Sample size calculations
Our clients often approach us when the method for obtaining a sample size is unclear or complex. Such computations are as much art as science, often relying on simulations. In calculating a sample size, Statistics Collaborative, Inc. (SCI) takes into account the timing, the frequency, and the method of analysis (including adjustment for multiple looks and multiple outcomes).
SCI works closely with our clients to ensure that the sample sizes we calculate be appropriate to their specific goals and objectives. We discuss reasonable estimates of treatment effects and variability with the clinical team. When past clinical data are limited, we use input from historical data and from experts in the subject matter.
SCI recognizes that trial sponsors are faced with real-world constraints on the sizes of their studies. Both financial resources and study participants may be in short supply. In the study of rare diseases, small trials may be unavoidable.
Examples of SCI's more novel sample-size calculations:
- Understanding uncertainty: We assisted with the design of a large prevention trial whose underlying event rate was likely to be both low and variable across sites. Moreover, the desired efficacy was in flux at the time of the trial because the experimental agent was still in development. For planning purposes, we calculated a range of sample sizes, describing each in terms of the “minimally detectable difference” the difference that would be statistically significant at the specified alpha level if it were observed at the trial's conclusion. This way of looking at sample size across varying background rates and experimental efficacy values allowed our client to consider clinical and statistical significance.
- Balancing efficacy and safety: One of our clients has developed a new formulation of an existing marketed compound. It presents a potentially serious, but rare, safety concern. SCI calculated a sample size to balance efficacy with safety. We recommended a sample size providing sufficient power for non-inferiority analyses of two efficacy outcomes, while also ensuring that the trial be large enough that investigators would have a high probability of detecting the rare adverse event.
- Post-hoc assessments: Taking into account lessons learned in an unsuccessful trial, we recalculated the sample size to allow the client to envision the study that would have been required to demonstrate statistical significance for a smaller treatment effect and a higher prevalence of missing data than initially anticipated.