Prediction of 7-year’s Conversion from Subjective Cognitive Decline to Mild Cognitive Impairment

Ling Yue, Dan Hu, Han Zhang, Junhao Wen, Ye Wu, Wei Li, Lin Sun, Xia Li, Jinghua Wang, Guanjun Li, Tao Wang, Dinggang Shen, Shifu Xiao

Abstract

Subjective cognitive decline (SCD) represents a high-risk, yet poorly understood, stage occurring years before the clinical onset of Alzheimer’s disease (AD). This study analyzed 76 individuals with SCD over a seven-year period, utilizing baseline T1-weighted MRI and a battery of 25 clinical measurements. Using a machine learning-based model, researchers successfully discriminated between individuals who remained stable (sSCD) and those who progressed to Mild Cognitive Impairment (pSCD). The findings suggest that a combination of stroke history, education level, baseline cognitive scores, and specific neuroimaging markers—specifically the left amygdala and white matter at the right superior temporal sulcus—can predict long-term cognitive decline. This research represents the largest retrospective study of SCD-to-MCI conversion with a seven-year follow-up, offering a potential roadmap for early screening and intervention.

Introduction

The neuropathological hallmarks of Alzheimer’s disease often begin 15–20 years before the first clinical symptoms manifest. By the time a patient is diagnosed with Mild Cognitive Impairment (MCI), significant and often irreversible neuronal loss has already occurred. Subjective Cognitive Decline (SCD)—a state where individuals report memory lapses despite performing normally on standard neuropsychological tests—is increasingly recognized as the earliest detectable stage of this progression.

However, SCD is a heterogeneous category. While some individuals will progress to clinical impairment, many remain stable for years. Identifying the "progressors" (pSCD) from the "stable" (sSCD) group using non-invasive baseline measures is a critical challenge for clinical trial recruitment and early intervention. This paper addresses this gap by utilizing a unique seven-year longitudinal dataset to identify a predictive "signature" of cognitive decline.

Methodology

Feature Selection and Classification

The researchers analyzed 223 baseline features, including 25 clinical/demographic variables and 198 MRI-derived structural features. To manage the high dimensionality of the data and avoid overfitting, a three-stage feature selection process was employed:

  1. Relief Algorithm: This algorithm was used to identify features that are effective in combination, rather than just in isolation.
  2. Cost-Sensitive Support Vector Machines (CSVM): Given that the dataset was imbalanced (52 stable vs. 24 progressive cases), a cost-sensitive approach was used to penalize the misclassification of the progressive group more heavily, ensuring the model remained sensitive to high-risk individuals.
  3. Validation: The model was validated using nested leave-one-out cross-validation (LOOCV). To ensure biological relevance, the identified markers were further tested against independent datasets, including the Alzheimer’s Disease Neuroimaging Initiative (ADNI).

Dataset Characteristics

The primary data was drawn from the China Longitudinal Ageing Study (CLAS), involving 76 SCD individuals with baseline T1-MRI and a follow-up evaluation seven years later. This timeframe is significantly longer than the 2–5 year follow-ups typical of existing literature, providing a more robust view of the conversion process.

Key Results and Predictors

The refined machine learning model, utilizing only the top five features, achieved an AUC of 0.7997 with an overall accuracy of 69.74%. The study identified five primary predictors of 7-year conversion:

  1. History of Stroke: A significant clinical risk factor for accelerated decline.
  2. Education Level: Lower education levels were associated with a higher risk of progression.
  3. Baseline MoCA Score: Lower initial scores on the Montreal Cognitive Assessment (MoCA) predicted future impairment.
  4. Left Amygdala Volume: Decreased volume (atrophy) in the left amygdala was a key structural marker.
  5. White Matter at the Right STS: Increased white matter volume at the banks of the right superior temporal sulcus (wmSTSbanks).

Discussion: Biological Insights

The Compensation Hypothesis

One of the most intriguing findings was the enlargement of white matter volume at the banks of the right superior temporal sulcus (STS) in progressors. While AD is typically characterized by atrophy, the authors suggest this early increase may represent a compensatory mechanism. In this phase, the brain may increase connectivity or structural density to maintain functional performance while other regions begin to fail.

The Amygdala Paradox

While a shrunk left amygdala was a general predictor of progression, a secondary observation emerged: within the progressive group, those with larger amygdalas at baseline actually experienced a faster rate of decline later. This aligns with the "cognitive reserve" theory, where a larger brain structure may mask early pathological changes, leading to a more rapid and severe collapse once the disease eventually overcomes the brain's structural resilience.

Synergy of Features

The study emphasizes that neither imaging nor clinical data alone is sufficient for long-term prediction. The most robust results came from the synergy of demographic history, cognitive screening, and structural MRI metrics, suggesting that multi-modal screening is essential for identifying high-risk SCD populations.

Limitations and Future Directions

Despite the long follow-up period, the sample size of 76 subjects remains relatively small for machine learning applications. Additionally, the study lacked PET imaging (amyloid/tau) or APOE genotyping, meaning the researchers could not definitively confirm that every conversion to MCI was due to Alzheimer's pathology specifically. Future work should focus on integrating these structural markers with fluid biomarkers (blood or CSF) and increasing the frequency of follow-up evaluations to better map the trajectory of decline.

Conclusion

This study demonstrates that the far-future cognitive decline of individuals with SCD is predictable using baseline measures. By identifying a specific signature of five clinical and structural markers, clinicians may eventually be able to identify at-risk individuals nearly a decade before clinical impairment, opening a vital window for preventative therapies.

Summary updated

Abstract

Abstract Subjective cognitive decline (SCD) is a high-risk yet less understood status years before Alzheimer’s disease (AD). This work included 76 SCD individuals with two (baseline and seven years later) neuropsychological evaluations and a baseline T1-MRI. A machine learning-based model was trained based on 198 baseline neuroimage features and a battery of 25 clinical measurements to discriminate 24 progressive SCDs from 52 stable SCDs. The SCD progression was satisfactorily predicted with combined features. A history of stroke, a low education level, a low baseline MoCA score, a shrunk left amygdala, and enlarged white matter at the banks of the right superior temporal sulcus favor the progression. This is to date the largest retrospective study of SCD-to-MCI conversion with the longest follow-up, suggesting predictable far-future cognitive decline for the risky populations with baseline measures only. These findings provide valuable knowledge to the future neuropathological studies of AD in its prodromal phase.
BibTeX
@inproceedings{yue2020prediction,
  author = {Ling Yue and Dan Hu and Han Zhang and Junhao Wen and Ye Wu and Wei Li and Lin Sun and Xia Li and Jinghua Wang and Guanjun Li and Tao Wang and Dinggang Shen and Shifu Xiao},
  title = {{Prediction of 7-year’s Conversion from Subjective Cognitive Decline to Mild Cognitive Impairment}},
  booktitle = {Human},
  year = {2020},
  doi = {10.1002/hbm.25216},
}
Figure