TBD: Development of an adverse drug reaction prediction model using machine learning
Viola Savy Dsouza
Prasanna School of Public Heath, Manipal Academy of Higher Education (MAHE) India
To develop and validate the predictive model to identify patients who are at increased risk for an adverse drug reaction
Data from the tertiary hospital in India was collated and was used to develop an ADR risk score. Variables associated with ADRs were identified using logistic regression analysis and used to compute the ADR risk score.
One of the strongest predictors of ADRs was determined to be the number of medicines, co morbidity (i.e. the presence of two or more diseases), and a previous history of ADR. These variables were utilised in the ADR risk score calculation. We sorted the scores into three categories. 0% to 33% was deemed low risk, 34% to 66% was deemed moderate hazard, and 67% to 99% was deemed high risk. The area under the receiver operator characteristic curve, which measures the score's ability to predict ADRs, was 0.58 (95% confidence interval: 0.44 to 0.63)
Innovative techniques for predicting, detecting, and intervening in adverse drug events might well be based on machine learning. Utilizing existing electronic health information to better identify people at risk for adverse drug reactions may enhance patient care. This study therefore presents a simple and effective strategy for identifying patients who are at increased risk for an ADR. This strategy may be beneficial in clinical practise as a tool to identify high-risk patients and in research to target a population that can benefit from interventions designed to prevent drug-related disease. Future research must contain comprehensive data, and the tool must be incorporated into the health information system.