In zrmacc/SurvUtils: Utility Functions for Survival Analysis

Utility Functions for Survival Analysis

Zachary R. McCaw
Updated: 2024-08-04

knitr::opts_chunk$set(cache = FALSE, fig.align = "center")

suppressPackageStartupMessages({
  library(dplyr)
  library(SurvUtils)
})

Installation

devtools::install_github(repo = "zrmacc/SurvUtils")

Data Generation

Generates survival data with exponential event times and censoring. Optionally, the subject-specific event rate may depend on a set of covariates and/or a gamma-frailty.

data <- SurvUtils::GenData(
  base_event_rate = 1.0,
  censoring_rate = 0.25,
  n = 100,
  tau = 4.0
)
head(data)

Estimation

One Sample

Kaplan-Meier

• Tabulates the cumulative hazard and survival functions, along with variance estimates and confidence intervals.

km_tab <- SurvUtils::TabulateKM(data)
head(km_tab)

Event Rate, Percentile, Restricted Mean Survival

• Calculate the event rate at a point in time.

# Rate.
SurvUtils::OneSampleRates(data, tau = 1.0)

# Percentile: median.
SurvUtils::OneSamplePercentiles(data, p = 0.5)

# RMST.
SurvUtils::OneSampleRMST(data, tau = 1.0)

Two Sample

Generate Data

data0 <- SurvUtils::GenData(
  base_event_rate = 1.0,
  censoring_rate = 0.25,
  n = 100,
  tau = 4.0
)
data0$arm <- 0

data1 <- SurvUtils::GenData(
  base_event_rate = 0.5,
  censoring_rate = 0.25,
  n = 100,
  tau = 4.0
)
data1$arm <- 1
data <- rbind(data0, data1)

Compare Rates

SurvUtils::CompareRates(data, tau = 1.0)

Compare RMSTs

SurvUtils::CompareRMSTs(data, tau = 1.0)

Compare Cox Models

Compare the predictive performance of Cox models based on different sets of covariates with respect to their c-statistics on held-out data via k-fold cross validation.

# Simulate data.
n <- 1000
x1 <- rnorm(n)
x2 <- rnorm(n)
data <- SurvUtils::GenData(
  covariates = cbind(x1, x2),
  beta_event = c(1.0, -1.0),
  simple = FALSE
)

# Evaluate.
eval <- CompreCoxCstat(
  status = data$status,
  time = data$time,
  x1 = data %>% dplyr::select(x1, x2),
  x2 = data %>% dplyr::select(x1)
)

head(round(eval, digits = 3))

Inference

For a tutorial on influence functions and the perturbation bootstrap, see this vignette.

Plotting

# Generate data.
arm1 <- SurvUtils::GenData(base_event_rate = 0.8)
arm1$arm <- 1
arm0 <- SurvUtils::GenData(base_event_rate = 1.0)
arm0$arm <- 0
data <- rbind(arm1, arm0)

One Sample

Standard Kaplan-Meier

x_breaks <- seq(from = 0.0, to = 4.0, by = 0.50)
data0 <- data %>% dplyr::filter(arm == 0)
fit0 <- Temporal::FitParaSurv(data0)  # Optional parametric fit. 
q_km <- SurvUtils::PlotOneSampleKM(data0, fit = fit0, x_breaks = x_breaks, x_max = 4)
q_nar <- SurvUtils::PlotOneSampleNARs(data0, x_breaks = x_breaks, x_max = 4)
cowplot::plot_grid(
  plotlist = list(q_km, q_nar),
  align = "v",
  axis = "l",
  ncol = 1,
  rel_heights = c(3, 1)
)

AUC

x_breaks <- seq(from = 0.0, to = 4.0, by = 0.50)
data0 <- data %>% dplyr::filter(arm == 0)
q_auc <- SurvUtils::PlotOneSampleAUC(data0, x_breaks = x_breaks, x_max = 4, tau = 3)
q_nar <- SurvUtils::PlotOneSampleNARs(data0, x_breaks = x_breaks, x_max = 4)
cowplot::plot_grid(
  plotlist = list(q_auc, q_nar),
  align = "v",
  axis = "l",
  ncol = 1,
  rel_heights = c(3, 1)
)

Two Sample

x_breaks <- seq(from = 0.0, to = 4.0, by = 0.50)
contrast <- Temporal::CompParaSurv(data)  # Optional parametric fit. 
q_km <- SurvUtils::PlotTwoSampleKM(data, contrast = contrast, x_breaks = x_breaks, x_max = 4)
q_nar <- SurvUtils::PlotTwoSampleNARs(data, x_breaks = x_breaks, x_max = 4)
cowplot::plot_grid(
  plotlist = list(q_km, q_nar),
  align = "v",
  axis = "l",
  ncol = 1,
  rel_heights = c(3, 1)
)

zrmacc/SurvUtils documentation built on Sept. 28, 2024, 8:43 a.m.