Analyze the MRC dataset. The SPSS save file can easily be imported
into R using the Hmisc
spss.get
function. The goal of the analysis
is to assess whether there is an increase in sudden death associated
with one of the treatments. Note that this study was published in BMJ
in 1992 and now a Vanderbilt investigator has added a new variable
(sudden death). Please note that this is real patient data that we
must keep confidential.
As part of your analysis please do the following:
- Read the BMJ paper
- Use a simple command to change the names of the variables to lower case and another command to change underscores in the names to periods. The latter may be handled by modifying the following code:
gsub('_','.',variablenames)
. Note: as of 22Dec04 spss.get
does this automatically.
- Test whether there is an increase in sudden death associated with one of the treatments.
- Use survival methods to assess factors associated with time to sudden death (
dl.death
and sdeath
). Consider treatment (tmtnew
), gender
, age
, and other baseline factors such as EKG results (e.g., bl.ecgm3
). Also, assess baseline heart rate, baseline cholesterol, and smoking status.
- Are there important interactions, such as sex-treatment, or treatment with baseline factors?
- Determine whether the statistical methods in the BMJ paper were appropriate.
- What conclusions would you make regarding these results? Are there subgroups that might be at increased risk or questions that might require another trial?
The important variables are:
sdeath |
Was this a sudden death? 0=no, 1=yes |
tmtnew |
Treatment group 1=Diuretic (moduretic), 2=B-blocker (tenormin), 3=Placebo |
dl.death |
years to death, end of study, or lost to followup |
age |
age in years |
sex |
gender of the patient 0=women, 1=men |
dl.mi |
Years to MI/end of followup |
bl |
baseline |
bl.smoke |
baseline smoking 0=no, 1=yes |
bl.ecgm3 |
vertricular hypertroyphy on ecg 0=no, 2=yes |
bl.ecgm4 |
ST depression on ecg |
bl.chol |
baseline cholesterol |
bl.hr |
baseline heart rate |
Variable and value labels are included in the data file, so you can ignore some of the numeric codes listed above.
Note: The ID for accessing the article and the dataset is
articles
. Please contact
FrankHarrell for the password.