Interval censored data: Cox proportional hazard and surival difference in R

Question

there is quite a lot of information (internet and textbooks) on how to do survival analysis in R with the survival package. But I don't find any information on how to do this when you have left censored data.

---

Problem background:

I have a self constructed data set with published survival data. Usually the event time and the date of the last follow-up (right censoring) is given. There is however one study that only states that the event happened before day 360. So I left censored this data.

What I want to do:

I want to analyse the complete data set with left truncation, events, and right truncation. I want to plot the Kaplan-Meier curve by gender and then

1. do a log-rank test
2. do a Cox regression

What I need:

I am able to create a Surv object with type = interval2. But this does neither allow to calculate survdiff, nor coxph of the survival package.

The intcox package was removed from CRAN and I don't find what I search in the icenReg or interval packages.

---

Can anyone please give me a hind how to solve my problem or where to find practical information on this? I am already spending days on this one.

Many thanks!

Hohenzollern

I believe you might get a better answer on the stats page in stack overflow. Try here: stats.stackexchange.com – Mach

Seems to me that interval2-type censoring would create the same problems as "time-dependent covariates" for which Therneau generally avoids offering "full-time" predictions, since the covariates cannot be assumed to be constant across a span of time and therefore there cannot be a sensible "diff" for any two particular cases. If you want to specify a well-defined set of covariates for two cases with values at all times, then post some R-code that presents them. (The "practical information" to seek would be Therneau's posts in Rhelp Archives.) – Indebted

Answer 1

You can fit a Cox-PH model with both right and left censoring in icenReg by using the ic_sp function. You can fit this using the standard Surv response variable, i.e.

fit <- ic_sp(Surv(L, R, type = 'interval2') ~ treatment, data = myData)

or a little more succinctly with

fit <- ic_sp(cbind(L, R) ~ treatment, data = myData)

Log-rank tests are not available in icenReg, but can be found in the interval package.

Vasti