PheWAS

PheWAS performs phenome-wide association analyses on electronic health records and longitudinal medication data to identify therapeutic and adverse drug effects.

Key Features:

• Propensity Score Methods: Incorporates propensity score (PS) methods to mitigate confounding by indication in observational EHR studies.
• Handling Sparse Data: Employs logistic regression techniques, including Firth's penalized maximum likelihood (PML), to manage low exposure and event rates and complete separation.
• PS Adjustment versus Matching: Evaluates PS adjustment relative to PS matching using Monte Carlo simulations, reporting superior power retention with controlled Type I error for PS adjustment.
• Logistic Regression Comparison: Compares Wald's default method to Firth's PML, indicating that Firth's approach yields reliable p-values under complete separation while maintaining controlled Type I error.
• Drug Effect Identification: Identifies latent therapeutic and adverse drug effects in EHR data, demonstrated for pediatric antibiotic exposures such as ampicillin and gentamicin.

Scientific Applications:

• Genotype-Phenotype and Drug-Phenotype Discovery: Extends PheWAS approaches to discover phenotype associations with therapies and adverse drug outcomes beyond genetic variants.
• Clinical Research and Pharmacovigilance: Supports clinical research, personalized medicine, and pharmacovigilance by analyzing medication-outcome associations in EHRs.

Methodology:

Uses Monte Carlo simulations, implements propensity score methods to adjust for confounding by indication, and employs logistic regression approaches including Wald's method and Firth's penalized maximum likelihood to address sparse-data challenges.