findcutpoints {CutpointsOEHR}

Find two optimal cutpoints using optimal equal-HR method

Description

Use optimal equal-HR method to determine two optimal cutpoints of a continuous predictor that has a U-shape relationship with survival outcomes based on Cox regression model.

Usage

findcutpoints(cox_pspline_fit, data, nquantile = 100, exclude = 0.05,
  eps = 0.01, shape = "U")

Arguments

cox_pspline_fit

Cox model with psplined x, e.g. coxph(Surv(t,d)~pspline(x,df=0,caic=T),data=test).

data

a dataframe contain survival outcome and a continuous variable which needs to find two optimal cutpoints.

nquantile

an integer; the default value is 100, which means using the 100-quantiles of log relative hazard to find cutpoints.

exclude

a decimals; it is used for excluding extreme values of log relative hazardthe. The default value is 0.05, which log relative hazard values smaller than 5th percentile or larger than 95th percentile are excluded.

eps

a decimals; the default value is 0,01. It restrict the difference between the log relative hazard values of two cadidate cutpoints to be less than 0.01.

shape

a string; equals "U" or "inverseU"

Details

A function to find two optimal cutpoints

Examples

#### Example 1. Find two optimal cutpoints in an univariate Cox model
# Fit an univariate Cox model with pspline
require(survival)
result <- coxph(Surv(t,d)~pspline(x,df=0,caic=TRUE),data=test)
# Visualize the relationship
# Explore whether there is a U-shaped relationship between x and log relative hazard
termplot(result,se=TRUE,col.term=1,ylab='log relative hazard')
# Find two opitmal cutpoints using optimal equal-HR method.
cuts <- findcutpoints(cox_pspline_fit = result, data = test, shape='U')
cuts$optimal # output two optimal cutpoints

#### Example 2. Find two optimal cutpoints in a multivariate Cox model
# Fit a multivariate Cox model with pspline
# The independent variable which is need to find cutpoints should be placed before other covariates.
# To find cutpoints of x, Surv(t,d)~pspline(x)+x1 should be used instead of Surv(t,d)~x1+pspline(x)
require(survival)
result <- coxph(Surv(t,d)~pspline(x,df=0,caic=TRUE)+x1,data=test)
# The rest procedure is the same as example 1
# Visualize the relationship
# Explore whether there is a U-shaped relationship between x and log relative hazard
termplot(result,se=TRUE,col.term=1,ylab='log relative hazard')
# Find two opitmal cutpoints of the first independent variable.
cuts <- findcutpoints(cox_pspline_fit = result, data = test, shape='U')
cuts$optimal # output two optimal cutpoints