (excluding atrophy-negative individuals). DAT = dementia of the Alzheimer’s type; MCI = mild cognitive impairment; MTL = medial temporal lobe; ROI = region of interest; SCD = subjective cognitive decline; SuStaIn = Subtype and Stage Inference.

Figure 2

The two atrophy subtypes and the atrophy-negative group exhibit different cross-sectional clinical profiles. (A–C) The top row shows standardized β-coefficients from ordinary least squares linear regression models predicting each continuous dependent variable, with atrophy subtype as the predictor of interest. (D–F) The bottom row shows unstandardized estimates of the effect of subtype in binomial logistic regression models predicting each binary dependent variable, with subtype as the predictor of interest. Initially, models were fitted to compare participants with (A and D) hippocampal-sparing atrophy and (B and E) limbic-predominant atrophy against the atrophy-negative group, before (C and F) models comparing the two atrophy subtypes against each other were fitted. The displayed effects are controlled for diagnostic group when predicting demographic variables (including APOE ε4 status), for age, sex and diagnostic group when predicting fluid biomarkers and for age, sex, education and diagnostic group when predicting cognitive scores. Error bars visualize 95% confidence intervals. EXEC = executive functions and mental processing speed; MEM = learning and memory; NfL = neurofilament light chain; PACC-5 = Preclinical Alzheimer Cognitive Composite.

Figure 3

Visualization of cognitive trajectories across pseudo-longitudinal SuStaIn stages and longitudinal annual assessments. (A) PACC-5 scores, (B) MEM and (C) EXEC domain scores, as well as (D) MEM–EXEC scores, were fitted across SuStaIn stages using natural cubic regression splines that controlled for age, sex and years of education. The open diamond denotes denotes the knot position at SuStaIn stage = 5. (D) Negative values (below dashed line) represent a pronounced impairment of the MEM domain, whereas positive values (above dashed line) represent a pronounced impairment of the EXEC domain. Linear mixed effects models were used to estimate longitudinal PACC-5 slopes for (E) each atrophy subtype and (F and G) across SuStaIn stages (here displayed stratified by median split). (H–J) These analyses were repeated in only CU and SCD participants. CU = cognitively unimpaired; EXEC = executive functions and mental processing speed; MEM = learning and memory; PACC-5 = Preclinical Alzheimer Cognitive Composite; SCD = subjective cognitive decline; SuStaIn = Subtype and Stage Inference.

Figure 4

Kaplan–Meier survival curves displaying the estimated probability of remaining without an incident mild cognitive impairment diagnosis over time among cognitively unimpaired subjects and those with subjective cognitive decline. Visualized are: (A) the effect of atrophy subtype and the effects of baseline SuStaIn stage (here displayed stratified by median split) in both the (B) limbic-predominant and (C) hippocampal-sparing atrophy subtypes. SuStaIn = Subtype and Stage Inference.

Figure 5

Overview of the BioFINDER-2 model and subject-level generalizability testing across cohorts. (A) Two highly comparable atrophy progression patterns were identified in the DELCODE and BioFINDER-2 cohorts. (B) SuStaIn stages and (C) subtype probabilities generated using the internally and externally trained models were highly correlated. Note that the probability of limbic-predominant atrophy can be interpreted as one minus the probability of hippocampal-sparing atrophy. (D) Subtype assignments were highly consistent across internally and externally trained models. BC = Bhattacharyya coefficient; SuStaIn = Subtype and Stage Inference.