Statalisters
I believe that the smoothing behavior for the hazard function plots of
-stcurve- is less than ideal near the time boundaries. Consider the
program given below, which performs two analyses on the -drugtr- data
set from stata-press. Analysis A shows a marked and steady increase in
hazard at times between 13 and 21. Analysis B is the same as analysis A
except that all observations have been censored at time 17. The hazard
curves from these analyses are similar until about time 13. At later
times, hazard A increases sharply while hazard B decreases sharply. I
wonder if the latter behavior is due to the fact that the smoothing
algorithm at time 13 looks at some interval symmetric about 13,
recognizes that there are no failures to the right of 13 but fails to
see that there are no patients at risk at later times either.
I have not attempted to read the -stcurve- code and realize that
devising smoothing algorithms can be non-trivial. I wondered, however,
if the program was really working as the authors intended or it there
might be some way of improving its performance near the time boundaries.
Bill
William D. Dupont phone: 615-343-4100 URL
http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/WilliamDupont
***********************************
* Example of sub-optimal smoothing near a boundary for stcurve
*
log using stcurve.log, replace
set more on
use http://www.stata-press.com/data/r9/drugtr, clear
stset
tabulate died
*
* Analysis A
*
stcox drug, basehc(a)
stcurve, hazard outfile(a, replace)
more
replace died = died & studytime <= 17
replace studytime = min(17, studytime)
stset
tabulate died
*
* Analysis B
*
stcox drug, basehc(b)
stcurve, hazard outfile(b, replace)
more
*
* Graph hazard estimates from both analyses together
*
use b, clear
rename haz1 hazcensored
merge _t using a, sort
sort _t
line haz1 hazcensored _t, xlabel(0(5)30) xtick(0(1)33)
log close
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
