Hello,
I have a general question. I conducted a survival analysis using a multivariable Cox regression. Suppose that the variable of interest is dichotomous and time-varying (such as before and after receiving a certain medication). Is it appropriate to use the stcurve command to generate survival curves for before and after taking the medication? This is nearly the exact example from the Stata Journal in 2004 using the Stanford Heart Transplant data (http://www.stata-journal.com/sjpdf.h...iclenum=st0066). However, one of the statisticians who is helping me with the analysis says that such survival curves should only be based on a variable which will remain fixed during the analysis period (i.e. does not change when t>0), and that it is not appropriate to graphically depict these two curves when the variable is time-varying because interpreting such a graph would be difficult. In this situation, she states one should only report the hazard ratio and confidence interval. My questions are:
1. Is it appropriate to generate two survival curves for a time-varying dichotomous variable? Any references which I may show our statistician?
2. If the above is yes, is it still appropriate if subjects switch back and forth between the two groups several times during the analysis period? I am performing another analysis looking at ratios of blood products during a massive transfusion, and therefore a subject can cross a certain ratio threshold several times during the course of receiving the blood products.
3. If it is not appropriate to generate these graphs, is there another way to estimate the baseline risk so that a clinically relevant endpoint (such as number needed to treat) be estimated? A big reason I would like to generate the curves is so that I can report the estimated number needed to treat by the method described in https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1117211/
Thank you to everyone for their help.
Ronald Chang
I have a general question. I conducted a survival analysis using a multivariable Cox regression. Suppose that the variable of interest is dichotomous and time-varying (such as before and after receiving a certain medication). Is it appropriate to use the stcurve command to generate survival curves for before and after taking the medication? This is nearly the exact example from the Stata Journal in 2004 using the Stanford Heart Transplant data (http://www.stata-journal.com/sjpdf.h...iclenum=st0066). However, one of the statisticians who is helping me with the analysis says that such survival curves should only be based on a variable which will remain fixed during the analysis period (i.e. does not change when t>0), and that it is not appropriate to graphically depict these two curves when the variable is time-varying because interpreting such a graph would be difficult. In this situation, she states one should only report the hazard ratio and confidence interval. My questions are:
1. Is it appropriate to generate two survival curves for a time-varying dichotomous variable? Any references which I may show our statistician?
2. If the above is yes, is it still appropriate if subjects switch back and forth between the two groups several times during the analysis period? I am performing another analysis looking at ratios of blood products during a massive transfusion, and therefore a subject can cross a certain ratio threshold several times during the course of receiving the blood products.
3. If it is not appropriate to generate these graphs, is there another way to estimate the baseline risk so that a clinically relevant endpoint (such as number needed to treat) be estimated? A big reason I would like to generate the curves is so that I can report the estimated number needed to treat by the method described in https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1117211/
Thank you to everyone for their help.
Ronald Chang
Comment