OBJECTIVES: RWE studies traditionally define exposure as drug name(s), which doesn’t account for the continuum of structural-activity differences between drugs. Despite the utility of QSAR in drug development little work has been done in extending to RWE exposure definitions. Our assessment examine chemical predictors from published QSAR adverse event (AE) models to AE reporting frequencies to show how QSAR based exposure definitions may unlock hidden associations.

METHODS: QSAR models used were antihistamines and beta-blockers with sedation and NSAIDs and Cox-2 inhibitors with myocardial infarction.

EBGM scores, the standard data mining output for AE reporting from the FDA (2022Q1; accessed via a vendor tool) were calculated. Positive EBGM scores are predictive of safety signals.

RESULTS: For antihistamines (lipophilicity cLogP>3) - EBGM scores for sedation ranged from 6.02 to12.7 (median 7.58), for antihistamines (lipophilicity cLogP<3) - EBGM ranged from 1.23 to 7.66 (median2.53).

For non-sedating beta-blockers EBGM scores ranged from 1.02 to 1.03 (median 1.13), for sedating beta-blockers – EBGM ranged from 1.06 to 157.6 (median 1.66).

NSAIDs with increased Cox-2 inhibition - EBGM scores for MI ranged from 1.53 to 6.57 (median 4.91),NSAIDs with lower Cox-2 inhibition - EBGM ranged from 0.42 to 1.86 (median 0.78).

CONCLUSIONS: Overall, despite study limitations, the data shows a correlatory pattern between QSAR predictive models and AE reporting patterns helping to differentiate insights for similar drugs.

While we used FDA safety data for power reasons we see no clear reason why findings would not generalize to healthcare databases. Future RWE generation should when appropriate consider the nature of exposures as continuous, multidimensional variables rather than binary inputs, and, when combined with enriched clinical ontologies, could further help unlock the potential that Machine Learning and Artificial Intelligence can offer for RWE generation where volume of training data and minimizing information loss are key tenets.