How to conduct competing-risk based on proc surveyphreg using sas 9.4

[yuzheng20]

proc surveyphreg data= data varmethod = TAYLOR;
strata SDMVSTRA;
cluster SDMVPSU;
weight weight6;
model PERMTH_int*dead(0) =CLASS4/ eventcode=1 rl;
run;

[OsoGris]

I don't see an EVENTCODE option on the MODEL statement in PROC SURVEYPHREG.

[PamG]

@Rick_SAS I am also wondering if there is a way in SAS to implement competing risk for survey data?  I will be working on survey data and there doesn't seem to be much information online. 

[OsoGris]

unfortunately, PROC SURVEYPHREG does not have the EVENTCODE= option (like PROC PHREG does) to enable a competing risks analysis.  You could try using PHREG with a WEIGHT statement; the parameter estimates would be the same but not the standard errors as from SURVEYPHREG. 

[Season]

I also ran into your problem. Unfortunately, I conducted a search on the Internet and did not find a statistical method tailored for dealing with competing risks in complex survey data analysis. I did found a couple of papers dealing with competing risks in the setting of clustered data, but not complex survey data specifically. A recently published article on the comparison of the methods was also found: Systematic comparison of approaches to analyze clustered competing risks data | BMC Medical Research.... An R package is also available.

I think a naive approach based on @OsoGris's suggestion might be of some help: apply the survey weights via the WEIGHT statement in the PHREG procedure and manually calculate the variance of the regression coefficients by the jackknife method by strictly following its work flow (i.e., (1) calculate the regression coefficients on the entire sample at first, (2) delete a primary sampling unit in turn and adjust survey weights on the remaining sample, (3) calculate the regression coefficients on each sample thus formed and (4) apply the formula of estimating variances via the jackknife method to obtain the final result). The process is lengthy but workable. The possibly only (and a big!) question underlying this procedure is the lack of theoretical support.

Friends in this Community can also kindly offer their help if they have any ideas on this problem.

[Season]

Update on my reply: hours after posting this thread, I found not only that I could no longer  change its contents but also solutions to this problem.

An insidious tactic of finding the solution lies in correctly specifying the keyword in the search engine. I used "competing risks" and virtually retrieved no result on my problem. However, when substituting the words for "cause-specific hazard", a subdomain of competing risks analysis, things worked.

To summarize my finding, there is a research paper (Population-based absolute risk estimation with survey data | Lifetime Data Analysis) that discusses this issue. From the first sentence of its abstract (i.e., "Absolute risk is the probability that a cause-specific event occurs in a given time interval in the presence of competing events."), I came to know "absolute risk", a term whose meaning puzzled me before, could also serve as a keyword in search engines in this setting. I then found a book entitled Absolute Risk: Methods and Applications in Clinical Management and Public Health (Chapman & Hall/CRC..., whose cover contained a picture depicting competing risks. It was exciting to find in the book that the intersection of complex survey data analysis and competing risks modeling was introduced as a few sections in this book, including an independent section on variance estimation. You- and of course, I- can refer to it for more details.