use http://www.stata-press.com/data/r8/leukemia, clear
stset weeks, fail(relapse)
sts list if treatment1 == 1
* 1.2 At 23 weeks, the survivor function drops from 0.54 to 0.45
* 1.3 Total in the Net Lost column is 12
sts list if treatment1 == 0
* 1.4 8 weeks
* 1.5 None
* 1.6 Median survival before relapse is better on Drug A (23 weeks) than standard treatment (8 weeks)
sts graph, by(treatment1)
graph export graph1.eps, replace
* 1.7 Yes, survival appears to be better on Drug A
sts graph, by(treatment1) yline(0.5)
graph export graph2.eps, replace
sts graph, by(treatment1) yline(0.5) lost
graph export graph3.eps, replace
* 1.9 12 on Drug A, 0 on standard treatment, as before
sts graph, by(treatment1) yline(0.5) lost gwood
graph export graph4.eps, replace
* 1.10 Confidence bands get wider since they are based on smaller numbers
sts test treatment1
* 1.11 Yes, there are far fewer relapses on Drug A than expected
sts test treatment1, wilcoxon
* 1.12 Yes, the results are very similar
sts graph, by(wbc3cat)
graph export graph5.eps, replace
* 2.1 Yes, survival is best in the Normal group and worst in the High group
tab wbc3cat treatment1, co
* 2.2 No, there are more in the High category on standard treatment than Drug A
* 2.3 Survival would be worse in the standard treatment arm
stcox treatment1
* 2.4 HR = 0.22, 95% CI = 0.10, 0.49
xi: stcox treatment1 i.wbc3cat
* 2.5 HR = 0.28, 95% CI = 0.12, 0.66
* 2.6 The beneficial effect of Drug A was exaggerated by the difference in white blood cell counts between the groups
stcoxkm, by(treatment1)
graph export graph6.eps, replace
* 2.7 Yes
stcoxkm, by(wbc3cat)
graph export graph7.eps, replace
* 2.8 Yes
xi: stcox treatment1 i.wbc3cat, sca(sca*) sch(sch*)
stphtest
* 2.9 Yes
stphtest, detail
* 2.10 No, the PH assumption holds for all three variables
sts graph, by(treatment2)
graph export graph8.eps, replace
* 3.1 Survival on drug B is worse than standard treatment for the first 10 weeks
* 3.2 After 10 weeks, survival is better on Drug B than standard treatment
stcoxkm, by(treatment2)
graph export graph9.eps, replace
* 3.3 The observed curves cross, the predicted curves do not
xi: stcox treatment2 i.wbc3cat
* 3.4 No, the HR is 0.85, but this reduction is not statistically significant
drop sca* sch*
xi: stcox treatment2 i.wbc3cat, sca(sca*) sch(sch*)
stphtest
* 3.5 No: the test shows a significant departure from proportional hazards
stphtest, detail
* 3.6 Ony the treatment variable does not satisfy the PH assumption
sts list
gen id = _n
stset weeks, fail(relapse) id(id)
stsplit split_time, at(10)
sts list
* 3.10 Yes
list id weeks relapse split_time _t0 _t in 1/45
gen t1 = treatment2 * (split_time == 0)
gen t2 = treatment2 * (split_time == 10)
xi: stcox t1 t2 i.wbc3cat
* 3.13 HR for the first 10 weeks == 2.13 (95% CI = 0.83, 5.47)
* 3.14 HR for after 10 weeks = 0.11 (95% CI = 0.02, 0.58)
* 3.15 Yes: risk was elevated at first, then reduced
drop sca* sch*
xi: stcox t1 t2 i.wbc3cat, sca(sca*) sch(sch*)
stphtest
* 3.16 Yes, the model is now appropriate
stphtest, det
* 3.17 None of the variables depart from the PH assumption