Hi,
I am looking to calculate the sample size for a clinical trial. The endpoint is median time to an event, where a positive effect would be a longer latency. The event is expected to occur in all individuals & is expected to be left skewed.
The primary outcome will be analysed using quantile regression (adjusted for randomisation strata). Secondary analysis will consider time-to-event using log rank & Cox PH.
A pilot study was conducted in which the median (IQR) time to event was 3.0 (0.5-15.1) in the control arm & 9.1 (4.2-21.3) in the treatment arm.
I am having difficulty estimating the sample size. I can use power twomeans, although this doesn't match my inputs & analysis method. The SD is estimated using Wan et al. 2014. This is my code:
Can anyone advise on the appropriate power method for quantile regression?
I have tried to use simulation using a method I found on this Statalist post. Though I am stuck at how to generate the random inputs. Aside from the median (IQR) & knowing the data is left skewed (it relates to days to discharge from hospital), I don't have clear data on the expected distribution. I have considered using the user written command SKNOR ("ssc install sknor") for random input generation, though this seems to require mean and variance.
Secondly, I want to estimate what hazard ratio (Cox PH) we would be powered to detect (although I appreciate this will only be exploratory). Would the following power code make sense (where n is the sample size calculated for the primary outcome):
Thank you for all your help.
Megan
I am looking to calculate the sample size for a clinical trial. The endpoint is median time to an event, where a positive effect would be a longer latency. The event is expected to occur in all individuals & is expected to be left skewed.
The primary outcome will be analysed using quantile regression (adjusted for randomisation strata). Secondary analysis will consider time-to-event using log rank & Cox PH.
A pilot study was conducted in which the median (IQR) time to event was 3.0 (0.5-15.1) in the control arm & 9.1 (4.2-21.3) in the treatment arm.
I am having difficulty estimating the sample size. I can use power twomeans, although this doesn't match my inputs & analysis method. The SD is estimated using Wan et al. 2014. This is my code:
Code:
power twomeans 3.0 9.1, sd(13)
I have tried to use simulation using a method I found on this Statalist post. Though I am stuck at how to generate the random inputs. Aside from the median (IQR) & knowing the data is left skewed (it relates to days to discharge from hospital), I don't have clear data on the expected distribution. I have considered using the user written command SKNOR ("ssc install sknor") for random input generation, though this seems to require mean and variance.
Secondly, I want to estimate what hazard ratio (Cox PH) we would be powered to detect (although I appreciate this will only be exploratory). Would the following power code make sense (where n is the sample size calculated for the primary outcome):
Code:
power cox, power(0.8) effect(hratio) n(146)
Megan
