The U. S. Veterans Administration lung cancer trial studied 137 males with advanced inoperable lung cancer. The subjects were randomly assigned to either a standard chemotherapy treatment or a test chemotherapy treatment. Several additional variables were also measured on the subjects. The data in the file Veteran have been investigated in Kalbfleisch and Prentice. 21 A brief description of the variables is provided below:
Trt: 1 = standard chemotherapy, 2 = test chemotherapy
Celltype: 1 = squamous, 2 = smallcell, 3 = adeno, 4 = large
Time: survival time in days status: censoring
Status (1 = complete, 0 = censored)
Karno: Karnofsky performance scale index score (100 = good) (This score is used to quantify cancer patients’ functional impairment. The Karnofsky performance score is measured on a 0–100 scale in increments of 10, with 0 indicating that the subject is dead and 100 indicating that the subject shows no signs of the disease.)
Diagtime: months from diagnosis to randomization
Age: in years
Prior: prior therapy (0 = no, 1 = yes)
a. Create a graph with both Kaplan- Meier curves to compare the survival time (use the variable time) for subjects with the standard and the test chemotherapy treatment. What do you observe about the survival probabilities for the two groups of subjects?
b. Conduct the log- rank test and the Wilcoxon test to compare the survival curves of both treatment groups. Interpret the results.
c. It may be beneficial to incorporate health as a variable in the analysis. Patients with low Karnofsky scores are less healthy than patients with high Karnofsky scores. Create four groups with the Veteran data: trt = 1 and Karnofsky score low, trt = 1 and Karnofsky score high, trt = 2 and Karnofsky score low and trt = 2 and Karnofsky score high. Recall that it is often best to keep sample sizes as equivalent as possible when you determine what is a low or high Karnofsky score. Create a Kaplan- Meier curve for each of the four groups. Conduct the log- rank test and the Wilcoxon test to compare the survival curves of the four groups. (While we have only discussed using these tests to compare two groups, they can easily be extended to more than two groups.) Did incorporating health into your analysis impact your conclusions?