| Abstrakt: |
Since few medications are equally effective in all patients, physicians can maximize the risk/benefit ratio of therapy for their patients by limiting exposure based on baseline predictors of success. Traditional procedures typically evaluate the response of patients receiving the same treatment regimen without evaluating a comparator. However, when treatments are compared, such as in clinical trials, traditional procedures of identifying predictors must be modified to analyze the treatment effect on the primary outcome variable. We focus on clinical and statistical considerations that arise when developing baseline predictors through models which consider treatment differences. To illustrate an application of this method, we used data from 1,026 patients completing at least 6 months of double‐blind therapy in clinical trials comparing fluoxetine (N=522) with placebo (N=504) for weight loss. Stepwise regression procedures were used to identify baseline variables which were predictive of a beneficial fluoxetine treatment effect on last‐visit‐carried‐forward (LVCF) weight change. In this example, age, smoking activity, and uric acid concentration were the best baseline predictors of long‐term treatment effect relative to LVCF weight change. Patients were more likely to achieve long‐term benefit with fluoxetine if they were older, and/or were nonsmokers, and/or had high concentrations of uric acid at baseline. These predictors, developed through models keying on treatment effect, can be used to identify patients who are more likely to accrue benefits with active therapy beyond those expected with placebo therapy, thus enriching the treatment population so that a higher proportion of treated patients are successful. |
Plný text