A method to assess the proportion of treatment effect explained by a surrogate endpoint

Abstract

Randomized clinical trials are the standard for evaluating new drugs, devices and procedures. Traditional clinical trials entail not only considerable expense, but require considerable time to complete. The use of surrogate endpoints constitutes an effort to control cost and completion time for clinical trials. We propose a method to quantify the proportion of treatment effect explained by a surrogate endpoint based on a general model setting which includes the commonly used linear, logistic and Cox regression models. The interpretation of this quantitative measure is facilitated by graphical displays. To reduce the variability associated with the estimate, a meta‐analytic approach is proposed based on random effects models. An example using real clinical trial data is given to illustrate the proposed procedures. Copyright © 2001 John Wiley & Sons, Ltd.

New-Onset Breast Tenderness After Initiation of Estrogen Plus Progestin Therapy and Breast Cancer Risk

“…Using the method of Li and Meredith, we estimated the proportion of the effect of CEE + MPA on breast cancer risk that was explained by new-onset breast tenderness 24 . Using the incidence estimates of new-onset breast tenderness and the parameter estimates from a Cox proportional hazards model, this formula calculated the proportion of the treatment effect explained by the surrogate marker, new-onset breast tenderness, divided by the overall effect of treatment.…”

Exploring the Relationship Between Surrogates and Clinical Outcomes: Analysis of Individual Patient Data vs. Meta-regression on Group-Level Summary Statistics

“…To control the variability, each study was simulated 100 times to generate the IPD. We considered these two values since 3.2% was the average BMD increase over placebo across all studies (Li et al, 2001) and 8% BMD increase over placebo was used by Wasnich and Miller (2000). Specifically, we applied a Poisson regression model to estimate the association between BMD and fracture risk reduction.…”

“…Clinical trials that reported a reduction in fracture risk also increased BMD significantly, although the fracture risk reduction appeared not to be linearly proportional to the BMD increases. For risedronate, Li et al (2001) conducted an analysis using IPD from the two vertebral fracture trials (Reginster et al, 2000;Harris et al, 1999) and reported that 28% of the vertebral fracture risk reductions was explained by BMD increase. However, the corresponding BMD increases over placebo were 6.2%, 5.8%, and 2.7%, reflecting a wide range of BMD response from agent to agent.…”

“…Extensive statistical research has been conducted to develop advanced methods to validate a surrogate endpoint (Buyse and Molenberghs, 1998;Freedman et al, 1992;Li et al, 2001;Molenberghs et al, 2001). Specifically, a great deal of attention has been put on the meta-analytic techniques to provide more robust and precise estimates (Buyse et al, 2000;Daniel and Hughes, 1997;Li et al, 2001). Two statistical approaches have been used by researchers in the scientific exchange of clinical data: 1) analysis based on individual patient data (IPD) and, 2) meta-regression based on summary statistics.…”

“…Specifically, a great deal of attention has been put on the meta-analytic techniques to provide more robust and precise estimates (Buyse et al, 2000;Daniel and Hughes, 1997;Li et al, 2001). For example, to explore the relationship between BMD increases and fracture risk reductions in the treatment of osteoporosis, analyses based upon IPD have reported between 4% and 28% of the antifracture efficacy was explained by BMD across three antiresorptive agents including alendronate, risedronate, and raloxifene (Cummings et al, 2002;Li et al, 2001;Sarkar et al, 2002). The analysis based on IPD models the relationship between the surrogate and clinical outcome for patients directly and is able to include all relevant patient information including patient-level covariates.…”