Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Assessing the significance of longitudinal data in Alzheimer's Disease forecasting

Batuhan K. Karaman, Mert R. Sabuncu

TL;DR

This study addresses predicting Alzheimer's disease progression by leveraging longitudinal, multimodal patient data. It introduces LongForMAD, a transformer encoder that processes sequences of visits augmented with horizon information, enabling predictions across multiple future time points for CN- and MCI-baseline groups. The results demonstrate that longer longitudinal histories, particularly MRI-derived biomarkers, substantially enhance predictive accuracy (notably for CN-to-MCI), though gains can saturate or slightly decline with very long histories due to potential overfitting; cognitive assessments also provide strong present-day signals. The findings support clinical adoption of longitudinal data for early detection and monitoring, and point to future work that could further improve performance via imaging encoders and interpretability analyses.

Abstract

In this study, we employ a transformer encoder model to characterize the significance of longitudinal patient data for forecasting the progression of Alzheimer's Disease (AD). Our model, Longitudinal Forecasting Model for Alzheimer's Disease (LongForMAD), harnesses the comprehensive temporal information embedded in sequences of patient visits that incorporate multimodal data, providing a deeper understanding of disease progression than can be drawn from single-visit data alone. We present an empirical analysis across two patient groups-Cognitively Normal (CN) and Mild Cognitive Impairment (MCI)-over a span of five follow-up years. Our findings reveal that models incorporating more extended patient histories can outperform those relying solely on present information, suggesting a deeper historical context is critical in enhancing predictive accuracy for future AD progression. Our results support the incorporation of longitudinal data in clinical settings to enhance the early detection and monitoring of AD. Our code is available at \url{https://github.com/batuhankmkaraman/LongForMAD}.

Assessing the significance of longitudinal data in Alzheimer's Disease forecasting

TL;DR

This study addresses predicting Alzheimer's disease progression by leveraging longitudinal, multimodal patient data. It introduces LongForMAD, a transformer encoder that processes sequences of visits augmented with horizon information, enabling predictions across multiple future time points for CN- and MCI-baseline groups. The results demonstrate that longer longitudinal histories, particularly MRI-derived biomarkers, substantially enhance predictive accuracy (notably for CN-to-MCI), though gains can saturate or slightly decline with very long histories due to potential overfitting; cognitive assessments also provide strong present-day signals. The findings support clinical adoption of longitudinal data for early detection and monitoring, and point to future work that could further improve performance via imaging encoders and interpretability analyses.

Abstract

In this study, we employ a transformer encoder model to characterize the significance of longitudinal patient data for forecasting the progression of Alzheimer's Disease (AD). Our model, Longitudinal Forecasting Model for Alzheimer's Disease (LongForMAD), harnesses the comprehensive temporal information embedded in sequences of patient visits that incorporate multimodal data, providing a deeper understanding of disease progression than can be drawn from single-visit data alone. We present an empirical analysis across two patient groups-Cognitively Normal (CN) and Mild Cognitive Impairment (MCI)-over a span of five follow-up years. Our findings reveal that models incorporating more extended patient histories can outperform those relying solely on present information, suggesting a deeper historical context is critical in enhancing predictive accuracy for future AD progression. Our results support the incorporation of longitudinal data in clinical settings to enhance the early detection and monitoring of AD. Our code is available at \url{https://github.com/batuhankmkaraman/LongForMAD}.
Paper Structure (15 sections, 4 figures, 4 tables)

This paper contains 15 sections, 4 figures, 4 tables.

Table of Contents

  1. Introduction and Related Literature
  2. Materials and Methods
  3. Dataset
  4. Input features
  5. Model
  6. Training
  7. Evaluation
  8. Experiments
  9. Experimental Details
  10. Results
  11. Impact of longitudinal history duration
  12. MRI-only case
  13. Complete-set case
  14. Impact of data collection frequency
  15. Conclusion

Figures (4)

  • Figure 1: Schematic representation of the Longitudinal Forecasting Model for Alzheimer's Disease (LongForMAD) architecture, illustrating the workflow from input to the final prediction output.
  • Figure 2: $\Delta$AUROC scores obtained with the addition of longitudinal MRI data. Legend format is "longitudinal history start year (data collection frequency)", year 0 being the present point. Error bars indicate the standard error.
  • Figure 3: AUROC scores for CN to MCI and MCI to AD conversion with up to three years of longitudinal history inclusion for the input feature set containing all data modalities. The x-axis represents the start year of the patients' longitudinal history relative to present point, year 0 being the present point. AUROC values are averaged across 5 follow-up years. Error bars indicate the standard error.
  • Figure 4: $\Delta$ AUROC scores obtained with the addition of biennial and annual longitudinal history data to the present point data for the input feature set containing all data modalities. The history duration is 2 years. Error bars indicate the standard error across these splits.