Oncology Example

Oncology Example

Sample Size for survival analysis

What sample size is required for a study to have 90% power for a comparison of two year survival after surgery for a brain tumor for patients treated with a new combination of chemotherapy and radiation versus those treated with standard methods? The study design calls for randomization to one of two parallel groups and the primary outcome is time to death due to any cause. Accrual occurs uniformly over the two year study, with no losses to follow up expected.

Data will be analyzed using a log-rank test with a two-sided 5% significance level. In previous studies with the standard method about 2/3 of patients survived six months and 20% survived one year following surgery. The new therapy would be considered markedly better if it increased six months survival by about 15%.

nQuery Advisor® can provide sample size and power for standard exponential survival models, but in addition, nQuery Advisor® provides a simulation solution for power of the log-rank test given sample size which allows considerable flexibility in specifications of survival, accrual, and dropout patterns to match previous experience.

• Select File … New, and in the Study Goal and Design Box, select Survival, Two Groups, and Test.
• Select Log-rank test, user specified survival rates, accrual, dropouts simulation.
• Fill in the desired significance level and the number of periods for which survival, accrual, and dropout parameters will be specified
• Select Compute effect size from the Assistants menu or click on the button marked /δ\ .

• Enter end of period times, accrual percentages, hazard rates or proportion expected to survive to the end of the period, and exponential dropout rates for each period.

• Use Edit Row Names to label the rows and click on Plot.

• Select Plot survival vs time to obtain a plot of the expected survival curves.
• You may edit either the plot or the side-table to obtain the desired survival curves.
• When the side table is completed and saved, return to the main table, specify the number of simulated experiments, a random seed for the simulation, and a preliminary choice of sample size.

In this example, we used trial and error with 1,000 simulations to get approximately the desired sample size and then 10,000 simulations to check power to within about ±1%.

For this example, it looks as if 200 patients per group will provide 90% power for the specified survival pattern, and accrual and dropout rates.

nQuery Advisor® provides tables, plots, and standardized sample size justification statements which can be printed directly or copied to clipboard and pasted into your manuscript or proposal.

Plot of Power Vs. n per group for column 4 of previous table