PII-072 - LEVERAGING SUPERVISED MACHINE LEARNING TO IDENTIFY PREDICTORS OF ADVERSE EVENTS WITH STIMULANT TREATMENT FROM REAL-WORLD DATA

B. Cao¹, A. Dunn², S. Lee¹, A. Nagar¹; ¹UMB, Baltimore, MD, USA, ²University of Maryland, Baltimore, Baltimore, MD, USA.

Background: We aimed to develop machine learning (ML) models to predict the risk of individual attention-deficit hyperactivity disorder (ADHD) patients experiencing serious clinical outcome on stimulant treatment. Real-world data from the FDA Adverse Event Reporting System (FAERS) database was used, incorporating factors such as patient demographics, medical history, and drug attributes.

Methods: From January 2014 to March 2024, FAERS reports listing drugs containing methylphenidate, amphetamine and lisdexamfetamine as primary suspects were analyzed after removing duplicate reports. A variety of traditional statistical and ML algorithms, such as logistic regression and random forests (RF), were applied to build predictive models for signal detection. To improve accuracy, we used hyperparameter tuning in combination with cross validation to generate the most optimal model. The validated model was used to predict whether a patient would experience any of serious clinical outcome. Serious clinical outcome was defined as death, life-threatening outcome, hospitalization, disability, congenital anomaly, required intervention to prevent permanent impairment/damage, or other serious outcome.

Results: Among all the developed models in this study, the RF ML model demonstrated robust predictive performance with the area under the receiver operating characteristic curve (AUC-ROC) of 0.75, accuracy of 0.72, precision of 0.72, F1 score of 0.79 and recall of 0.88, when applied to the test dataset. The analysis revealed several key risk factors associated with a higher likelihood of serious clinical outcome. These factors included age, different ADHD drugs, gender, taking stimulants for indications other than ADHD such as narcolepsy, rechallenge status, and duration.

Conclusion: To the best of our knowledge, this is the first time an ML approach was used to predict the likelihood of serious clinical outcome from the FAERS database and identify risk factors for individual patients taking various stimulant medicines. This model suggests that children and males are more likely to experience certain serious clinical outcome. The result also supports the current clinical practice guideline which suggests that, if a serious adverse event exists, the prescribers should discontinue the ongoing stimulant treatment or switch to a different stimulant drug.