[S] Relative survival

Therneau, Terry M., Ph.D. (therneau@mayo.edu)
Tue, 27 Oct 1998 08:12:05 -0600


Hakulinen's method for computing an expected surival curve is explicitly
included in the survexp function. For relative survival regression models,
I prefer the exposition of the problem in Barry (1983), The analysis of
mortality by the subject years method, Biometrics, p173-184.
1. Use survexp with the individual=T option to get the predicted
survival of each subject
surv <- survexp(futime ~1, data=mydata, ratetable=survexp.us)
2. Compute the cumulative hazard for the subject
cumhaz <- -log(surv)
3. Use the cumulative hazard as though it were the "time" variable in
poisson regression
fit <- glm(status ~ x1 + x2 + offset(log(cumhaz)), poisson,
data=mydata)

The glm function is fitting the model
observed death rate = (population death rate) * exp(b0 + b1*x1 + ...)

Terry T.

Technical note: You can use the "~1" construct in Splus5.0 if "mydata" has
the variables "age, sex, year" found in the rate table, by those names.
In 3.4 you have to use the ~ratetable(age=var1, year=var2, sex=var3)
construction to explicitly tell survexp the mapping between the ratetable's
variable names and the ones in your data set. I'm not sure about 4.5.

Terry M. Therneau, Ph.D. (507) 284-3694
Head, Section of Biostatistics (507) 284-9542 FAX
Mayo Clinic therneau.terry@mayo.edu
Rochester, Minn 55905
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news