CalculateTimeROC: Calculate Time-Dependent ROC

CalculateTimeROCR Documentation

Calculate Time-Dependent ROC

Description

This function computes the time-dependent Receiver Operating Characteristic (ROC) curve for a given survival model. It uses the 'timeROC' package to compute ROC statistics based on predicted risk scores from the model and actual survival outcomes. This is particularly useful for evaluating the predictive accuracy of survival models at a specified time probability.

Usage

CalculateTimeROC(model, newx, s, acture.y, modelname, time_prob = 0.9)

Arguments

model

A survival model object from which predictions will be made.

newx

A matrix or data frame of new data points for which predictions need to be made.

s

The value of the penalty parameter 'lambda' at which predictions are requested. This can be a specific value or a character string specifying a criterion, such as 'lambda.min' or 'lambda.1se' which are commonly used in models like those from 'glmnet'.

acture.y

A data frame containing the actual survival data, expected to have at least two columns: 'time' which is the survival time, and 'status' which is the event indicator (1 if the event occurred, 0 otherwise).

modelname

A character string specifying the name of the model, used for identification in outputs or plots.

time_prob

A numeric value specifying the quantile of interest for the ROC calculation. The default is 0.9, which calculates the ROC at the 90th percentile of the survival time.

Value

An object of class 'timeROC' that contains the time-dependent ROC curve information.

Examples

# Assuming 'fit' is a Cox proportional hazards model fitted using `coxph` or `glmnet`
new_data <- data.frame(x1 = rnorm(100), x2 = rnorm(100))
actual_outcome <- data.frame(time = rexp(100, rate = 0.1), status = rbinom(100, size = 1, prob = 0.5))
roc_info <- CalculateTimeROC(fit, newx = new_data, s = "lambda.min", acture.y = actual_outcome, modelname = "Cox Model")

IOBR/IOBR documentation built on Nov. 13, 2024, 5:22 a.m.