INTEGRATING PATIENT GENOMIC PROFILES, MEDICATION RECORDS AND INTERACTION KNOWLEDGE FOR PERSONALIZED MEDICATION SAFETY ANALYTICS - (GENE2DOSE)

Abstract

Adverse drug reactions are a leading cause of avoidable hospital admissions, and a considerable share of them can be traced to two factors that are rarely checked together: interactions between the medicines a patient is taking, and genetic variations that change how the patient's body processes a drug. A patient who is a poor metaboliser through the CYP2C19 enzyme, for example, may receive little benefit from clopidogrel, while a patient with certain HLA alleles may develop a severe skin reaction to carbamazepine. This paper presents Gene2Dose, a data integration and analytics system that combines patient genomic profiles, medication records and drug interaction knowledge to provide personalised medication safety analysis. The system builds a pipeline that ingests pharmacogenomic test results in variant call format, medication orders and dispensing records from the hospital information system, laboratory values such as kidney and liver function, and curated knowledge from public pharmacogenomic guidelines and drug interaction databases. Genetic variants are translated into star-allele diplotypes and metaboliser phenotypes, medicines are mapped to standard drug codes, and all records are linked to a pseudonymised patient identifier in a clinical data warehouse. A knowledge graph connects genes, phenotypes, drugs, enzymes and known interactions so that indirect risks can be traced. On this integrated data, Gene2Dose performs three kinds of analysis. A rule engine applies guideline recommendations to each patient's phenotype and current medicines, flagging drugs that should be avoided or given at an adjusted dose. A drug interaction module checks every pair of active medicines, including interactions that act through a shared metabolising enzyme whose activity is already reduced by the patient's genotype. A gradient boosting model, trained on historical records, estimates the probability that a patient will experience an adverse drug event in the following thirty days, using the number of flagged risks, age, renal function and polypharmacy as features.