Extracting Biomedical Terms from Postpartum Depression Online Health Communities

Area of Science:

• Computational linguistics
• Medical informatics
• Public health

Background:

• Online health communities provide unique social support for patients with postpartum depression (PPD).
• Understanding patient-generated content is vital for improving PPD care.
• Existing methods for analyzing online health discussions often lack specificity.

Purpose of the Study:

• To create and annotate a dataset of online PPD discussion forums.
• To assess the performance of MetaMapLite (MMLite) and Human Phenotype Ontology (HPO) for biomedical term recognition in PPD contexts.
• To identify and model the primary topics discussed by PPD forum users.

Main Methods:

• Scraping and structuring text from BabyCenter.com PPD discussion forums.
• Utilizing MMLite and HPO for automated biomedical term annotation.
• Developing a data extraction pipeline for corpus creation (10,584 posts).
• Evaluating MMLite performance with precision (86.7%), recall (81.3%), and AUC (0.714).

Main Results:

• A comprehensive annotated corpus of PPD online discussions was successfully created.
• MMLite demonstrated high accuracy in identifying PPD-relevant biomedical terms.
• Key topics identified include exposures, phenotypes, health conditions, behaviors, and timing.

Conclusions:

• The developed dataset and annotation methodology provide a valuable resource for PPD research.
• Automated tools like MMLite can effectively analyze online health communities for clinical insights.
• This work facilitates deeper investigation into the multifaceted experiences of PPD.