Predicting User Reported Drug Side Effects Using a Gated Neural Network

The detection of Adverse Drug Events (ADE) or side effects of different drugs are necessary to minimize potential health risks of patients. Given the prevalence of user reported contents on Twitter and various health forums, recent research has focused on the automatic discovery of potential side effects of drugs from these online platforms. However, it is not clear if these reported side effects are solely due to the drug or there are other confounding factors that influence a patient’s experience of side effects. In this work, we seek to characterize the reported side effects along with their severity for different drugs, based on patients’ past drug evaluations and pre-existing medical conditions. We analyze a large dataset collected from an online health community patientslikeme.com, and observe that there exists a strong correlation between a patient’s existing health condition(s) and the side effects she reports across different drugs. We develop a multi-objective deep neural network architecture with gating mechanism, to predict the possible side effects and their overall severity level for a drug, for a given patient. Experimental results from a real world drug evaluations dataset demonstrate the effectiveness of our approach over state-of-the-art baselines. Furthermore, our adaptation of Mixture of Experts approach imbues the network with added explainability, allowing it to justify its predictions of side effects with respect to the evaluated drug, the patient and her conditions.