In a recent study published in Nature Mental Health, a group of researchers evaluated if a neurobiological model of the default-mode network (DMN) effective connectivity can predict future dementia diagnosis at the individual level.
Study: Early detection of dementia with default-mode network effective connectivity. Image Credit: Komsan Loonprom/Shutterstock.com
Background
There is a significant interest in reducing dementia’s growing burden, with Alzheimer’s disease (AD) as the leading cause. Early detection of neural changes could enable personalized prevention strategies.
Resting-state functional magnetic resonance imaging (rs-fMRI) maps brain connectivity and shows altered patterns in AD, but traditional methods lack precision for individual risk prediction. Effective connectivity analysis, modeling causal brain interactions, offers better detection.
Early DMN dysconnectivity patterns are linked to genetic risk factors for AD and social isolation, suggesting their potential as preclinical biomarkers. Further research is needed to validate effective connectivity analysis for early dementia diagnosis and refine prevention strategies.
About the study
Controlling for confounders like age, sex, ethnicity, and head motion, the present study used data from the United Kingdom Biobank (UKB). An initial sample of 148 dementia cases was identified, with ten matched controls for each case.
After preprocessing, the final sample included 103 cases and 1,030 controls, with 81 cases undiagnosed at the time of MRI data acquisition.
MRI data were acquired using Siemens Skyra 3 T scanners, focusing on T1-weighted and rs-fMRI data. Preprocessing involved segmenting and normalizing images and estimating head motion.
Effective connectivity was estimated using spectral dynamic causal modeling (DCM), fitting a fully connected model for each participant and using parametric empirical Bayes modeling for group-level differences.
An elastic-net regularized logistic regression model, with k-fold cross-validation, was used to classify dementia cases based on effective connectivity features. Prognostic models predicted the time until diagnosis. The study also compared the predictive power of effective connectivity with structural MRI features and assessed functional connectivity and cognitive data.
Further analysis examined the association between DMN effective connectivity and modifiable risk factors like hypertension, diabetes, and social isolation, as well as AD polygenic risk scores. Ethical approval and informed consent were obtained for the study.
Study results
After exclusions for image quality and excessive in-scanner head motion, the final sample comprised 103 dementia cases (22 with prevalent dementia and 81 who later developed dementia) and 1,030 matched controls.
The incident cases had a median time to diagnosis of 3.7 years. The total sample had a mean age of 70.4 at the time of MRI data acquisition, and cases and controls were matched on age, sex, ethnicity, handedness, and geographical location of the testing center.
Cases performed worse than controls in four cognitive tests, reflecting possible cognitive decline or reduced cognitive reserve.
Blood Oxygen Level Dependent (BOLD) time-series were extracted from ten pre-defined DMN regions, including the precuneus, anterior and dorsomedial prefrontal cortices, and medial and lateral temporal cortices. A fully connected DCM estimated the effective connectivity between each region-of-interest (ROI) pair.
Bayesian model reduction and averaging estimated the simplest effective connectivity map explaining group-level differences between cases and controls, controlling for age, sex, and head motion.
Fifteen connectivity parameters significantly differed, with increased inhibition from the Ventromedial Prefrontal Cortex (vmPFC) to Left Parahippocampal Formation (lPHF) and Left Intraparietal Cortex (lIPC) to lPHF, and attenuated inhibition from Right Parahippocampal Formation (rPHF) to Dorsomedial Prefrontal Cortex (dmPFC).
An elastic-net logistic regression model, trained on these parameters, predicted future dementia diagnosis with an area under the curve (AUC) of 0.824. A sensitivity analysis using the full model of 100 parameters yielded a slightly reduced AUC of 0.816. Effective connectivity also predicted the time until diagnosis.
Thirty-seven connectivity parameters were associated with the time until diagnosis, including the three largest differences. An elastic-net linear regression model showed a positive correlation between actual and predicted time until diagnosis (Spearman’s ρ = 0.53).
Comparative analyses with other MRI-based markers, including volumetric and functional connectivity data, showed that effective connectivity parameters had superior diagnostic performance.
Volumetric models yielded moderate diagnostic value (AUC of 0.671) and chance-level prognostication. Functional connectivity models were performed at the chance level for both diagnosis and prognostication. Cognitive data alone had moderate diagnostic performance (AUC of 0.628) and chance-level prognostication.
Effective connectivity changes were examined for associations with dementia risk factors. The AD polygenic risk score is strongly associated with the effective connectivity index, suggesting these changes reflect Alzheimer’s pathology.
Social isolation was the only modifiable risk factor significantly associated with the effective connectivity index.
Mediation analysis showed that DMN effective connectivity partially mediated the relationship between genetic risk and dementia incidence, as well as the association between social isolation and dementia.
Conclusions
The study reveals that a neurobiologically informed DMN effective connectivity model can accurately predict dementia onset.
The classifier outperformed those based on volumetric and functional connectivity data and past structural MRI-based models. Clinically, rs-fMRI could identify early neural network signatures of dementia, aiding the early use of disease-modifying drugs.
Effective connectivity predicts dementia development and time until diagnosis better than traditional biomarkers. The study also links DMN connectivity changes to Alzheimer’s risk and social isolation, highlighting its potential as an early detection biomarker.