/ADR_prediction

Demo

Primary LanguageHTML

Prediction of adverse drug reaction outcomes using FDA Adverse Event Reporting System Data (FAERS)

According to Chan et. al. (Br J Clin Pharmacol. 2016): “Adverse drug reactions (ADRs) are a major cause of morbidity and mortality, and leading cause of hospital admission. ADRs negatively affect patients' quality of life and confidence in medications, which consequently leads to worse treatment outcomes”. Our model will be valuable for healthcare professionals by providing insights about ADRs to improve drug prescriptions.The Food and Drug Administration (FDA) database of adverse drug reaction (FAERS) is a database which contains information on adverse event and medication errors which submitted to FDA by consumers, healthcare professionals or manufacturers. The adverse drug reactions (ADR) outcome includes death, hospitalization, life-threatening, disability, congenital anomaly, and/or other serious outcome. This data has been quarterly reported from 2004 to 2019 and for each quarter, 7 datasets are available: DEMO (patient demographic and administrative information), DRUG (drug/biologic information), REAC (adverse events), OUTC (patient outcomes), RPSE (report sources), THER (drug therapy start and end dates), and INDI (indications for use for the reported drugs) making it 900 files in total. For the purpose of demonstration, I only looked at data reported in Q4 of 2018 with ~300,000 reported outcomes. Moreover, there is a similar dataset for Animals ADRs recently made publicly available. The techniques used in human ADR prediction may be transferrable to animal dataset as well. If you are interested to use ADR dataset, you can try:

  1. To use machine learning models to predict ADR outcome from other features available in the dataset and find the most predictive features.
  2. Investigate if the outcome predictor can be improved for specific indications such as cardiovascular diseases.
  3. Each patient in the data may have a sequence of outcomes. Is it possible to use these sequential patterns to improve of the prediction?
  4. Is it possible to utilize other biological/chemistry information to cluster the drugs into subtypes and correlate it with ADR outcomes? For instance, investigating if drugs with specific chemical structure are significantly correlated with particular outcomes.