In ndphillips/iwillsurvive: Easily Create and Visualize Survival Models

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

iwillsurvive r packageVersion("iwillsurvive")

The goal of iwillsurvive is to make it easy to estimate and visualize simple survival models. It does this by providing an intuitive functional interface and user-friendly in-line messages, notes, and warnings, while leveraging the gold-standard survival package for all statistical methods.

Installation

iwillsurvive is hosted at https://github.com/ndphillips/iwillsurvive. Here is how to install it:

devtools::install_github(
  repo = "https://github.com/ndphillips/iwillsurvive",
  build_vignettes = TRUE
)

Example

library(iwillsurvive)
library(dplyr)

I'll now give a very brief overview of the basic survival model that iwillsurvive works with. For a more thorough and informative discussion, check out Emily C. Zabor's Survival Analysis in R. It's awesome.

Raw data

We'll start with the cohort_raw dataset which represents the results of a (fictional) clinical trial testing the effectiveness of a drug in extending survival from a patient's first line of therapy start date.

Here are the first 8 patients:

cohort_raw %>%
  slice(1:8) %>%
  knitr::kable()

Here's what the key columns mean:

|Column|Definition| |:-----|:---------| |patientid|A character referring to an individual patient in the form "FXXXXX"| |condition|A character indicating which condition the patient was in, unique values are: r paste(unique(cohort_raw$condition), collapse = ", ")| |lotstartdate|A date indicating when a patient started their first line of therapy after diagnosis (will be used as the index date)| |lastvisitdate|A date indicating the last known date that a patient was alive (will be used as the censor date)| |dateofdeath|A date indicating the date of death of patients who died during the study period (will be used as the event date)|

Research Question

Below is our main research question:

What is the difference in median survival from lot1start to death (or censor) for patients in the placebo versus drug condition?

Survival data

Before we can estimate the survival model, we need to define some key columns:

|Variable|Definition| |:----|:-----| |followup_date|The date at which the event occurs (if known), otherwise the last date the patient was known to be alive| |followup_days|The number of days from indexdate to followupdate| |eventstatus|A logical column indicating whether or not the patient died. TRUE = Yes, FALSE = No.

To calculate these variables, we can use iwillsurvive's derive functions: Use the derive_*() functions to calculate key derived columns:

• followup_date: dateofdeath, if known, and censordate, otherwise
• followup_days: Days from index_date (in our case, lotstartdate) to followup_date
• event_status: A logical column indicating whether or not the event (dateofdeath) is known.

cohort <- cohort_raw %>%
  derive_followup_date(
    event_date = "dateofdeath",
    censor_date = "lastvisitdate"
  ) %>%
  derive_followup_time(index_date = "lotstartdate") %>%
  derive_event_status(event_date = "dateofdeath")

Here's how the new columns look for the first 8 patients:

cohort %>%
  select(patientid, followup_date:event_status) %>%
  slice(1:8) %>%
  knitr::kable()

Fitting survival models

Use iwillsurvive() to fit the survival model. We'll set the follow up time to be followup_days and specify "condition" as a term (i.e.; covariate) to be used in the model.

cohort_iws <- iwillsurvive(cohort,
  followup_time = "followup_days",
  terms = "condition",
  event_title = "Death",
  index_title = "LOT1 Start"
)

print method

Print the object to see summary information:

cohort_iws

knitr::include_graphics("https://raw.githubusercontent.com/ndphillips/iwillsurvive/master/inst/figures/print_iwillsurvive.png")

Plotting followup times

Use plot_followup() to visualize the observed follow-up times for each patient ordered by the length of their follow-up and colored by their event status (not by condition)

plot_followup(cohort_iws)

Plotting Kaplan-Meier curves (the plot method)

Use plot() to plot the Kaplan-Meier survival curve. If you don't include any arguments, you'll get the 'default' curve options.

plot(cohort_iws)

You can fully customize the look of your Kaplan-Meier curve (see ?plot.iwillsurvive) to see all the optional arguments:

plot(cohort_iws,
  add_confidence = FALSE,
  add_median_delta = FALSE,
  censor_pch = 3,
  censor_size = 5,
  legend_position_x = c(600, 400),
  legend_nudge_y = c(.25, .3),
  median_flag_nudge_y = .15,
  anchor_arrow = TRUE,
  palette = "Dark2",
  title = "My Title",
  subtitle = "My Subttitle",
  risk_table_title = "My Risk Table Title"
)

Understanding iwillsurvive objects

The iwillsurvive() function returns an object of class iwillsurvive. Internally, it is a list containing many objects from the original data, to a survival object:

names(cohort_iws)

The .$data object contains the original data

cohort_iws$data

The .$fit object contains the survival object (created using the survival::survfit() function)

cohort_iws$fit

The .$fit_summary object contains summary information:

cohort_iws$fit_summary

iwillsurvive and the survival package

iwillsurvive uses the survival package under the hood for all model estimation. For that reason, you should always be able to get the 'same result' using the survival package as you would using the iwillsurvive package.

For example, here's how to directly replicate the same result we got using survival:

library(survival)

# Fit the model
fit_survival <- survival::survfit(
  survival::Surv(followup_days, event_status,
    type = "right"
  ) ~ condition,
  data = cohort
)

# Print method
fit_survival

ndphillips/iwillsurvive documentation built on June 4, 2024, 3 p.m.