Drug adverse events classification using social media content

Abstract

On-time detection of possible adverse events a drug may have has been a critical issue for the pharmaceutical industry, although it undergoes rigorous clinical trials there still can be adverse effects once it reaches the market, this is known as post-market drug safety surveillance. The ordinary way to collect these was through physicians who prescribe the drug reporting back to the pharmaceutical company. But this process consumes time and has the risk of missing important drug adverse reactions. The recent popularity of social media has led people to communicate extensively about their aspects in day-to-day life, this includes the communications of the experience regarding the drugs and their adverse events. This makes social media a rich resource for monitoring drugs after they reach the market. In this research, we experiment with machine learning models including deep learning models using social media contents manually verified by health care professionals for the presence of drug adverse events. The Social media data has been acquired through popular health care social media channels from their respective APIs. Well-known Text classification algorithms such as SVM and Logistic Regression provide the best accuracy for ADR mining, CNN's which has recently shown high accuracy levels for text classification also shows high levels of accuracy for ADR classification tasks.