Calibration Curve

Authors

[Editor] Hu Zheng;

[Contributors]

The calibration curve is used to evaluate the consistency / calibration, i.e. the difference between the predicted value and the real value.

Setup

System Requirements: Cross-platform (Linux/MacOS/Windows)
Programming language: R
Dependent packages: data.table; jsonlite; survival; rms; ggplotify

# Install packages
if (!requireNamespace("data.table", quietly = TRUE)) {
  install.packages("data.table")
}
if (!requireNamespace("jsonlite", quietly = TRUE)) {
  install.packages("jsonlite")
}
if (!requireNamespace("survival", quietly = TRUE)) {
  install.packages("survival")
}
if (!requireNamespace("rms", quietly = TRUE)) {
  install.packages("rms")
}
if (!requireNamespace("ggplotify", quietly = TRUE)) {
  install.packages("ggplotify")
}

# Load packages
library(data.table)
library(jsonlite)
library(survival)
library(rms)
library(ggplotify)

Data Preparation

Data frame of multi columns data (Numeric allow NA). i.e the survival data (status with 0 and 1).

# Load data
data <- data.table::fread(jsonlite::read_json("https://hiplot.cn/ui/basic/calibration-curve/data.json")$exampleData$textarea[[1]])
data <- as.data.frame(data)

# convert data structure
res.lrm <- lrm(as.formula(paste(
  "status ~ ", 
  paste(colnames(data)[3:length(colnames(data))], collapse = "+"))),
  data = data, x = TRUE, y = TRUE)

lrm.cal <- calibrate(res.lrm, method = "boot", B = length(rownames(data)))

# View data
head(data)

  time status age sex ph.ecog ph.karno pat.karno meal.cal wt.loss
1  306      2  74   1       1       90       100     1175      NA
2  455      2  68   1       0       90        90     1225      15
3 1010      1  56   1       0       90        90       NA      15
4  210      2  57   1       1       90        60     1150      11
5  883      2  60   1       0      100        90       NA       0
6 1022      1  74   1       1       50        80      513       0

Visualization

# Calibration Curve
p <- as.ggplot(function() {
  plot(lrm.cal,
       xlab = "Nomogram Predicted Survival",
       ylab = "Actual Survival",
       main = "Calibration Curve"
       )
})


n=168   Mean absolute error=0.064   Mean squared error=0.00486
0.9 Quantile of absolute error=0.093