Programming the statistical procedures from SAS

Proc phreg for case-cohort counting process: How to obtain correct baseline survival to create graph

Reply
New User
Posts: 1

Proc phreg for case-cohort counting process: How to obtain correct baseline survival to create graph

Running phreg for case-cohort in SAS 9.4 using the Barlow 1999 methodology (sample code attached). The model must use either counting process or entry= syntax. When the baseline survival estimates are graphed, survival plummets at time t=1 to about .60 - as about 40% of the data are failures. However, the failures occur dispersed over 10 years - the survival estimates in the baseline output dataset are ignoring the entry=  (time of entry) and thus all failures are graphed at t=1 (also attached). Any suggestions on how to obtain correct survival plots?  Diradj group = option and ods survivalplots produce same result. Output = data may have a related issue: survival estimates in the output dataset vary widely with time, thus the graphics do not display correctly (example below).  Suggestions on the web to create explicit risk sets that enter only at the time of failure would seem to be inappropriate for survival plots (haven't tried - but the entry= poses same problem). Method=PL appears not an option with counting process, etc.  Perplexed. Thanks for any suggestions.

Also changed the Barlow methodology slightly to weighting according to Therneau and Li - thus only 1 obs per ID in the dataset. Did not resolve problem. 

 

From the output dataset

Obs group start time survival

422 3 1 65 0.63618
425 3 1 69 0.75759
426 3 1 70 0.88699
429 3 1 75 0.29766 - time=  75 has much lower survival than time =81 for the same group variable
431 3 1 81 0.76885
432 3 1 84 0.90297
434 3 1 89 0.48455
435 3 1 89 0.73385
436 3 1 90 0.73385

 

 

Ask a Question
Discussion stats
  • 0 replies
  • 77 views
  • 0 likes
  • 1 in conversation