Automated classification of adverse events in pharmacovigilance

Shantanu Dev, Shinan Zhang, Joseph Voyles, Anand S. Rao

Abstract

Adverse Events (AEs) are a significant concern in healthcare, since it is among the leading causes of morbidity and mortality[12]. According to the Food and Drug Administration (FDA), between 2006 and 2014, there was a 232% increase in AE cases reported to have caused mortality[13]. In fact, the volume of all AE cases reported to the FDA has increased by almost five fold since 1997[13]. Pharmaceutical companies are struggling to handle the increased case volume due to manual logging of individual cases. This is not a sustainable solution as we see the volume of AE case logs increase exponentially [12,13]. In this paper, we discuss our work and findings for implementing a pharmacovigilance automation solution. This solution explores machine learning techniques in being able to identify serious vs non-serious adverse event narrative logs. While developing our methodology, we explored both traditional machine learning and deep learning techniques. Our final model achieved a mean F1-Score of 95% and an MCC score of 0.80 on the AE case narratives1.
BibTeX
@inproceedings{dev2017automated,
  author = {Shantanu Dev and Shinan Zhang and Joseph Voyles and Anand S. Rao},
  title = {{Automated classification of adverse events in pharmacovigilance}},
  booktitle = {BIBM},
  year = {2017},
  doi = {10.1109/bibm.2017.8217777},
}

Loading related content...

Figure