survreg not appropriate for double-censored data?

Question

I'm consulting with a local quantitative person about some data, it's double-censored. Left censored is below limit of quantitation. Right censored is saturation of the assay. I want to regress it against a simple numeric predictor. I was just told that using survival analysis may not be appropriate because "the censoring is not independent of the outcome". Aren't so-called Tobit models also censored "independent of the outcome", but they can be run via a wrapper function in AES that just calls survreg?

I'm simply confused.

Answer 1

There's no problem with this type of censoring.

A review by Leung et al. explains the potential problems with censoring. In your case, the upper limit (right censoring) is essentially what they call "Type I censoring," in a clinical context "a study in which every subject is under observation for a specified period $C_0$ or until failure." In your case, an actual measurement is equivalent to an observed time "until failure" and your saturation level is equivalent to $C_0$. A similar argument applies to the left censoring of values below the limit of quantitation. These are equivalent to what's considered administrative censoring based directly on study design.

Problems arise when there is censoring due to things like loss to follow up in clinical studies. For example, if someone is close to death, too sick to show up for a clinic visit, and further information isn't available, the fact that the individual was lost to follow up might be considered troublingly informative. But that's not the case in your situation.