Machine learning in medicine: a practical introduction to natural language processing.

Area of Science:

• Computational linguistics
• Bioinformatics
• Health informatics

Background:

• Unstructured text data from medical records and patient feedback is valuable for clinical research.
• Natural Language Processing (NLP) techniques transform text into analyzable datasets for statistical and machine learning (ML) models.
• This paper introduces contemporary text-data analysis techniques using freely available software.

Purpose of the Study:

• To provide a practical introduction to modern NLP and ML techniques for analyzing text data.
• To demonstrate the application of these techniques using publicly available drug review data.
• To offer reproducible code for similar research studies.

Main Methods:

• Conducted lexicon-based sentiment analysis on patient reviews of four drugs (Levothyroxine, Viagra, Oseltamivir, Apixaban).
• Employed unsupervised ML (Latent Dirichlet Allocation, LDA) to cluster drugs based on review content.
• Developed and compared three supervised ML algorithms (logistic regression, SVM, ANN) for predicting review sentiment.

Main Results:

• Levothyroxine and Viagra reviews showed higher positive sentiment compared to Oseltamivir and Apixaban.
• LDA identified distinct drug clusters, including mental health medications and contraceptives, with associated themes.
• Supervised ML models achieved classification accuracies for sentiment prediction ranging from 0.664 to 0.720, with SVM performing best.

Conclusions:

• This study offers a conceptual overview of text analysis techniques applicable to large text volumes.
• Reproducible code is provided, enabling researchers to apply these NLP and ML methods in their studies.
• The findings highlight the potential of NLP and ML in extracting meaningful insights from patient-generated text data.