Detecting Adverse Drug Reaction (ADR) with data mining and predicting its intensity with machine learning

Abstract

Adverse Drug Reaction (ADR) is one of the many uncertainties which are considered as a fatal threat in the field of pharmacy and medical diagnosis. Utmost care is taken to test a new drug thoroughly before it is introduced and made available to the public; but these pre-clinical trials are not enough on their own to ensure safety. Many ADRs are discovered in the later stages of consumption which could not be found out during the pre-clinical trials. The increasing concern to the ADRs has motivated the development of statistical, data mining and machine leaning methods to detect the Adverse Drug Reactions. With the availability of electronic health Records (EHRs) it has become possible to detect ADRs with the mentioned technologies. In this work, we have proposed a hybrid model of data mining and machine learning to identify different Adverse Reactions and predict the intensity of the final outcome. We have used the Proportionality Reporting Ratio (PRR) along with the precision point estimator test called the Chi-Square test to mine the different associations between drug and symptoms called the drug-ADR association. This output from the data mining technique is used as an input to the machine learning algorithms such as Random Forest and Support Vector Machine (SVM) to predict the intensity of the final outcome of ADR, depending on a patient’s demographic data such as gender, weight, age, etc. We have performed the analysis on a total count of 88000 data taken from the publicly available dataset of FDA and achieved an accuracy of 91% to predict ‘death’ as the final outcome from an ADR.