Crunching the numbers. Responsibly.
Alzheimer's disease (AD) is a major cause of dementia and cognitive decline. Here, we assessed how episodic memory (EM) network dysfunction, a hallmark of AD, is related to the longitudinal progression of AD biomarkers, neurodegeneration and cognition using data from the DZNE DELCODE study. This data set includes over 1000 longitudinal functional magnetic resonance imaging measurements of EM network function. We related activation and deactivation of EM to individual disease progression scores from a disease progression model. Voxel-wise analyses revealed widespread loss of deactivation and activation with disease progression. Trajectories for the loss of deactivation were nonlinear, associated with amyloid- and tau-positivity and visually preceded trajectories of cognitive decline. The relationship between deactivation and cognitive decline was partly independent of neurodegeneration. Our results provide evidence that synaptic dysfunction and neurodegeneration are independent drivers of cognitive decline, providing a rationale for targeting synaptic dysfunction along the AD cascade.
Trajectories for the loss of deactivation were nonlinear, associated with amyloid- and tau-positivity and visually preceded trajectories of cognitive decline.
model-derived timing and a small nonlinear effect support the direction but limit temporal certainty
The relationship between deactivation and cognitive decline was partly independent of neurodegeneration.
cross-sectional adjusted associations measure cognitive level rather than longitudinal decline
Our results provide evidence that synaptic dysfunction and neurodegeneration are independent drivers of cognitive decline.
observational BOLD associations and mediation cannot identify independent causal drivers
Voxel-wise analyses revealed widespread loss of deactivation and activation with disease progression.
FWE-corrected full-sample results support the pattern, though the activation finding weakened in sensitivity analysis
Derived from the full evaluation — not a separate score.
Strengths
The analysis draws on 493 baseline fMRI participants and longitudinal scans across four time points, providing unusual scale for an AD task-fMRI study.
↳ Methods, “Participants”; Results, “Longitudinal activation change is moderated by disease stage”
Voxelwise FWE-controlled analyses are supplemented by biomarker-specific adjustment, mediation models, residual-variability checks, and longitudinal moderation tests.
↳ Results, Figs. 2–5; Methods, “Statistical analyses”
The paper considers atrophy, white-matter hyperintensities, effective connectivity, vascular contributions, dedifferentiation, regression to the mean, and measurement variability.
↳ Introduction, competing mechanisms; Discussion, limitations
Limitations
Adjusted associations and cross-sectional mediation do not establish that synaptic dysfunction is an independent driver or probable cause of cognitive impairment, nor that the inferred cascade is confirmed.
↳ Abstract, final sentence; Discussion, extended ATN model paragraph; Fig. 6 legend
The fMRI trajectory is derived from activity clusters selected for disease-stage association, and no uncertainty interval is reported for its fastest-change time. The disease model also assumes a shared progression rate.
↳ Methods, spline analysis and disease-progression model; Results, Fig. 2c; Discussion, limitations
CSF data were available for 218 of 493 participants, the model was fitted in a selected 208-person subgroup, and follow-up availability differed strongly across diagnostic groups without a reported missingness analysis.
↳ Methods, “Participants” and “Participant selection for the model fit”; Results, longitudinal analysis
The strongest contribution is the integration of a large longitudinal task-fMRI resource with a multivariate AD disease-progression model and several adjusted or longitudinal follow-up analyses. The reported associations support a nonlinear loss of deactivation and partial statistical separation from medial temporal atrophy, but the inferred timing remains model-dependent and the within-person change–change correlations were null. The main score constraint is the shift from association to claims that dysfunction is an independent driver, a probable cause, and part of a confirmed cascade. The extensive limitations section improves transparency but directly acknowledges that larger multimodal longitudinal samples are needed to confirm the proposed extended ATN model.
Nabu’s assessment, alongside the field’s view.
Are you an author of this paper?
Sound3.1
Confidence mediumThe study meaningfully extends prior AD task-fMRI research by positioning episodic-memory network activity on a multivariate disease-progression axis and adding a longitudinal moderation analysis. The advance is incremental because the biological sequence remains model-derived rather than independently identified.
“We generated a computational disease progression marker using a multivariate DPM approach.”
The analysis combines a sizeable multicentre cohort, voxelwise FWE-controlled inference, a Gaussian-process disease model, adjusted biomarker analyses, and longitudinal mixed models. Methodological Rigour is limited by differential follow-up, selected biomarker availability, model-dependent staging, and causal conclusions unsupported by an observational identification strategy.
“our data confirm a neuropathological cascade”
The progression from disease staging through voxelwise, biomarker-specific, sensitivity, and longitudinal analyses is readily traceable. However, “independent drivers,” “confirm,” and “probable cause” communicate materially stronger causal conclusions than the adjusted associations establish.
“network dysfunction as a probable cause for cognitive impairment”
The paper engages competing accounts, contrary timing evidence, vascular explanations, dedifferentiation, and extensive measurement limitations. Contextual integration is undermined when the conclusion claims confirmation despite the limitations stating that larger multimodal longitudinal samples are needed to confirm the model.
“in order to confirm our extended ATN model, larger samples”
Caveats4 of 4 checks
The paper contains a reporting inconsistency in the number of measurements used for the disease-progression model, alongside several smaller numerical or reporting discrepancies. These issues warrant caution but do not establish that the reported estimates are unreliable.
Written consent, ethics approval, study registration, source-data availability, and public model code are declared. No text-grounded research-conduct concern was established.
Flags: 3 declared / 5 total
86 references in manuscript 84 of 84 checkable references found in an index 2 have no canonical index record — counted, but not index-checkable 2 references confirmed by manual review
No retraction notice found in Retraction Watch.
Sources: Retraction Watch ✓
Where this paper’s evidence sits on the path from initial observation to real-world use.
The work characterises associations under observational conditions but does not validate a diagnostic tool, treatment target, or intervention. Further longitudinal multimodal replication is explicitly required before practical use.
AI-generated, human-governed. Something look off? Contact us to request a review.