Predicting Short-term MCI-to-AD Progression Using Imaging, CSF, Genetic Factors, Cognitive Resilience, and Demographics

Yogatheesan Varatharajah, Vijay K. Ramanan, Ravishankar K. Iyer, Prashanthi Vemuri, Michael W. Weiner, Paul Aisen, Ronald Petersen, Clifford R. Jack, Andrew J. Saykin, William J. Jagust, John Q. Trojanowki, Arthur W. Toga, Laurel Beckett, Robert C. Green, John C. Morris, Leslie M. Shaw, Zaven S. Khachaturian, Greg Sorensen, Marı́a C. Carrillo, Lew Kuller, Marc Raichle, Steven M. Paul, Peter J. Davies, Howard Fillit, Franz Hefti, David M. Holtzman, M. Marcel Mesulam, William C. Potter, Peter J. Snyder, Adam J. Schwartz, Tom Montine, Ronald G. Thomas, Michael Donohue, Sarah Walter, Devon Gessert, Tamie Sather, Gus Jiminez, Archana B. Balasubramanian, Jennifer Mason, Iris Sim, Danielle Harvey, Matt A. Bernstein, Nick C. Fox, Paul M. Thompson, Norbert Schuff, Charles DeCarli, Bret Borowski, Jeff Gunter, Matthew L. Senjem, David Jones, Kejal Kantarci, Chad Ward, Robert A. Koeppe, Norm Foster, Eric M. Reiman, Kewei Chen, Chester A. Mathis, Susan Landau, Nigel J. Cairns, Erin Franklin, Lisa Taylor‐Reinwald, Virginia Lee, Magdalena Korecka, Michal Figurski, Karen Crawford, Scott Neu, Tatiana Foroud, Steven Potkin, Kelley Faber, Sungeun Kim, Kwangsik Nho, Leon J. Thal, Neil Buckholtz, Marilyn Albert, Richard Frank, John Hsiao, Jeffrey Kaye, Joseph F. Quinn, Lisa C. Silbert, Betty Lind, Raina Carter, Sara Dolen, Lon S. Schneider, Sonia Pawluczyk, Mauricio Beccera, Liberty Teodoro, Bryan M. Spann, James B. Brewer, Helen Vanderswag, Adam Fleisher, Judith L. Heidebrink, Joanne Lord, Sara S. Mason, Colleen S. Albers, David S. Knopman, Kris Johnson, Rachelle S. Doody, Javier Villanueva‐Meyer, Valory Pavlik, Victoria ShibleyScientific 2019 · DOI

Abstract

In the Alzheimer's disease (AD) continuum, the prodromal state of mild cognitive impairment (MCI) precedes AD dementia. Identifying MCI individuals at risk of progression is critical for clinical management. This study developed generalizable multivariate models integrating high-dimensional data—including multimodal neuroimaging, cerebrospinal fluid (CSF) biomarkers, genetic factors, and measures of cognitive resilience—to identify MCI individuals who progress to AD within three years. The models achieved clinically relevant accuracy (~93%). While markers of AD pathophysiology (amyloid, tau, neuronal injury) accounted for most of the predictive variance, the expression of CR1 (complement receptor 1) was found to add unique, independent predictive value.

Introduction

Mild Cognitive Impairment (MCI) serves as a critical window for intervention in the Alzheimer’s Disease (AD) spectrum, with 10–15% of patients progressing to full dementia annually. However, the high variability in progression rates poses a significant challenge for clinical management and trial recruitment. Identifying "rapid progressors" requires navigating high-dimensional datasets where the number of features often dwarfs the sample size, leading to risks of overfitting. This research addresses these challenges by developing a robust machine learning framework that integrates multimodal biomarkers to predict short-term progression with high sensitivity.

Data Modalities and Participants

The study utilized data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), focusing on 135 individuals with MCI at baseline. The cohort was divided into 39 "Progressors" (MCI-P) and 96 "Non-Progressors" (MCI-NP).

Conceptual diagram of the multimodal factors integrated to predict the progression from MCI to AD.

The predictive models incorporated 94 features across five modalities:

  • CSF Biomarkers: Amyloid-beta and Tau levels.
  • Neuroimaging: MRI measures of atrophy and PET scans for glucose metabolism and amyloid deposition.
  • Genetics: APOE status and the expression of top AD-associated susceptibility genes.
  • Cognitive Resilience: Years of education and IQ (ANART scores).
  • Demographics: Age and gender.

Methodological Framework

To ensure the generalizability of the findings, the researchers employed a rigorous machine learning pipeline designed to mitigate overfitting in high-dimensional spaces.

The machine learning workflow, featuring nested cross-validation and feature selection using the Joint Mutual Information (JMI) criterion.

Key methodological pillars included:

  1. Linearity Assessment: Using a "histogram of projections" method, the authors confirmed that the data was linearly separable. This justified the use of linear classifiers like Linear SVM and GLM-ElasticNet, which are more robust than non-linear alternatives in small-sample contexts.
  2. Feature Selection: The Joint Mutual Information (JMI) criterion was used to identify features that are individually relevant and non-redundant.
  3. Nested Cross-Validation: A 5-fold outer and 10-fold inner cross-validation loop was used to optimize hyperparameters and assess performance on unseen data.
  4. LASSO Regression: This was applied to distill the 94 features into a "minimal set" of the most powerful independent predictors.

Key Insights and Discussion

The Independent Value of CR1

A major finding of this study is the role of the CR1 (Complement Receptor 1) gene. While APOE ε4 is the most recognized genetic risk factor, its predictive power often overlaps with amyloid markers. In contrast, CR1 expression—linked to immune pathways—provided independent predictive value, suggesting it captures a distinct pathological process not fully represented by standard amyloid or tau markers.

Cognitive Resilience and Brain Structure

The study observed that measures of cognitive resilience, such as education and IQ, lost their predictive power once MRI-derived brain volume measures were included. This suggests that the protective effects of cognitive resilience may be physically manifested in preserved brain structure (e.g., in the frontal lobes), which neuroimaging captures directly.

Sufficiency of Linear Models

The high correlation between linear and non-linear model performance indicates that linear boundaries are sufficient for this classification task. In medical informatics, simpler linear models are often preferred as they are less prone to memorizing noise in small datasets.

Results

The optimized Linear SVM model achieved an Area Under the ROC Curve (AUC) of 0.93 and a sensitivity of ~93%. The researchers identified a lean set of variables—including specific MRI regions like the Insula and Parahippocampal gyrus, CSF Tau/Amyloid, and CR1 expression—that can predict progression with accuracy comparable to the full feature set.

Future Directions

While the results are promising, the study is limited by its sample size (N=135). Validation in larger, more diverse cohorts is necessary to confirm the predictive utility of CR1. Furthermore, future work should investigate whether less invasive blood-based biomarkers can achieve similar accuracy, potentially reducing the reliance on expensive PET scans and invasive CSF collection.

Reference: Varatharajah, Y., et al. (2019). Predicting Short-term MCI-to-AD Progression Using Imaging, CSF, Genetic Factors, Cognitive Resilience, and Demographics. Scientific Reports, 9:2235.

Summary updated

Abstract

In the Alzheimer's disease (AD) continuum, the prodromal state of mild cognitive impairment (MCI) precedes AD dementia and identifying MCI individuals at risk of progression is important for clinical management. Our goal was to develop generalizable multivariate models that integrate high-dimensional data (multimodal neuroimaging and cerebrospinal fluid biomarkers, genetic factors, and measures of cognitive resilience) for identification of MCI individuals who progress to AD within 3 years. Our main findings were i) we were able to build generalizable models with clinically relevant accuracy (~93%) for identifying MCI individuals who progress to AD within 3 years; ii) markers of AD pathophysiology (amyloid, tau, neuronal injury) accounted for large shares of the variance in predicting progression; iii) our methodology allowed us to discover that expression of CR1 (complement receptor 1), an AD susceptibility gene involved in immune pathways, uniquely added independent predictive value. This work highlights the value of optimized machine learning approaches for analyzing multimodal patient information for making predictive assessments.
BibTeX
@inproceedings{varatharajah2019predicting,
  author = {Yogatheesan Varatharajah and Vijay K. Ramanan and Ravishankar K. Iyer and Prashanthi Vemuri and Michael W. Weiner and Paul Aisen and Ronald Petersen and Clifford R. Jack and Andrew J. Saykin and William J. Jagust and John Q. Trojanowki and Arthur W. Toga and Laurel Beckett and Robert C. Green and John C. Morris and Leslie M. Shaw and Zaven S. Khachaturian and Greg Sorensen and Marı́a C. Carrillo and Lew Kuller and Marc Raichle and Steven M. Paul and Peter J. Davies and Howard Fillit and Franz Hefti and David M. Holtzman and M. Marcel Mesulam and William C. Potter and Peter J. Snyder and Adam J. Schwartz and Tom Montine and Ronald G. Thomas and Michael Donohue and Sarah Walter and Devon Gessert and Tamie Sather and Gus Jiminez and Archana B. Balasubramanian and Jennifer Mason and Iris Sim and Danielle Harvey and Matt A. Bernstein and Nick C. Fox and Paul M. Thompson and Norbert Schuff and Charles DeCarli and Bret Borowski and Jeff Gunter and Matthew L. Senjem and David Jones and Kejal Kantarci and Chad Ward and Robert A. Koeppe and Norm Foster and Eric M. Reiman and Kewei Chen and Chester A. Mathis and Susan Landau and Nigel J. Cairns and Erin Franklin and Lisa Taylor‐Reinwald and Virginia Lee and Magdalena Korecka and Michal Figurski and Karen Crawford and Scott Neu and Tatiana Foroud and Steven Potkin and Kelley Faber and Sungeun Kim and Kwangsik Nho and Leon J. Thal and Neil Buckholtz and Marilyn Albert and Richard Frank and John Hsiao and Jeffrey Kaye and Joseph F. Quinn and Lisa C. Silbert and Betty Lind and Raina Carter and Sara Dolen and Lon S. Schneider and Sonia Pawluczyk and Mauricio Beccera and Liberty Teodoro and Bryan M. Spann and James B. Brewer and Helen Vanderswag and Adam Fleisher and Judith L. Heidebrink and Joanne Lord and Sara S. Mason and Colleen S. Albers and David S. Knopman and Kris Johnson and Rachelle S. Doody and Javier Villanueva‐Meyer and Valory Pavlik and Victoria Shibley},
  title = {{Predicting Short-term MCI-to-AD Progression Using Imaging, CSF, Genetic Factors, Cognitive Resilience, and Demographics}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-38793-3},
}
Figure