Survival Analysis

---

Survival Analysis from CID (2012)

yingtools2 contains functions for running univariate and multivariate cox proportional hazards modeling, but this requires a few dependencies:

library(plyr)
library(dplyr)
library(survival)
library(yingtools2)
data("cid94")

Of note, be sure to load plyr before dplyr

---

1. Univariate Analysis

• If we wanted to assess whether patient age at bmt predicts Enterococcus domination (greater than 30% relative abundance), left censoring up to first stool sample collected, we could create a cox proportional hazards model using:

  stcox(data=pt.cid94,"enterodom30", "agebmt", starttime="firstsampday")
  

---

2. Univariate Analysis with Multiple Predictors

• To perform univariate analyses for a vector of predictors, similarly you could run:

  xvars <-  c("agebmt","sex","primary.dx","priorabx.14d","bmt.tcell.depletion","bmt.cord","fever","vanco_iv","fluoroquinolone","metronidazole","betalactam")
  univariate.stcox(data=pt.cid94, "enterodom30", xvars, starttime="firstsampday") 
  

• Notice how time dependent variables are marked with (td) in the output. Time dependent variables have corresponding columns in the dataframe with the same name with _day added to the variable name which contains the day (relative to time 0) that event occurred. Anytime you run stcox on a dataframe containing a variable and column of the same name with _day at the end, that variable will be treated as time dependent. If you remove the _day column, that variable will no longer be treated as time-dependent.

---

3. Multivariate Analysis

• To test the same set of predictors on Enterococcus domination, then testing variables with pvalues less than 0.2 in a multivariate analysis, you can run:

  univariate.stcox("enterodom30", xvars, starttime="firstsampday", data=pt.cid94, multi = T, multi.cutoff = 0.2) 
  

• Remember, it is important to align all variables by the same time zero before running these kinds of analyses