Survival Analysis with visR

Introduction

This tutorial illustrates a typical use case in clinical development - the analysis of time to a certain event (e.g., death) in different groups. Typically, data obtained in randomized clinical trials (RCT) can be used to estimate the overall survival of patients in one group (e.g., treated with drug X) vs another group (e.g., treated with drug Y) and thus determine if there is a treatment difference.

For a more thorough introduction to Survival Analysis, we recommend the following tutorial.

In this example, we will work with patient data from NCCTG Lung Cancer dataset that is part of the survival package. Another vignette presents an example using a data set following the CDISC ADaM standard.

library(ggplot2)
library(visR)

Global Document Setup

# Metadata Title
DATASET <- paste0("NCCTG Lung Cancer Dataset (from survival package ", 
                  packageVersion("survival"), ")")

# Save original options()
old <- options()  

# Global formatting options
options(digits = 3)

# Global ggplot settings
theme_set(theme_bw())

# Global table settings 
options(DT.options = list(pageLength = 10, 
                          language = list(search = 'Filter:'), 
                          scrollX = TRUE))

lung_cohort <- survival::lung

# Change gender to be a factor and rename some variables to make output look nicer
lung_cohort <- lung_cohort %>%  
  dplyr::mutate(sex = as.factor(ifelse(sex == 1, "Male", "Female")))  %>%  
  dplyr::rename(Age = "age", Sex = "sex", Status = "status", Days = "time")

# Restore original options()
options(old)

Cohort Overview (Table one)

Visualizing tables, like the table one or risk tables, is a two-step process in visR . First, a data.frame (or tibble) is created by a get_XXX() function (e.g. get_tableone()). Secondly, the data.frame can be displayed by calling the function render(). The advantage of this process is that data summaries can be created, used and adjusted throughout an analysis, while at every step data summaries can be displayed or even be downloaded.

Populations are usually displayed as a so-called table one. Function get_tableone creates a tibble that includes populations summaries.

# Select variables of interest and change names to look nicer
lung_cohort_tab1 <- lung_cohort %>%  
  dplyr::select(Age, Sex) 

# Create a table one
tab1 <- visR::get_tableone(lung_cohort_tab1)

# Render the tableone
visR::render(tab1, title = "Overview over Lung Cancer patients", datasource = DATASET)

Function render nicely displays the tableone. Additionally, visR includes a wrapper function to create and display a tableone in only one function call.

# Use wrapper functionality to create and display a tableone
visR::tableone(lung_cohort_tab1, title = "Overview over Lung Cancer patients", datasource = DATASET)

Creating and visualizing a tableone with default settings is very simple and can be done with one line of code. However, there are further customization options.

In both the get and the wrapper functions, a stratifier can be defined and the column displaying total information can be removed.

# Create and render a tableone with a stratifier and without displaying the total
visR::tableone(lung_cohort_tab1, strata = "Sex", overall = FALSE,
               title = "Overview over Lung Cancer patients", datasource = DATASET)

visR’s render supports three different rendering engines to be as flexible as possible. By default, render uses gt. Additional engines are datatable (dt) to include easy downloading options…

# Create and render a tableone with with dt as an engine
visR::tableone(lung_cohort_tab1, strata = "Sex", overall = FALSE,
               title = "Overview over Lung Cancer patients", datasource = DATASET, 
               engine = "dt")

…and kable for flexible displaying in various output formats (html by default, latex supported).

# Create and render a tableone with with kable as an engine and html as output format
visR::tableone(lung_cohort_tab1, strata = "Sex", overall = FALSE, 
               title = "Overview over Lung Cancer patients", datasource = DATASET, 
               engine = "kable", output_format="html")

Called with html as an output format, a html view is displayed; called with latex a string containing latex code is printed.

Time-to-event analysis

# Select variables of interest and change names to look nicer
lung_cohort_survival <- lung_cohort %>%  
  dplyr::select(Age, Sex, Status, Days)  

# For the survival estimate, the censor must be 0 or 1
lung_cohort_survival$Status <- lung_cohort_survival$Status - 1

# Estimate the survival curve
lung_suvival_object <- lung_cohort_survival %>% 
  visR::estimate_KM(strata = "Sex", CNSR = "Status", AVAL = "Days")
lung_suvival_object
#> Call: ~survival::survfit(formula = survival::Surv(Days, 1 - Status) ~ 
#>     Sex, data = data)
#> 
#>              n events median 0.95LCL 0.95UCL
#> Sex=Female  90     37    529     376      NA
#> Sex=Male   138     26    840     806      NA

# Create a risktable
rt <- visR::get_risktable(lung_suvival_object)

# Display the risktable
visR::render(rt, title = "Overview over survival rates of Lung Cancer patients", datasource = DATASET)

# Display a summary of the survival estimate
visR::render(lung_suvival_object %>% visR::get_summary(), title = "Summary", datasource = DATASET)

# Display test statistics associated with the survival estimate
visR::render(lung_suvival_object %>% visR::get_pvalue(), title = "P-values", datasource = DATASET)

# Display qunatile information of the survival estimate
visR::render(lung_suvival_object %>% visR::get_quantile(), title = "Quantile Information", datasource = DATASET)

# Display a cox model estimate associated with the survival estimate
visR::render(lung_suvival_object %>% visR::get_COX_HR(), title = "COX estimate", datasource = DATASET)
#> tidyme S3 default method (broom::tidy) used.

# Create and display a Kaplan-Meier from the survival object
gg <- visR::visr(lung_suvival_object)
gg 

# Add a confidence interval to the Kaplan-Meier and display the plot
gg %>% visR::add_CI() 

# Add a confidence interval and the censor ticks to the Kaplan-Meier and display the plot
gg %>% visR::add_CI() %>% visR::add_CNSR(shape = 3, size = 2)

# Add a confidence interval and the censor ticks and a risktable to the Kaplan-Meier and display the plot
gg %>% visR::add_CI() %>% visR::add_CNSR(shape = 3, size = 2) %>% visR::add_risktable()

Competing Risks

In addition to classic right-censored data, the {visR} package supports the estimation of time-to-event outcomes in the presence of competing events. The package wraps the {tidycmprsk} package, and exports functions for cumulative incidence estimation and visualization.

The function estimate_cuminc() estimates the cumulative incidence of the competing event or outcome of interest. The syntax is nearly identical to estimate_KM(); however, the outcome status variable (passed to the CNSR= argument) must be a factor where the first level indicates censoring, the second level the competing event of interest, and subsequent levels are the other competing events. Visualization functions, visr(), add_CI(), add_CNSR(), and add_risktable() share the same syntax as the Kaplan-Meier variants.

visR::estimate_cuminc(
  tidycmprsk::trial,
  strata = "trt",
  CNSR = "death_cr",
  AVAL = "ttdeath"
) %>%
  visR::visr(
    legend_position = "bottom", 
    x_label = "Months from Treatment",
    y_label = "Risk of Death"
  ) %>%
  visR::add_CI() %>%
  visR::add_risktable(statlist = c("n.risk", "cum.event"))
#> Warning: `visr.tidycuminc()` was deprecated in visR 0.4.0.
#> ℹ Please use `ggsurvfit::ggcuminc()` instead.
#> This warning is displayed once every 8 hours.
#> Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
#> generated.