Developing and validating a prediction tool for cerebral amyloid angiopathy neuropathological severity

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning

Background:

• Cerebral amyloid angiopathy (CAA) is a cerebrovascular disease diagnosed post-mortem.
• Accurate, non-invasive methods for assessing CAA severity are needed for clinical application.
• Predicting CAA severity is crucial for understanding disease progression and associated risks.

Purpose of the Study:

• To develop and validate machine learning models for predicting CAA severity.
• To create a digitized web-based tool (Florey CAA Score - FCAAS) for clinical use.
• To explore the potential of the FCAAS framework for predicting amyloid-related imaging abnormalities (ARIAs).

Main Methods:

• Developed machine learning models (FCAAS) using an auto-score-ordinal algorithm.
• Validated models on data from three aging and dementia cohort studies.
• Digitized models into a web-based tool and conducted a pilot trial.

Main Results:

• The FCAAS-4 model achieved an AUC-ROC of 0.74 and a Harrell generalized c-index of 0.72.
• Pilot trial using the web tool showed improved performance with AUC-ROC of 0.82 and c-index of 0.79.
• The models demonstrate significant predictive capability for CAA severity.

Conclusions:

• The FCAAS models show promise for predicting CAA severity, enabling clinical risk stratification.
• The developed framework has potential for predicting the development of ARIAs, given the known link between CAA and ARIAs.
• The FCAAS tool offers a valuable approach for non-invasive assessment of CAA severity.