Using sentiment analysis to quantify the relative desirability and acceptability of drug-product attributes

Patient adherence to medication may be influenced by preferences regarding drug-product attributes. Product quality complaints (PQC) and medical information (MI) requests are a potential source of evidence on patients' preferences. The objective of this study was to demonstrate an approach for extracting and categorizing PQC and MI data using natural language processing (NLP) and quantify the relative desirability and acceptability of attributes of oral solid dosage formulations using sentiment analysis.

Data of over 20 000 records of PQC and MI requests from Europe between 2022 and 2023 were extracted and analyzed with the use of pretrained artificial intelligence (AI) and NLP components. First, the data extraction pipelines were built through an iterative frequency analysis process, which involved developing a refined set of key search terms, synonyms, and derivatives relevant to the physical attributes of drug formations. These data pipelines served to create a training set for a transformer-based classification model. Then, sentiment analysis assessed the strength and direction of patient sentiment surrounding each relevant search term.

The highest frequency of communication records and inquiries were related to "swallowability," "palatability," and "texture," with an overlap between swallowability and size categories. Palatability-related attributes elicited the strongest negative sentiment from patients.

This study utilized de novo and pretrained algorithms featuring established AI and NLP and demonstrates the potential of NLP and sentiment analysis in extracting valuable insights about patients' preferences from PQC and MI data.

The findings of this study can have implications on drug development to avoid attributes viewed negatively by patients, thereby improving patients' medication-taking experience, treatment adherence, and satisfaction.