Roland Andersson wrote:
<..., as there is probably violation of
proportional hazard due to the difference in age.>
Violation of proportional hazard assumption can be easily detected via
Schoenfeld residuals (please, type - help stcox - in Stata 9/2 SE).
<I wonder if a stratified analysis would be better (with appropriate
age-intervals as strata)...>
In my opinion, a stratified analysis by age would be fine if, accordingly to
the e xistent epidemiological literature on this topic (that, unfortunately,
I do not know)differently aged people experience different hazard functions.
Two other comments come to my mind thinking over your research endeavour:
- provided that it is consistent with the existing epidemiological
literature and with the clinical evidence, wouldn't it be wise to include a
squared age terms in the RHS of Cox regression to detect any turning point
of the failure process?
- would it be interesting to consider categorical variable to tag the
different clinical subtypes of appendicitis?
Kind Regards,
Carlo
-----Messaggio originale-----
Da: [email protected]
[mailto:[email protected]] Per conto di roland andersson
Inviato: martedì 8 settembre 2009 10.59
A: [email protected]
Oggetto: st: How to avoid uncontrolled confounding in Coxregression analysis
I have a large national dataset and plan to analyse socioeconomic
differences in risk of having appendicitis. Persons born after 1949 to
end of 1990 and alive at 1/1/1991 (n=4.500.000) are included in the
dataset. The dataset contain the date when a person was operated for
appendicitis between 1/1/1991 and 31/12/2003 and date of deaths.
Previous collaborators have used SPSS and made Coxregression with time
from 1/1/1991 to the appendicitisdiagnosis or censoring as
analysistime, and have entered age at operation as covariates to
control for the differnces in age.
I am not content with the results and think that there may be
uncontrolled confounding from age, as there is probably violation of
proportional hazard due to the difference in age.
I wonder if a stratified analysis would be better (with appropriate
age-intervals as strata) or if I can use age as analysistime with the
persons coming under observation at different ages.
Is this a correct form to make stset:
stset dateexit [fweight = count], failure(app) enter(time dateenter)
exit(time dateexit) origin(time datebirth) scale(365.25)
with dateenter=1/1/1991
failure event: app != 0 & app < .
obs. time interval: (origin, dateexit]
enter on or after: time dateenter
exit on or before: time dateexit
t for analysis: (time-origin)/365.25
origin: time datebirth
weight: [fweight=count]
I would appreciate your comments.
Roland Andersson, MD PhD
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
