Purpose. There is significant interest in developing early prediction models of drug toxicity and utilizing pharmacogenomics to minimize the risk of adverse drug reactions. However, the potential long-term outcomes of such strategies are challenging to evaluate at the time of trial design and evaluation. Our objective is to develop a quantitative, model-based protocol simulation approach for evaluating the clinical and economic effects of adverse drug outcomes related to genetic variation at early stages of drug or test development, using warfarin pharmacogenomics as a case-study. Methods. We implemented a previously published (Hamberg et al., 2007) population pharmacokinetic/pharmacodynamic (PK/PD) model of warfarin distribution and effect that incorporates the effects of genetic variation in the CYP2C9 and VKORC1 genes and other relevant demographic variables. We simulated outcomes (INR distribution) of a non-pharmacogenomic-based warfarin dosing protocol, and plan to simulate various pharmacogenomic-based dosing protocols. We will then integrate these results with methods commonly used in outcomes research, such as discrete event (or continuous time) simulation and probabilistic sensitivity analysis to project the long term economic implications of the various intervention strategies. Results. INRs were modeled for 500 simulated patients using the same patient demographics (median and range) as those reported in the Hamberg analysis. A nomogram that consisted of 5 mg starting warfarin dose with subsequent dosing based on daily INR monitoring (Kovacs et al, 2003), was implemented in the clinical simulation. Baseline INRs were uniformly distributed over a range of 0.9 to 1.3. The INR at day 6 after initiation of therapy ranged from 0.97 to 10.31 with a median of 3.61. Median INR grouped by CYP2C9 expression ranged from 3.17 for *1*1 patients to 5.29 for *3*3 patients. INR variations are linked to the risks of bleeding and stroke, and ultimately to the pharmacoeconomic outcomes of costs and quality-adjusted life years. Conclusions. ‘P-cubed' (P3) modeling is likely feasible when sufficient population PK and PD data are available and valid linkages can be made to long-term clinical and economic outcomes. It may also serve as a tool to explore the robustness of such linkages and probe alternative therapeutic scenarios. Although our findings are preliminary to date, P3-modeling may provide a useful quantitative framework to help inform pharmacogenomic trial design, regulatory decisions, and potentially clinical guidelines and reimbursement policies.