survivalContour

General info

Show the predicted survival or cumulative incidence function (competing risks data) over time for single continuous covariate in the form of contour plot or 3D contour plot. The estimate for survival probability and cumulative incidence function(CIF) are based on various survival models. The models that survivalContour supports are listed below. The corresponding R shiny app is availble at survivalContourRshinyapp.

Install

library(devtools)
install_github("YushuShi/survivalContour")

Usage

survivalContour(
  data,
  model,
  contCov,
  D3 = FALSE,
  CI3D = FALSE,
  contCovName = NULL,
  nCovEval = 30,
  otherCov = NULL,
  strataName = NULL
)

data The original data used fit the model.
model The model used to analyze the data.
contCov Specify which continuous covariate is used for the y-axis of the contour plot.
D3 Should "survivalContour" generate a 2D contour plot, or a 3D contour plot.
CI3D Should "survivalContour" generate a semi-transparent layer of confidence intervals for 3D contour plot. This option is only applicable for Cox model or stratified Cox model with or without interval censored data, and parametric or spline models. Fine and Gray model for competing risks data or deep neural network models are not applicable.
contCovName A character string on the name of the continuous covariate shown on the contour plot. If not provided by the user, the printed name will be the covaraite name.
nCovEval Number of covaraite values where predicted survival is evaluated.
otherCov A vector of values of other covariates except for the continuous covariate used to generate contour plot. The default value is NULL, and the mean of continuous covariates and the most frequent values of categorical covariates are used.
strataName The print name of the stratifying covariate used in stratified Cox model. If not provided by the user, "survivalContour" will use the name of the stratifying covariate.

Output

A colored contour plot showing the association between survival time and continuous covariate value.

Details

Model	Package	survivalContour function Input object type	Confidence interval in 3D plot
Cox model	survival	coxph	Y
stratified Cox model	survival	coxph	Y
Cox model with interval censored data	mets	phreg	Y
stratified Cox model with interval censored data	mets	phreg	Y
parametric and spline model	flexsurv	flexsurvreg	Y
Fine and Gray model for competing risks data	riskRegression+prodlim	FRG	N
Fine and Gray model for competing risks data with interval censored data	intccr	ciregic	N
random survival forest	randomForestSRC	rfsrc	N
coxtime,deepsurv,deephit,loghaz,pchazard	reticulate+survivalmodels	pycox	N

Reference

Shi Y, Zhang L, Do KA, Jenq RR, Peterson CB (2023) survivalContour: Visualizing predicted survival for continuous covariate via colored contour plot

Note

The survival contour plot focuses on displaying predicted survival, and does not aim to provide any insights regarding model fit. We recommend users to calculate Harrell's Concordance Index (C-index) to evaluate model fit before sharing results from survivalContour. Complex deep neural network-based models, like deephit or deepsurv, often require significant hand tuning and the best practices for fitting such intricate models is beyond the scope of this paper. Additionally, flexible models may be susceptible to overfitting, so it is always advisable to use an independent dataset for validation.

Examples


library(devtools)
library(mets)
library(flexsurv)
library(prodlim)
library(riskRegression)
library(intccr)
library(reticulate)
library(survivalmodels)
library(survivalContour)
library(survival)
# Cox model without interval censoring
dataPath<-system.file("extdata","veteran.csv",
                      package = "survivalContour")
data<-read.csv(dataPath,row.names = 1)
coxResult<-coxph(Surv(time,status)~PRS+age+diagtime,data=data)

survivalContour(data,coxResult,"PRS")
survivalContour(data,coxResult,"PRS",D3=TRUE)
survivalContour(data,coxResult,"PRS",D3=TRUE,CI3D=TRUE)

paraResult<-flexsurvreg(Surv(time,status)~.,data=data,dist="weibull")
survivalContour(data,paraResult,"karno")
survivalContour(data,paraResult,"karno",D3=TRUE)
survivalContour(data,paraResult,"karno",D3=TRUE,CI3D=TRUE)

splineResult<-flexsurvspline(Surv(time,status)~.,data=data)
survivalContour(data,splineResult,"karno")
survivalContour(data,splineResult,"karno",D3=TRUE)
survivalContour(data,splineResult,"karno",D3=TRUE,CI3D=TRUE)
# stratified data analysis
dataPath<-system.file("extdata","veteran2.csv",
                      package = "survivalContour")
data<-read.csv(dataPath,row.names = 1)
coxResult<-coxph(Surv(time,status)~.-celltype+strata(celltype),data=data)
survivalContour(data,coxResult,"karno")
survivalContour(data,coxResult,"karno",D3=TRUE)
survivalContour(data,coxResult,"karno",D3=TRUE,CI3D=TRUE)
#########################
dataPath<-system.file("extdata","bcdeter2.csv",
                      package = "survivalContour")
data<-read.csv(dataPath,row.names = 1)
coxResult<-phreg(Surv(lower,upper,type="interval2")~cont+treat,data=data)
survivalContour(data,coxResult,"cont")
survivalContour(data,coxResult,"cont",D3=TRUE)
survivalContour(data,coxResult,"cont",D3=TRUE,CI3D=TRUE)


coxResult<-phreg(Surv(lower,upper,type="interval2")~cont+strata(treat),data=data)
survivalContour(data,coxResult,"cont")
survivalContour(data,coxResult,"cont",D3=TRUE)
survivalContour(data,coxResult,"cont",D3=TRUE,CI3D=TRUE)

################
#Fine and Gray model
dataPath<-system.file("extdata","Paquid.csv",
                      package = "survivalContour")
data<-read.csv(dataPath,row.names = 1)
FGResult<-FGR(Hist(time,status)~DSST+MMSE,data=data,cause=1)
survivalContour(data,FGResult,"DSST")
survivalContour(data,FGResult,"DSST",D3=TRUE)

#Fine and Gray interval censored data
dataPath<-system.file("extdata","sim1234.csv",
                      package = "survivalContour")
data<-read.csv(dataPath,row.names = 1)
FGResult<-ciregic(intccr::Surv2(v=lower,u=upper,event=event)~continuousCov,data=data,alpha = c(0, 1), nboot = 0, do.par = FALSE)
survivalContour(data,FGResult,"continuousCov")
survivalContour(data,FGResult,"continuousCov",D3=TRUE)
#######################
dataPath<-system.file("extdata","data1.csv",
                      package = "survivalContour")
data1<-read.csv(dataPath)
data1$num_comorb<-factor(data1$num_comorb)
data1$race<-factor(data1$race)
data1$cancer<-factor(data1$cancer)
data1$diab<-factor(data1$diab)
data1$sex<-factor(data1$sex)
data1$dementia<-factor(data1$dementia)
data1$time<-data1$duration
data1$duration<-NULL
data1$status<-data1$event
data1$event<-NULL
trainModel<-deepsurv(data=data1,frac=0.2,
                     activation="relu",
                     num_nodes=c(64L,64L),
                     dropout=0.2,
                     early_stopping=TRUE,
                     epochs=1000L,
                     patience=50L,
                     batch_norm = TRUE,
                     batch_size=256L,
                     shuffle=TRUE)

survivalContour(data1,trainModel,"resp")
survivalContour(data1,trainModel,"resp",D3=TRUE)

trainModel<-deephit(data=data1,frac=0.2,
                    activation="relu",
                    num_nodes=c(64L,64L),
                    dropout=0.2,
                    early_stopping=TRUE,
                    epochs=1000L,
                    patience=50L,
                    batch_norm = TRUE,
                    batch_size=256L,
                    shuffle=TRUE)

survivalContour(data1,trainModel,"resp")
survivalContour(data1,trainModel,"resp",D3=TRUE)

trainModel<-coxtime(data=data1,frac=0.2,
                    activation="relu",
                    num_nodes=c(64L,64L),
                    dropout=0.2,
                    early_stopping=TRUE,
                    epochs=1000L,
                    patience=50L,
                    batch_norm = TRUE,
                    batch_size=256L,
                    shuffle=TRUE)

survivalContour(data1,trainModel,"resp")
survivalContour(data1,trainModel,"resp",D3=TRUE)

trainModel<-loghaz(data=data1,frac=0.2,
                   activation="relu",
                   num_nodes=c(64L,64L),
                   dropout=0.2,
                   early_stopping=TRUE,
                   epochs=1000L,
                   patience=50L,
                   batch_norm = TRUE,
                   batch_size=256L,
                   shuffle=TRUE)

survivalContour(data1,trainModel,"resp")
survivalContour(data1,trainModel,"resp",D3=TRUE)

trainModel<-pchazard(data=data1,frac=0.2,
                     activation="relu",
                     num_nodes=c(64L,64L),
                     dropout=0.2,
                     early_stopping=TRUE,
                     epochs=1000L,
                     patience=50L,
                     batch_norm = TRUE,
                     batch_size=256L,
                     shuffle=TRUE)

survivalContour(data1,trainModel,"resp")
survivalContour(data1,trainModel,"resp",D3=TRUE)

########################

dataPath<-system.file("extdata","data1.csv",
                      package = "survivalContour")
data1<-read.csv(dataPath)
data1$num_comorb<-factor(data1$num_comorb)
data1$race<-factor(data1$race)
data1$cancer<-factor(data1$cancer)
data1$diab<-factor(data1$diab)
data1$sex<-factor(data1$sex)
data1$dementia<-factor(data1$dementia)
model1<-rfsrc(Surv(duration,event)~.,data1)
survivalContour(data1,model1,"resp",contCovName = "Respiratory rate")
survivalContour(data1,model1,"resp",contCovName = "Respiratory rate", D3=TRUE)

yushushi / survivalcontour Goto Github PK