MedWAS

Phenome- and lab-wide association studies have yielded a wealth of clues about the molecular basis and comorbidities of many common human diseases. These analyses are centered on well-curated sources of electronic health record (EHR) data, from phecodes and laboratory values, respectively. While useful, these methods do not leverage the extensive amounts of prescription medication data available in the EHR. While including prescription medication may accelerate translational research, there is little precedent for their analysis and their utility for genetic analysis. Barriers to including prescription medication data include uneven data quality, and levels of high-dimensionality. We are proposing to extend upon these methods by developing a medication-wide association scan (“medWAS”), that will borrow from the rich and longitudinal prescription data available in EHR. In a first step, we will clean prescription data by creating libraries of common prescriptions for multiple medical and psychiatric conditions. We will do so by mapping prescription medication to RXNorm codes, which should be portable across sites. We will create “medcodes” by aggregating longitudinal prescription data; we will require the presence of prescription medication for a particular RXNorm category on at least two separate occasions. As a proof of concept, we will systematically evaluate medcodes for antidepressants. First, we will calculate the heritability of antidepressant use followed by polygenic risk analyses with clinically defined major depressive disorders from large independent cohorts. Our method has the potential to query relationships between prescription data and other sources of data from the EHR, including genetics, and increase the amount of EHR data to enable discoveries.