e-HAIL Research in Progress: Automatically Extracting Medication Timelines from Social Media Data

Bipolar disorder (BD) is a life-threatening mental illness characterized by mood swings between depressive lows and (hypo-)manic highs. Those who suffer from the disorder commonly report frustrations in finding effective and tolerable medication regimens, resulting in sustained symptom morbidity despite major advances in evidence-based medicine. Clinical BD-treatment research focuses on a narrowly defined patient profile by implementing strict RCT exclusion criteria. The resulting homogeneous samples are unrepresentative of the real-world BD population. The goal of this project is to enable alternative treatment research methodologies by exploring social media as a source of psychiatric medication data that can help bridge the treatment implementation gap between clinical research and practice.

To explore this problem, we created a dataset of text posts from BD forums on Reddit.  These online communities discuss a wide spectrum of experiences related to BD, and users commonly contextualize their experiences by including a chronological description of their treatment history.  The goal is to automatically extract any self-reported medication timelines from these posts via temporal relation extraction (TRE) and temporal reasoning.

We will discuss the dataset creation, machine learning analyses, and research questions.