Generate Cox-model predicted Kaplan-Meier plots

Description

The main purpose of this function is to be used to create plots within the shiny app created by shine_coxph(). For this reason the argument it takes, KM.hat, is created through a process delineated in the example. This can make the function more complicated if you want to use it outside of the shiny app, although it is fully possible to do so.

Usage

cox_KM_plots(KM.hat, clrs = NULL, confint, ylab = "Prob")

Arguments

Value

Plot of predicted survival curve(s)

Examples

library(survival)
# First colon is split into three treatment arms to compare predicted
# survival across arms
split_colon <- split(colon, colon$rx)

colon_arm1 <- split_colon$Obs
colon_arm2 <- split_colon$Lev
colon_arm3 <- split_colon$`Lev+5FU`

# One coxph model is fit for each treatment

colon1ph <- coxph(Surv(time, status) ~sex +  age + obstruct + nodes,
                  colon_arm1, x = TRUE, model = TRUE)

colon2ph <- coxph(Surv(time, status) ~ sex + age + obstruct + nodes,
                  colon_arm2, x = TRUE, model = TRUE)

colon3ph <- coxph(Surv(time, status) ~ sex + age + obstruct + nodes,
                  colon_arm3, x = TRUE, model = TRUE)

# Creating list of models
cox.fit.list <- vector("list", 3)
cox.fit.list[[1]] <- prep_coxfit(colon1ph)
cox.fit.list[[2]] <- prep_coxfit(colon2ph)
cox.fit.list[[3]] <- prep_coxfit(colon3ph)

# Creating new data row for predictions
new.data <- colon[1, ]
# Creating KM.hat object
n.models=length(cox.fit.list)
KM.hat=vector('list',n.models)
lp=rep(NA,n.models)
names(KM.hat)=names(cox.fit.list)
for (i in 1:n.models)
{
 km.hat=predict_one_coxfit(cox.fit.list[[i]],new.data)
 lp[i]=attr(km.hat,'lp')
 sfit=list(time=km.hat$time,surv=km.hat$surv)
 class(sfit)='survfit'
 KM.hat[[i]]=sfit
}
# Plot
cox_KM_plots(KM.hat)

Create table of Cox-model predicted probabilities

Description

Generates tables of predicted probabilities at specified time or vector of times. The KM.hat object contains time and predicted survival probability information as a list of survfit objects.

Usage

cox_times_table(KM.hat, fixTimes = NULL)

Arguments

Details

The main purpose of this function is to be used within the shiny app for the purpose of creating predicted probability tables for user-inputted times. For this reason it is not expressly recommended to use this function outside the context of the shiny app, but it is still possible to do so if desired. The time or vector of times are inputted as characters due to the use of this function in the shiny app, where times are inputted as numbers separated by a comma

Value

Table of predicted probabilities, one column for each time, and one row for each curve

Examples

library(survival)
library(shinyCox)
# First colon is split into three treatment arms to compare predicted
# survival across arms
split_colon <- split(colon, colon$rx)

colon_arm1 <- split_colon$Obs
colon_arm2 <- split_colon$Lev
colon_arm3 <- split_colon$`Lev+5FU`

# One coxph model is fit for each treatment

colon1ph <- coxph(Surv(time, status) ~sex +  age + obstruct + nodes,
                  colon_arm1, x = TRUE, model = TRUE)

colon2ph <- coxph(Surv(time, status) ~ sex + age + obstruct + nodes,
                  colon_arm2, x = TRUE, model = TRUE)

colon3ph <- coxph(Surv(time, status) ~ sex + age + obstruct + nodes,
                  colon_arm3, x = TRUE, model = TRUE)

# Creating list of models
cox.fit.list <- vector("list", 3)
cox.fit.list[[1]] <- prep_coxfit(colon1ph)
cox.fit.list[[2]] <- prep_coxfit(colon2ph)
cox.fit.list[[3]] <- prep_coxfit(colon3ph)

# Creating new data row for predictions
new.data <- colon[1, ]
# Creating KM.hat object
n.models=length(cox.fit.list)
KM.hat=vector('list',n.models)
lp=rep(NA,n.models)
names(KM.hat)=names(cox.fit.list)
for (i in 1:n.models)
{
 km.hat=predict_one_coxfit(cox.fit.list[[i]],new.data)
 lp[i]=attr(km.hat,'lp')
 sfit=list(time=km.hat$time,surv=km.hat$surv)
 class(sfit)='survfit'
 KM.hat[[i]]=sfit
}

# Function takes KM.hat object and a time or vector of times
cox_times_table(KM.hat, fixTimes = "100")

Get confidence intervals for predicted survival curves

Description

Creates confidence levels for plotting predicted survival curves.

Usage

get_confint(p, se, conf.type, conf.int, ulimit = TRUE)

Arguments

Value

list of length two, containing the lower and upper confidence levels

Examples

library(survival)
library(shinyCox)
colondeaths <- colon[colon$etype == 2, ]
split_colon <- split(colondeaths, colondeaths$rx)

colon_arm1 <- split_colon$Obs
colon1ph <- coxph(Surv(time, status) ~ factor(extent) + nodes + strata(surg)
                  + factor(differ),
                  colon_arm1,
                  x = TRUE, model = TRUE)

new.data = cbind.data.frame(`factor(extent)` = 3,
                         `surg` = "surg=0",`factor(differ)` = 2,`nodes` = 5)


coxfit = prep_coxfit(colon1ph)
coxlist = surv_pred_info(colon1ph)

for_ci = predict_se(coxlist, coxfit, new.data)

get_confint(for_ci$surv, for_ci$std.err, conf.int = 0.95,
            conf.type = "log-log")

Wrapper to create survival::coxph() object suitable for shine_coxph()

Description

Performs survival::coxph() with model = TRUE and x = TRUE as defaults. Checks that Cox model is appropriate for use with shine_coxph().

Usage

make_coxph(formula, data, ...)

Arguments

Value

Object of class "coxph" representing the fit

Examples

library(survival)
ovarianph <- make_coxph(Surv(futime, fustat) ~ age + strata(rx),
data = ovarian)

Compute Cox-model predicted survival function

Description

Computes Cox-model predicted survival function for one new data row using coxfit list object created by prep_coxfit().

Usage

predict_one_coxfit(coxfit, newdata)

Arguments

Value

data.frame of predicted survival probabilities over time, one column is time, one is probability

Note

This function's primary use is within the shiny app, where a coxph object is not available. It can be used outside of that context but that is the main purpose of this function, and why it only accepts the return object of prep_coxfit(). In the context of the shiny app, the new data is taken from user inputs.

Examples

# First, fit model using coxph
library(survival)
bladderph <- coxph(Surv(stop, event) ~ rx + number + size, bladder,
model = TRUE, x = TRUE)
# Use coxph object with function
bladderfit <- prep_coxfit(bladderph)
# Take first row of bladder as 'new data'
newdata <- bladder[1, ]
predictions <- predict_one_coxfit(bladderfit, newdata)

Creates predicted survival and standard errors for confidence intervals

Description

Adapted from parts of survival::survfit.coxph(), computes predictions for standard errors based on surv_pred_info() output and newdata from the shiny app.

Usage

predict_se(listsurv, coxfit, newdata)

Arguments

Value

a list of number of subjects for each curve, times at which the curve has a step, number at risk for each time, number of events at each time, number censored at each time (no event but exit risk set), estimated survival, cumulative hazard at each transition, and standard error of the cumulative hazard.

Examples

  library(survival)
  library(shinyCox)
  colondeaths <- colon[colon$etype == 2, ]
  split_colon <- split(colondeaths, colondeaths$rx)

  colon_arm1 <- split_colon$Obs
  colon1ph <- coxph(Surv(time, status) ~
  factor(extent) + nodes + strata(surg) + factor(differ), colon_arm1, x =
  TRUE, model = TRUE)

  new.data = cbind.data.frame(`factor(extent)` = 3, `surg` =
  "surg=0",`factor(differ)` = 2,`nodes` = 5)


  coxfit = prep_coxfit(colon1ph)
  coxlist = surv_pred_info(colon1ph)

  predict_se(coxlist, coxfit, new.data)

Create simplified coxph() object for shiny app

Description

Simplifies coxph() output and checks that predictions match those of the original object

Usage

prep_coxfit(coxph.result, tol = 1e-07)

Arguments

Value

list containing baseline survival estimates, linear predictor estimates, predictor types, coefficient estimates, mean and range of numeric predictors, levels of categorical predictors, strata if any, coxph() formula, table of hazard ratios, table with proportional hazard assumption results, number of subjects, and number of events

Examples

# First, fit model using coxph
library(survival)
bladderph <- coxph(Surv(stop, event) ~ rx + number + size, bladder,
model = TRUE, x = TRUE)
# Use coxph object with function
bladderfit <- prep_coxfit(bladderph)

Generates a shiny app for predictions from Cox model(s)

Description

Writes a shiny app to visualize predicted survival curves from one or multiple Cox models. One feature of this function is that the shiny app, once created, will not contain any identifiable data, containing only information necessary for predictions.

Usage

shine_coxph(..., app.dir = NULL, theme = c("default", "dashboard"))

Arguments

Value

A list containing Cox model information along with the shiny app code. The app is written to the directory while the function is operating.

Notes

There are some requirements in order for this function to run without error: in your original survival::coxph() function or functions, model = TRUE and x = TRUE are required arguments (used to create the simplified "coxph" object). Currently, this function does not support penalized models (e.g., as created by ridge() and pspline()). Multiple strata terms and strata by covariate interaction terms in the formula are also not currently supported, but workarounds are available by respectively using a new strata factor variable encompassing all combinations of desired stratum variable levels. Use of time-varying covariates (e.g. with tt()) and multi-state models is not supported in our function. The package is not intended to support Fine-gray models by survival::finegray() creating Cox models, but doing so will not result in an error.

Guidelines

This package is intended to visualize and present predicted survival functions for fitted Cox models. In regards to formula notation, the variable names used are ultimately what will be displayed in the application. Using functions in the formula will work, but with multiple nested functions it will fail. Using "." notation is not currently supported. The na.action is inherited from the Cox models, with omit being the only option with support at this time. For these reasons, we recommend creating all final variables (including suitable transformations) with meaningful names prior to using survival::coxph().

Examples

  library(survival)

  # Data used is from survival package, renamed for legibility
  names(leukemia)[names(leukemia) == "x"] <- "treatment"
  # Make Cox model, with x = TRUE and model = TRUE
  model1 <- coxph(Surv(time, status) ~ treatment,
  leukemia, x = TRUE, model = TRUE)

  # Use shine_coxph() to create shiny app in temporary directory
  shine_coxph("Model 1" = model1)

  # Get directory for shiny app (should be first, check file list if not)
  filedir <- list.files(tempdir())[1]

  # Run shiny app from temporary directory
  shiny::runApp(paste0(tempdir(), "/", filedir))
  # Remove app from directory once finished
  unlink(paste0(tempdir(),"/",filedir), recursive = TRUE)

Obtains information for standard errors of predictions

Description

Computes necessary information to calculate standard errors and confidence intervals in shiny app. This is adapted from parts of survival::survfit.coxph(). This function is meant to be used in conjunction with predict_se().

Usage

surv_pred_info(model, ctype, individual = FALSE, id, se.fit = TRUE, stype = 2)

Arguments

Value

A list of information needed for computing predicted standard errors.

Examples

library(survival)

colondeaths <- colon[colon$etype == 2, ]
split_colon <- split(colondeaths, colondeaths$rx)
colon_arm1 <- split_colon$Obs

colon1ph <- coxph(Surv(time, status) ~ factor(extent) + nodes + strata(surg)
                   + factor(differ),
                   colon_arm1,
                   x = TRUE, model = TRUE)
surv_pred_info(colon1ph)