[R] survival data with competing risks and time-dependent covariates!

[sdzhangping]

Dear friends:

  I’m very interested in the analysis of survival data of leukemic patients, which involves competing risks and time-dependent covariates (infections or relapse of leukemia after BMT). In my present work, I’d like to estimate the risk of chemotherapy, BMT or gvhd on the onset and prognosis of invasive fungal infections (IFI). In detail, the leukemic patients will receive several cycles of chemotherapy and some of them will receive BMT. The patients have three outcomes, that’s acquisition of IFI, death free from IFI or IFI free survival. My puzzle is: 

(1)Dose multi-state model fit this situation? 

(2)I tried to fit coxphf model:  fit = coxphf(Surv(start,stop,ifi==1)~bmt, data=ifi). The system always goes wrong when variable “gvhd” was fitted. [coxphf(Surv(start,stop,ifi==1)~gvhd, data=ifi)]. Is there something wrong with my data format?

Thank you!

Ping Zhang

Mar 26 2010

 

simulation dataset of six patients: 

id

start

stop

ifi

chemo

bmt

gvhd

death

 

1

0

7

0

1

0

0

0

#chemo at d7 and died at d30

1

7

30

2

0

0

0

1

 

2

0

8

0

1

0

0

0

#chemo1 at d7, ifi at d100, chemo2 at 

2

8

100

1

0

0

0

0

# d120, bmt at d200 and die at d450

2

100

120

0

1

0

0

0

 

2

120

200

0

0

1

0

0

 

2

200

450

0

0

0

0

1

 

3

0

6

0

1

0

0

0

#chemo1 at d6, ifi at d44, chemo2 at d60, 

3

6

44

1

0

0

0

0

# censor at d300

3

44

60

0

1

0

0

0

 

3

60

300

0

0

0

0

0

 

4

0

10

0

1

0

0

0

#chemo1 at d10, chemo2 at d60, 

4

10

60

0

1

0

0

0

#bmt at d120, censor at d300

4

60

120

0

0

1

0

0

 

4

120

300

0

0

0

0

0

 

5

0

9

1

0

0

0

0

ifi at d9, die at d30

5

9

30

0

0

0

0

1

 

6

300

301

0

0

1

0

0

#(left censorsed) bmt at d300, gvhd at 

6

301

330

0

0

0

1

0

#330, ifi at d340, die at d350

6

330

340

1

0

0

0

0

 

6

340

350

0

0

0

0

1