OII‐A‐4

BACKGROUND Conventional parametric estimates to assess dose proportionality (DP) are sometimes unreliable due to outliers and/or variability from model assumptions. A confidence interval (CI)-based approach using FDA bioequivalence criteria (0.80–1.25) has been proposed to assess DP in contrast to the power-law method (Smith, et al, 2000). Evaluation of robust methods was performed to determine if they improved estimation reliability and statistical inference. METHODS Conventional methods of ordinary least squares regression (OLS) for independent data and mixed-effects procedure for longitudinal data, and robust methods including Huber's Estimation, Least Trimmed Regression (LTS) and Least Absolute Deviation Regression (L1) for independent data and the recently developed Robust Estimation Equations (REE) (Hu, et al, 2001) for longitudinal data were applied to area under the curve (AUC) data from a parallel-group study and a fixed-sequence study. Monte Carlo simulation was used to compare the power in detecting dose proportionality for different estimation approaches. RESULTS The results are shown in Table 1 and Table 2. Table 1. Parallel-group design evaluated by different estimation approaches Methods Slope 90% CI: Lower limit 90% CI: Upper limit Dose proportionality (DP) Note: Acceptance interval for slope is (0.8219, 1.1781). OLS 1.1910 1.0194 1.3680 DP is suggested Huber 1.1002 0.9974 1.2030 DP is suggested LTS 1.1054 0.9895 1.2214 DP is suggested L1 1.1499 1.0427 1.2573 DP is suggested Table 2. Fixed-sequence design evaluated by different estimation approaches Methods Slope 90% CI: Lower limit 90% CI: Upper limit Dose proportionality (DP) Note: Acceptance interval for slope is (0.8219, 1.1781). PROC MIXED 1.2487 1.1829 1.3145 non-DP is documented %REE Macro 1.1451 1.0714 1.2188 DP is suggested CONCLUSIONS Robust methods based on CI estimation tend to lessen the impact of outliers or other departures from normality and could improve dose proportionality assessment. Further research is ongoing to study the magnitude of the power reduction in detecting dose proportionality of conventional (power-law) estimation against robust methods. Clinical Pharmacology & Therapeutics (2005) 79, P4–P4; doi: 10.1016/j.clpt.2005.12.013