Event history analysis helper package

Helper functions for event history analysis with the survival package.

Install

Install from GitHub repository.

library(devtools)
install_github('junkka/ehahelper')

Functions

ggsurv

Make a data.frame of a survfit or coxph object for visualization with ggplot2.

library(ehahelper)
library(survival)
library(ggplot2)
surv_object <- coxph(Surv(time, status) ~ strata(x), data = aml)
ggplot(ggsurv(surv_object), aes(time, surv, color=strata)) + geom_step()

gg_zph

ggplot2 implementation for visualizing scaled Schoenfield residuals form a cox.zph object.

bladder1 <- bladder[bladder$enum < 5, ] 
fit <- coxph(Surv(stop, event) ~ (rx + size + number) * strata(enum) + 
          cluster(id), bladder1)
x <- cox.zph(fit, transform = "identity")
gg_zph(x)

## `geom_smooth()` using method = 'loess'

gg_zph(x, log = TRUE)

## `geom_smooth()` using method = 'loess'

coxme tidyer

Convert coxme objects to tidy format.

library(broom)
library(coxme)
fit <- coxme(Surv(y, uncens) ~ trt + (1|center), eortc)
knitr::kable(tidy(fit, exp = T), digits = 3)

term	estimate	std.error	statistic	p.value	conf.low	conf.high
trt	2.031	0.064	11.03	0	1.791	2.304

fit_g <- glance(fit)
knitr::kable(as.data.frame(t(fit_g)), digits = 3)

	V1
n	2323.000
events	1463.000
Chisq	236.110
df	2.000
logLik	-10478.839
p	0.000
AIC	21015.051
BIC	21166.750
random_n_center	37.000
random_sd_center	0.329
random_variance_center	0.108

coxme augment

Using a coxme model, add fitted values and standard errors to original dataset.

eortc_augmented <- augment(fit, eortc)

knitr::kable(head(eortc_augmented))

	y	uncens	center	trt	.fitted	.se.fit
2	506.1603	1	1	1	0.2037681	0.0184739
3	294.3800	1	1	1	0.2037681	0.0184739
4	383.9152	1	1	0	-0.5048446	0.0457700
5	2441.8338	0	1	0	-0.5048446	0.0457700
6	2442.2923	0	1	0	-0.5048446	0.0457700
7	312.3571	1	1	1	0.2037681	0.0184739

coxme predict

Get predicted vales based on a mixed-effects Cox model, fitted using the coxme package. Extends the standard predict.coxme function by allowing for new data, and by calculating relative risks, either overall or within stratum.

new_data <- data.frame(trt = unique(eortc$trt))

predict_coxme(fit, newdata = new_data, type = "risk")

##         1         2 
## 1.2260138 0.6035994

mcjmigdal/ehahelper