Extracting Biomedical Terms from Postpartum Depression Online Health Communities

Online health communities play a vital role in supporting patients by connecting them to people with similar health conditions, thereby enabling interactions that may be distinct from those with their healthcare providers. Online health support groups can provide social support for people experiencing clinical conditions such as postpartum depression (PPD). In this paper, we describe our creation of a dataset of annotated PPD discussion forums and a preliminary assessment of topics that are discussed on the forums. Our approach leverages the capabilities of MetaMapLite (MMLite) and the Human Phenotype Ontology (HPO) concept recognition software to identify biomedical terms from BabyCenter.com online health communities. A data extraction pipeline wherein text from discussion forums on the topic of PPD is scraped and structured for annotation by the MMLite and HPO. The final corpus includes 10,584 posts with their all associated comments. Our analysis of the performance of MMLite to annotate biomedical terms relevant to PPD indicated a precision of 86.7%, recall of 81.3%, and AUC of 0.714. We propose a data model illustrating the main topics discussed among PPD forum users. Topics include: exposures, phenotypes, health conditions, behaviors, and timing. This resource has potential to enable investigating previously unexplored experiences with PPD.