Predicting the Risk of Alzheimer's Disease and Related Dementia in Patients with Mild Cognitive Impairment Using a Semi-Competing Risk Approach

Zhaoyi Chen, Yuchen Yang, Dazheng Zhang, Jingchuan Guo, Yi Guo, Xia Hu, Yong Chen, Jiang Bian

TL;DR

This paper tackles the challenge of predicting Alzheimer's Disease and Related Dementias (AD/ADRD) in individuals with Mild Cognitive Impairment (MCI), a crucial step for early intervention. They introduce a novel semi-competing risk model that accounts for the competing risk of death, a factor often ignored in previous studies, leading to more accurate risk estimations.

What Problem Does This Paper Address?

Alzheimer's disease (AD) and related dementias (AD/ADRD) are a growing public health concern, significantly impacting millions and incurring substantial healthcare costs. A key stage in the progression of these diseases is Mild Cognitive Impairment (MCI), which can precede the full onset of dementia. Early detection and prediction of the transition from MCI to AD/ADRD are vital for timely interventions. However, many existing prediction models fail to account for a critical factor: death. Individuals with MCI are often older and may have co-existing health conditions that increase their risk of mortality. If a patient dies before developing AD/ADRD, they are not included in the AD/ADRD incidence data, and ignoring this "competing risk" can lead to biased estimates of AD/ADRD risk. This paper aims to address this gap by developing a prediction model that explicitly considers death as a competing risk.

Methods

The researchers employed an "illness-death model" to jointly model two events: the transition from MCI to AD/ADRD (a non-terminal event) and death (a terminal event). This approach accounts for the correlation between these two events, which is crucial for accurate risk prediction. They used a semi-competing risk framework, treating AD/ADRD as the event of interest and death as a competing event. The model incorporates frailty parameters to capture unobserved patient-specific heterogeneity that might influence the risk of both events. The analysis was conducted using a Bayesian paradigm, which is computationally efficient for prediction.

Datasets

The study utilized the OneFlorida+ Clinical Research Network dataset. This is a large, longitudinal, and linked real-world data resource containing information on approximately 16.8 million Floridians, including data from Medicaid and Medicare claims, tumor registries, vital statistics, and EHRs. The dataset is a HIPAA-limited data set and adheres to the PCORnet Common Data Model.

Key Insights and Ideas

  • Death as a Competing Risk is Crucial: Ignoring death as a competing risk in AD/ADRD prediction models can lead to biased results and misinterpretations of risk.
  • Semi-Competing Risk Model for Accuracy: The illness-death model, which accounts for the correlation between AD/ADRD onset and death, provides a more accurate estimation of risk compared to standard Cox models that treat death as simple censoring.
  • Identifying Key Risk Factors: The study identified several significant risk factors for AD/ADRD progression, including older age, being Hispanic, depression, diabetes, cerebrovascular diseases, and stroke. These findings align with existing literature, lending credibility to the model.
  • Model Performance Comparison: While the semi-competing risk model theoretically offers advantages, the study found that in this specific dataset, the results were largely similar to those obtained from a standard Cox model. This suggests that for this particular population and outcome, the impact of death as a competing risk might be less pronounced than anticipated, or that the standard Cox model was sufficiently robust.

Key Results

The study identified 35,774 patients with MCI from the OneFlorida+ network, with 33,661 included in the final analysis. Of these, 5,890 (17.5%) developed AD/ADRD, and 27,771 (82.5%) did not. The mortality rate was higher in the AD/ADRD group (23.5%) compared to the no AD/ADRD group (13.5%).

When comparing the semi-competing risk model with the standard Cox model, the hazard ratios (HRs) for most predictors were similar. However, for variables like renal diseases, traumatic brain injury, and vision impairment, the confidence intervals were wider and not statistically significant in the semi-competing risk model, suggesting that accounting for death as a competing risk might alter the interpretation of these associations.

Limitations and Future Work

The authors acknowledge several limitations:

  • Single Competing Risk: Only death was considered as a competing risk. Other significant conditions like various cancers or other types of dementia could also act as competing risks.
  • Data Availability: The study was limited to structured EHR data. Social determinants of health (SDoH) and clinical narratives, which could provide valuable insights, were not readily available.
  • Outcome Definition: Only AD/ADRD onset was modeled. To fully understand progression, intermediate outcomes like neuropsychological test results would be beneficial.
  • Heterogeneity of AD/ADRD: The study grouped various AD/ADRD subtypes. Future work could explore the distinct progression patterns of these subtypes.

Future work suggested includes:

  • Incorporating More Predictors: Adding more predictors and exploring different modeling approaches to improve the semi-competing risk model.
  • Exploring Other Competing Events: Investigating the impact of different types of competing events on AD/ADRD prediction.
  • Intervention Development: Developing and evaluating interventions to reduce the risk of both AD/ADRD and death.
  • Generalizability: Assessing the generalizability of the findings to other populations and healthcare settings.
  • Model Precision and Sensitivity: Further investigating and reporting the precision and sensitivity of the developed model.

Summary updated

Abstract

Alzheimer’s disease (AD) and AD-related dementias (AD/ADRD) are a group of progressive neurodegenerative diseases. The progression of AD can be conceptualized as a continuum in which patients progress from normal cognition to preclinical AD (i.e., no symptoms but biological changes in the brain) to mild cognitive impairment (MCI) due to AD (i.e., mild symptoms but not interfere with daily activities), followed by increasing severity of dementia due to AD. Early detection and prediction models for the transition of MCI to AD/ADRD are needed, and efforts have been made to build predictions of MCI conversion to AD/ADRD. However, most existing studies developing such prediction models did not consider the competing risks of death, which may result in biased risk estimates. In this study, we aim to develop a prediction model for AD/ADRD among patients with MCI considering the competing risks of death using a semi-competing risk approach.
BibTeX
@article{chen2023predicting,
  author = {Zhaoyi Chen and Yuchen Yang and Dazheng Zhang and Jingchuan Guo and Yi Guo and Xia Hu and Yong Chen and Jiang Bian},
  title = {{Predicting the Risk of Alzheimer's Disease and Related Dementia in Patients with Mild Cognitive Impairment Using a Semi-Competing Risk Approach}},
  journal = {Informatics},
  year = {2023},
  doi = {10.3390/informatics10020046},
}
Figure