East bayesian power calculations
0 likes207 views
PROBLEM - Uncertainty about treatment effects impedes accurate clinical trial design and predictions of success. SOLUTION - A Bayesian approach using East® software allows incorporation of prior information on treatment effects through probability distributions, providing more formal and realistic assessments. BENEFITS - Bayesian calculations like probability of success and predictive power account for uncertainty, leading to more accurate power and decision-making at design and interim stages compared to point estimates. This improves trial success rates.
East bayesian power calculations
- 1. PROBLEM Sample size and power calculations can be highly depen- dent on the assumed magnitude of the treatment effect. Sensitivity analysis is typically performed by checking calcu- lations for a range of potential values. Indeed, East is ideally suited for performing sensitivity analysis of this kind. This ad hoc approach to sensitivity analysis can be comple- mented by a Bayesian approach, which addresses uncer- tainty about the treatment effect in a more formal fashion. The assurance (O’Hagan et al., 2005), or probability of success, is a Bayesian version of power, which corresponds to the unconditional probability that the trial will yield a significant result. Specifically, it is the expectation of the power, averaged over a prior distribution for the unknown treatment effect. This prior distribution expresses the uncertainty about the treatment effect, before the trial began, in terms of the relative plausibility of different parameter values. Depending on one's goals, a prior distribution can take one of many forms. It may be non-informative (e.g., uniform); it may represent the beliefs of an "enthusiastic" or "skeptical" stakeholder, or it may adopt a complex shape that represents diverse opinions from a group of experts. The probability of success is an important consideration for your clinical trial at the design stage. Another Bayesian measure, known as predictive power (Lan et al., 2009) aids decision making at the interim monitoring stage. During the course of a trial, it is often helpful to calculate the condi- tional power: the probability of obtaining a significant result when the trial ends, given the current results. If the condi- tional power is low, the trial may stop early for futility, or there may be an opportunity to re-estimate and increase the sample size. As with computing power in the design stage, the condition- al power calculation depends on the assumed treatment effect, such as an estimate at the interim. However, empiri- cal estimates may not be reliable. Thus, rather than assum- ing a single value for the treatment effect, one could calcu- late conditional power for several different values, and weigh them by the posterior distribution for the treatment effect. PROBLEM - Uncertainty about treatment effects impede clinical trial design SOLUTION - East® allows Bayesian calculations to formally address this uncertainty BENEFITS - Ability to incorporate prior information leads to more accurate predictions of trial success Bayesian Power Calculations: Probability of Success and Predictive Power …assurance should be a major consideration when designing a confirmatory trial. -Chuang-Stein, et al. (2011) Case Study
- 2. East offers the flexibility to use a standard parametric prior distribution, or a user-defined CSV file for more complex prior distributions. Bayesian measures, such as assurance and predictive power, are calculated from these priors. Suppose we are designing a group sequential clinical trial for a weight loss treatment, with the inputs displayed above. In particular, note that we target 90% power to detect a treat- ment difference of 3 kg. This calculation rests on the assumption that the magnitude of the treatment effect (3 kg) is known with certainty. A more realistic scenario is that the true treatment effect lies within some range of possible values. The uncertainty about the treatment effect can be represented, for example, as a Normal distribution with mean 3, and standard deviation of 2. In this more realistic scenario, East shows that the probability of success (72%) is lower than the desired 90% power. Once the trial is underway, the interim monitoring dashboard in East can compare the conditional power calculated at the estimated treatment effect, the predictive power based on a posterior distribution derived from a diffuse prior, and the Bayes predictive power based on a posterior distribution derived from the user-specified prior. The difference in these three estimate are striking and highlight the importance of incorporating prior beliefs into the decision making, especial- ly for futility analysis BENEFITS The Bayesian approach to statistical decision making is becoming increasingly accepted as a valuable way to manage uncertainty in clinical trial design and analysis. One key advantage is the ability to incorporate quantitative prior information to support calcula- tions and decision making. In fact, any prior information about the treatment effect - whether gained from previous trials or from expert opinions - can be accounted for in a power calculation. As the industry standard software for clinical trial design, East continues to incorporate Bayesian and related methodologies to improve clinical success rates. References Chuang-Stein, C., Kirby, S., Hirsch, I., & Atkinson, G. (2011). The role of the minimum clinically important difference and its impact on design- ing a trial. Pharmaceutical Statistics, 10, 250-256. Lan, K., Hu, P., & Proschan, M. (2009). A conditional power approach to the evaluation of predictive power. Statistics in Biopharmaceutical Research, 1, 131-136. O’Hagan, A., Stevens, J.W., & Campbell, M.J. (2005). Assurance in clinical trial design. Pharmaceutical Statistics, 4, 187-201. Bayesian Power Calculations: Probability of Success and Predictive Power SOLUTION www.cytel.com