Survival analysis question

[Christine Swoboda]

Hello all,

I am preparing to do a survival analysis with data regarding how long it took for patients with a group of rare disease diagnoses to get referred to a specialist. Only 36% of them were even referred to a specialist ever. Would it be appropriate to do a logistic regression for factors related to being referred at all and then a separate survival analysis only among those who were referred? Essentially, truncating the data so that only those with a days from diagnosis to referral value are even examined? This may be interesting because the factors relating to being referred at all may be different from the factors related to how long it took to be referred, and I'm a little wary of doing a cox model with so many failures. Additionally, I don't want to assign a time value for "end of study" and artificially assign those never referred to a certain number of days. Basically I'm just wondering if it would be appropriate for me to just say stset daystoreferral and then stcox sex age race icdcode.... whatever variables I'm interested in but only if referred ==1. If cox is not appropriate, should I use a different test?

Thanks in advance.

[Tiago Pereira]

The multivariable logistic and Cox regression models will assess factors associated with the odds and hazard of being referred, respectively. The Cox model is more appropriate than the logistic model if the follow-up times are not identical across participants and you have censored individuals.

I don't see how an analysis considering only the subset of referred ==1 could offer any meaningful insight regarding the factors associated with referral.