How to run a bunch of models in R using a Function

Question

library(survival)
justices <- read.csv("http://data.princeton.edu/pop509/justices2.csv")
PREDS = c("age", "year")
m = coxph(Surv(tenure, event == 1) ~ age + year, data = justices)
summary(m)

exp(coef(m)[1])
exp(confint(m,level=(1-0.05/1))[1,])


DVMOD <- function(PREDS, data){
  t <- coxph(paste0("Surv(tenure, event == 1) ~ "), PREDS + number + name, data = data)
  return((c(PREDS, coef(t)[1], confint(t)[1,])))
}

all_models <- lapply(PREDS,DVMOD, PREDS = PREDS, data=justices)

I wish to run separate coxph model for each variable in PREDS and then store the name of that variable with its hazard ratio and confidence bands.

Where is PREDS defined outside the DVMOD function as a list of variables? — Len Greski, May 15 '20 at 22:43
Have you looked into tidymodels? I understand that is the tidy approach to running a bunch of models. — Mark Neal, May 16 '20 at 07:41

Len Greski · Accepted Answer · 2020-05-16T00:38:49.443

If you cast the formula string with as.formula() you can change independent variables in the DVMOD() function as follows:

DVMOD <- function(PREDS, data){
     theFormula <- paste("Surv(tenure, event == 1) ~ ",PREDS," + number + name")
     t <- coxph(as.formula(theFormula), data = data)
     return((c(PREDS, coef(t)[1], confint(t)[1,])))
}
DVMOD("age",justices)
all_models <- lapply(c("age"),function(x,y){
     DVMOD(x,y)
},justices)

Since the model for year does not converge (i.e. it fails when calling DVMOD() directly with year as the value of PREDS), the lapply() with 2 variables fails, but it works with age.

...and the output:

> all_models
[[1]]
                                    age               2.5 %              97.5 % 
              "age"  "24.4400841925434" "-20.8849057264629"  "69.7650741115497" 

>

How to run a bunch of models in R using a Function

1 Answers1