Cancer survival analysis

[Doug]

If a new treatment appears to improve median overall survival in a group of patients, is there any way to determine whether this improvement results from 'all' those treated surviving a few months longer, or whether it results from only a few individuals benefiting for a much longer time than the median (i.e. basically being 'cured'), while the majority receive no benefit from the treatment? Presumably both scenarios could give a similar apparent survival improvement.

 

[ichbin]

Well, you could construct parameterized models for both scenarios and see which model fit the data better. But I would start by just plotting the survival times of treatment and control groups. Presumably in the few-are-cured scenario, you would see a group of outliers in the treatment group that weren't there in the control group.

 

[Doug]

Well, you could construct parameterized models for both scenarios and see which model fit the data better. But I would start by just plotting the survival times of treatment and control groups. Presumably in the few-are-cured scenario, you would see a group of outliers in the treatment group that weren't there in the control group.

How does one construct such parameterized models - I presume I need help from a statistician, or can this be attempted using a statistics package? Apologies for my complete ignorance!

 

[fed1]

I agree with ICHBIN that you should plot the survival curves and just comment on them, since this is likely to be better understood by audience.

I think that the statement

"improvement results from 'all' those treated surviving a few months longer"

corresponds to a null hypothesis that the Survival curve of the treated and the Survival curve of the untreated are simple location shifts of one and other?

I guess the alternative is they are not simple location shifts.