In elong0527/evacure: Evaluate Prognostic accuracy for mixture cure models

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(ISwR)     # Get melanoma dataset
library(survival)
library(ROCR)
library(evacure)
# devtools::load_all()

# Transfer covariates
data(melanom)
melanom$year <- melanom$days / 365.25
melanom$l.thick <- log(melanom$thick)
melanom$sex1    <- melanom$sex == 2 #( 1 = male)
melanom$ulc1     <- melanom$ulc == 1 #( 1 = present)

median(melanom$year) # median survival time
1 - sum(melanom$status == 1) / nrow(melanom) # Censoring proportion

Kaplan-Meier Curve

# Ulceration
a <- survfit(Surv(melanom$days/365.25, melanom$status == 1) ~ ulc, data = melanom)
plot(a, ylab = "Survival Probability", xlab = "Year", lty = c(2,1), mark.time = T)
legend("bottomleft", lty = c(1,2), legend = c("No","Yes"))

# Tumor thickness
b <- survfit(Surv(melanom$days/365.25, melanom$status == 1) ~ I(l.thick > median(l.thick) ) , data = melanom)
plot(b, ylab = "Survival Probability", xlab = "Year", lty = c(2,1), mark.time = T)
legend("bottomleft", lty = c(1,2), legend = c("Above Median","Below Median"))

CoxPH cure model with K-index

fit <- smcure1( Surv(melanom$days, melanom$status == 1) ~ sex1 + l.thick + ulc1,
               cureform = ~ sex1 + l.thick + ulc1, data = melanom, model = "ph",
               Var = T, em = "smcure"
)
printsmcure1(fit, Var = T)

Cox model with K-index

fit1 <- coxph(Surv(melanom$days, melanom$status == 1) ~ sex1 + l.thick + ulc1, data = melanom, x = TRUE)
risk <- predict(fit1, type = "risk")   # risk from Cox PH model
pi <- rep(1, nrow(melanom) )           # set cure probablity always be 1


k.ind(risk = risk, pi = rep(1, nrow(melanom) ), model = "PH") # K-index