Predicting patients' sentiments about medications using artificial intelligence techniques.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• The proliferation of digital health data necessitates advanced methods for analyzing patient feedback on medications.
• Sentiment Analysis (SA) in the medical domain offers valuable insights into patient experiences and treatment effectiveness.

Purpose of the Study:

• To develop and evaluate Artificial Intelligence (AI) models for predicting patient sentiments from medication reviews.
• To explore the impact of different word embedding techniques and classification scenarios on sentiment prediction accuracy.

Main Methods:

• Utilized a large medication review dataset for sentiment analysis across two, three, and ten class scenarios.
• Developed seven Machine Learning (ML) and Deep Learning (DL) models, incorporating Word2Vec and pre-trained embeddings (general and clinical domains).
• Implemented ensemble learning to create a deep ensemble model (DL_ENS) and introduced an interpretability technique.

Main Results:

• The deep ensemble model (DL_ENS) using PubMed and PMC embeddings achieved superior performance, with the highest accuracy (92.96%) and F1-Score (92.27%) in the two-class scenario.
• The DL_ENS model demonstrated strong predictive capabilities across all classification scenarios (two, three, and ten classes).
• The developed model offers transparency in decision-making, enhancing its utility for clinical applications.

Conclusions:

• Combining DL models into a DL_ENS with clinical domain embeddings provides accurate patient sentiment prediction for medications.
• The DL_ENS model serves as a valuable auxiliary tool for clinicians, supporting informed medication prescription.
• This research advances the application of AI in analyzing patient feedback for improved healthcare outcomes.