Conditional power calculations for clinical trials with historical controls

Korn, Edward L.; Freidlin, Boris

doi:10.1002/sim.2516

Cited by 23 publications

(27 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The sample size needed for the experimental group is obtained as n e = rT . Further, although the exponential assumption may seem restrictive for determining sample size, this approach is known to work well compared with a more general approach which adopts the proportional hazards assumption and uses a log-rank test (Korn and Freidlin, 2006). …”

Section: Review Of Standard Methodsmentioning

confidence: 99%

“…It requires the (unrealistic) assumption that the observed HC proportion of successes equals the true HC proportion of successes. As a result, many researchers (Kepner and Wackerly, 2001; Lee and Tseng, 2001; Korn and Freidlin, 2006) have demonstrated that this method could not preserve the nominal power and type I error when taking uncertainty about the HC data into account. To facilitate discussion about this uncertainty, we introduce the concepts of conditional power and conditional type I error for historical control studies with survival outcomes.…”

Section: Introductionmentioning

confidence: 99%

“…The above discussion suggests that the power and type I error of a particular HCT should be discussed in a “conditional” sense, given the randomly observed HC data. The uniform or unconditional power approaches (Lee and Tseng, 2001; Korn and Freidlin, 2006) calculate sample sizes so that the means of conditional powers and type I errors over all possible realizations of HC data equal to the desired levels. Such approaches take a hypothetical long-term view, assuming that the HCT will be performed infinite times, each time with a different random realization of HC data, and the averaged performance is considered important.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Sample size considerations for historical control studies with survival outcomes

Zhu

Zhang

Ahn

2016

Journal of Biopharmaceutical Statistics

View full text Add to dashboard Cite

Historical control trials (HCTs) are frequently conducted to compare an experimental treatment with a control treatment from a previous study, when they are applicable and favored over a randomized clinical trial (RCT) due to feasibility, ethics and cost concerns. Makuch and Simon developed a sample size formula for historical control (HC) studies with binary outcomes, assuming that the observed response rate in the HC group is the true response rate. This method was extended by Dixon and Simon to specify sample size for HC studies comparing survival outcomes. For HC studies with binary and continuous outcomes, many researchers have shown that the popular Makuch and Simon method does not preserve the nominal power and type I error, and suggested alternative approaches. For HC studies with survival outcomes, we reveal through simulation that the conditional power and type I error over all the random realizations of the HC data have highly skewed distributions. Therefore, the sampling variability of the HC data needs to be appropriately accounted for in determining sample size. A flexible sample size formula that controls arbitrary percentiles, instead of means, of the conditional power and type I error, is derived. Although an explicit sample size formula with survival outcomes is not available, the computation is straightforward. Simulations demonstrate that the proposed method preserves the operational characteristics in a more realistic scenario where the true hazard rate of the HC group is unknown. A real data application of an advanced non-small cell lung cancer (NSCLC) clinical trial is presented to illustrate sample size considerations for HC studies in comparison of survival outcomes.

show abstract

Section: Review Of Standard Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Sample size considerations for historical control studies with survival outcomes

Zhu

Zhang

Ahn

2016

Journal of Biopharmaceutical Statistics

View full text Add to dashboard Cite

show abstract

“…Rubinstein et al [11] give methodology for calculating the required sample size for randomized studies using the logrank statistic. Korn and Freidlin [12] show how this approach can be extended to one-armed studies compared against historical controls, if the patient data are available.…”

Section: Historically Controlled Studiesmentioning

confidence: 99%

Randomized Phase II Designs

Rubinstein

Crowley

Ivy

et al. 2009

Clinical Cancer Research

125

View full text Add to dashboard Cite

As the use of molecularly targeted agents, which are anticipated to increase overall and progression-free survival (OS and PFS), but not necessarily tumor response, has increased in oncology, there has been a corresponding increase in the recommendation and use of randomized phase II designs. Such designs reduce the potential for bias, existent in comparisons with historical controls, but also substantially increase the sample size requirements. We review the principal statistical designs for historically controlled and randomized phase II trials, along with their advantages, disadvantages, and statistical design considerations. We review the arguments for and against the use of randomization in phase II studies, the situations in which use of historical controls are preferred, and the situations in which use of randomized designs are preferred. We review methods used to calculate predicted OS or PFS values from historical controls, adjusted so as to be appropriate for an experimental sample with particular prognostic characteristics. We demonstrate how adjustment of the type I and type II error bounds for randomized studies can facilitate detection of appropriate target increases in median PFS or OS with sample sizes appropriate for phase II studies. While there continues to be differences among investigators concerning the use of randomization vs. historical controls in phase II trials, there is agreement that each approach will continue to be appropriate, and the optimal approach will depend upon the circumstances of the individual trial.

show abstract

“…Makuch and Simon (1980) developed a sample size formula for HCTs with binary outcomes (Makuch-Simon (M-S) approach). Korn and Freidlin (2006) compared the performance of various approaches to HCT sample size calculation. In assessing the power, however, only the sampling variability from the experimental treatment arm is taken into account, due to the consideration that the measurements from the historical control (HC) arm have been observed before the clinical trial.…”

Section: Introductionmentioning

confidence: 99%

Sample size calculation for studies comparing binary outcomes using historical controls

Zhang

Ahn

2013

Biometrical J

View full text Add to dashboard Cite

In historical control trials (HCTs), the experimental therapy is compared with a control therapy that has been evaluated in a previously conducted trial. Makuch and Simon developed a sample size formula where the observations from the HC group were considered not subject to sampling variability. Many researchers have pointed out that the Makuch-Simon sample size formula does not preserve the nominal power and type I error. We develop a sample size calculation approach that properly accounts for the uncertainty in the true response rate of the HC group. We demonstrate that the empirical power and type I error, obtained over the simulated HC data, have extremely skewed distributions. We then derive a closed-form sample size formula that enables researchers to control percentiles, instead of means, of the power and type I error accounting for the skewness of the distributions. A simulation study demonstrates that this approach preserves the operational characteristics in a more realistic scenario where the true response rate of the HC group is unknown. We also show that the controlling percentiles can be used to describe the joint behavior of the power and type I error. It provides a new perspective on the assessment of HCTs.

show abstract

Conditional power calculations for clinical trials with historical controls

Cited by 23 publications

References 20 publications

Sample size considerations for historical control studies with survival outcomes

Sample size considerations for historical control studies with survival outcomes

Randomized Phase II Designs

Sample size calculation for studies comparing binary outcomes using historical controls

Contact Info

Product

Resources

About