Using artificial intelligence to learn optimal regimen plan for Alzheimer's disease.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Medical Informatics

Background:

• Alzheimer's disease (AD) is a progressive neurological disorder with no cure.
• Managing AD patients requires sophisticated clinical skills due to frequent comorbidities.
• Optimizing treatment regimens for AD patients with coexisting conditions is challenging.

Purpose of the Study:

• To leverage reinforcement learning (RL) to learn clinical decision-making for AD patients.
• To develop an RL-based approach for optimizing treatment regimens using longitudinal electronic health record (EHR) data.
• To assess the performance of RL-generated treatment policies compared to clinician decisions.

Main Methods:

• Utilized data from 1736 patients in the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• Created five data cohorts based on AD, hypertension, and depression status.
• Modeled treatment selection as an RL problem with defined states, actions (medication combinations), and rewards (MMSE scores).

Main Results:

• RL models demonstrated the ability to generate optimal treatment policies.
• Optimal policies (policy iteration, Q-learning) showed improved performance on larger datasets compared to clinician policies.
• Performance varied with dataset size, with RL outperforming clinicians on larger datasets.

Conclusions:

• Reinforcement learning holds significant potential for generating optimal AD treatment plans from longitudinal patient data.
• This research paves the way for RL-based decision support systems to aid in managing AD with comorbidities.
• RL can enhance clinical decision-making for complex patient populations.