Causality assessment of adverse drug reactions: A machine learning approach

Abstract

Introduction: Tuberculosis, caused by Mycobacterium tuberculosis, has a reported incidence of 2.77 per 1 lakh population in 2022 with a mortality of 0.32 per million. Acquired Immunodeficency Syndrome is caused by Human Immunodeficiency virus (HIV) which is characterised by marked immune suppression resulting in opportunistic infections. There are various adverse drug reactions reported with the anti-tubercular and anti-retroviral therapy. The present study attempts for causality categorization by machine learning algorithm.

Materials and methods: The present study comprises of 60 cases of adverse drug reaction in patient on either anti-tubercular or anti-retroviral therapy. To predict the causality category, a neural network was designed with a single input layer, two middle layer and a output layer. The train model was assessed on test cases to find accuracy of prediction of causality category.

Results: Mean age of cases was 35.92  16.9 yrs. Mean weight of the cases were 50.36  12.8 kgs. The underlying disease were pulmonary tuberculosis 71.7%, MDR tuberculosis 11.7%, extra pulmonary tuberculosis 8.3% and PLHA 8.3%. Out of 60 cases of adverse drug reactions, 2 cases were hospitalised and 1 case died. Various adverse reactions noted were hepatitis (18.33%), peripheral neuropathy (16.67%), rashes (11.67%), vomiting (11.67%), itching (8.3%). Other rare reactions included visual disturbances, psychosis etc. Out of 60 cases,54 cases were of possible causality category and 6 cases were of probable category. The overall accuracy of trained neural network on test cases was 62.5%

Conclusion: Causality assessment can be done by machine learning algorithm, which may help in pharmacovigilance practices.