Readme file
SERIES
B
 Statistical
Methodology
Analysis of failure time data under competing censoring mechanisms,
A. Rotnitzky, A. Farall, A. Bergesio and D. Scharfstein
J. R. Statist. Soc. B, Volume 69, (2007), 307–327
To run the data analysis in Section 7:
a) in Splus run the commands in the file "construction_of_variables". These commands generate a file with
extension .mat which contains data matrices with composite variables used in the analysis. The
commands call the S-plus files actg175 and SINTO which contain the original datasets
The variables called by the commands in file "construction_of_variables" are
From data base "actg175":
pid=patient identifier (disguised identifier, NOT the true one on the original dataset)
ivdrug=indicator of intravenous drug use
karnof=Karnofsky score
age= age of patient
d.onrx=date of start of treatment
d.offrx=date of off treatment
d.endaid= date of aids
d.death=date of death
cd4base=baseline cd4
d.marker= date cd4 measured
markvalu= time dependent cd4
From data base "SINTO":
pid= patient identifier
VISITDT=date of visit
TOXCODE=toxicity code
TOXDT, RESOLDT=start and end dates of symptom
b) in Matlab run the commands ACTGCONSINT and ACTGSINSINT. These commands compute the estimators
of P(T>t*) as in the paper; in ACTGCONSINT estimation is conducted under models for censoring that
include the covariate "time dependent symptom", in ACTGCONSINT the censoring models do not include
time dependent symptoms. The commands call functions in the files ESTIPI2, ESTIMAPI3, ESTIMAR2
y ESTIMAR3.
Andrea Rotnitzky
Department of Biostatistics
Harvard School of Public Health
655 Huntington Avenue
Boston
MA 02115
USA
E-mail: arotnitzky@utdt.edu
|
