Abdulaali Almutairi, PharmD
I have done a retrospective study comparing three arms and the outcome of interest occurred in two arms. I have questions about handling zero outcomes when using Cox-model or log-rank test. I would like to explorer other statistical tests that can help in this situation.
Client: Abdulaali Almutairi, PharmD PhD candidate, Health and Pharmaceutical Outcomes Track
College of Pharmacy, University of Arizona advisor Dr. Ivo Abraham.
Consultants: Dean Billheimer, Chen Chen, Shahin Mohammadi, Mirjana Glisovic-Bensa (author)
October 24, 2019, 1-2pm
Almutairi attached file containing the characteristics of the included sample and the results from the Cox-proportional model and log-rank tests.
The question that he would like the statistical consulting to address is: what is the best method to handle the treatment arm with zero outcomes when comparing three treatment arms.
Almutairi has data sets from SEER (cancer register data set ) and Medicare patients with melanoma. Ipilimumab was approved in in 2011 and Pembrolizumab and Nivolumab were approved in 2014. He is interested in seeing how many patients develop autoimmune disease Colitis.
He explained Inclusion/exclusion criteria: 65+, has Medicare coverage, has colitis. End of censoring time 2016.
Almutairi tred doing CP-Hazards model in SAS but was getting error. Issu was that one of the arms had zero events.
IV. Next Steps
We suggested the following:
-We need to find out what happens if we do not have events, what happens with Cox Proportional Hazards Model. Does it brake? Contact Dr. Hsu look at the textbook for survival analysis. What happens to logrank test.
-Add event on the date that is one after the last censoring time.
-What if we make different assumption of the distribution of times.