Analysis of randomized comparative clinical trial data for personalized treatment selections

Abstract

Suppose that under the conventional randomized clinical trial setting, a new therapy is compared with a standard treatment. In this article, we propose a systematic, 2-stage estimation procedure for the subject-level treatment differences for future patient's disease management and treatment selections. To construct this procedure, we first utilize a parametric or semiparametric method to estimate individual-level treatment differences, and use these estimates to create an index scoring system for grouping patients. We then consistently estimate the average treatment difference for each subgroup of subjects via a nonparametric function estimation method. Furthermore, pointwise and simultaneous interval estimates are constructed to make inferences about such subgroup-specific treatment differences. The new proposal is illustrated with the data from a clinical trial for evaluating the efficacy and toxicity of a 3-drug combination versus a standard 2-drug combination for treating HIV-1-infected patients.

Fig. 1.

(a): The estimated density function of the parametric score with respect to week 24 CD4 changes; (b): Estimated group averages of week 24 CD4 changes over the score for 2- and 3-drug combination groups; (c): Estimated treatment differences (thick curve), 3-drug combo minus 2-drug combo, with respect to week 24 CD4 changes over the score and the corresponding 95% pointwise (dashed curve) and simultaneous (shaded region) confidence intervals.1

Fig. 2.

(a): The estimated density function of the parametric score with respect to week 24 HIV-RNA; (b): Estimated averages of week 24 RNA over the score for 2- and 3-drug combination groups; (c): Estimated treatment differences (thick curve), 3-drug combo minus 2-drug combo, with respect to week 24 RNA over the score and the corresponding 95% pointwise (dashed curve) and simultaneous (shaded region) confidence intervals.

Fig. 3.

Summary of the simulation results for (a) n₀ = n₁ = 500 and (b) n₀ = n₁ = 5000. Left panel: the empirical average of the estimated treatment difference (solid black curve) versus the true treatment difference (solid gray curve). Shown also are the 45o reference line. Middle panel: the estimated (gray line) and empirical (black line) standard error. Right panel: empirical coverage level.