PHARMACOGENOMICS IMPLEMENTATION: QUANTITATIVE COMPARISON OF EVIDENCE LEVELS FOR PHARMACOGENOMIC- VS. CLINICAL-BASED PERSONALIZED MEDICINE

   Purpose: The evidence requirement for implementing pharmacogenomic-based testing in the clinic is not well defined, whereas clinical decisions based on drug-drug interactions(DDI) are routinely made generally based on limited empirical evidence. The objective of our study was to compare the evidence levels for pharmacogenomic-based warfarin dosing algorithm vs. warfarin-amiodarone DDI.

   Methods: We developed two analogous decision analytic models and conducted value of information(VOI) analyses to quantitatively compare the evidence from two interventions in atrial fibrillation patients: pharmacogenomic-based algorithm for warfarin dosing and dose reduction of warfarin following amiodarone initiation. The key model differences were intervention specific parameters. We used the baseline clinical event rates from ARISTOTLE(main study) trial for pharmacogenomic-based intervention and rates from re-analysis of ARISTOTLE study(analysis of amiodarone use) for DDI-based intervention. Relative risk estimates were taken from a recent meta-analysis of RCTs for pharmacogenomic-model, whereas for DDI-model, from a large observational study.  We used US payer perspective and a lifetime horizon. We estimated the probability of making a non-optimal decision, expected value of perfect information(EVPI) per patient, and EVPI at the population level.

   Results: The relative risk of major hemorrhage for CYPC29 gene variant (exposure for pharmacogenomics) vs wild type was 2.26(95%CI:1.36, 3.75), and the hazard ratio for major hemorrhage related hospitalization following concomitant warfarin and amiodarone(exposure for DDI) therapy was 2.45(95%CI:1.49, 4.02). The treatment effects(prevention of major hemorrhage) for the pharmacogenomic-and DDI-based interventions were 0.60 and 0.41, respectively. Initial simulation results indicate the QALYs improvements were 0.011 and 0.033 for pharmacogenomics- and DDI-based interventions, respectively. The probability of making a non-optimal decision was 15.6 and 10.1 percent for pharmacogenomics- and DDI-based intervention. The EVPI was $119 and $187 per patient and $179 million and $280 million at the AF population level for pharmacogenomics- and DDI-based interventions. The population EVPI for pharmacogenomic-based intervention decreased to $49 million with an assumed test cost of $0.

   Conclusions: The evidence levels for warfarin pharmacogenomics and warfarin-amiodarone DDI appear to be similar. The value of perfect information is higher for DDI because of greater uncertainty in the stroke risk due to dose reduction of warfarin. Our findings suggest that policies for implementation of pharmacogenomics-based testing should be comparable to the DDI-based clinical decisions, which is not the case currently in both clinical and reimbursement guidelines.