survfit.risk: Non-parametric absolute risk estimation

In skoval/coxph.risk: Computes absolute risk for cohort data

Description

Method to estimate a fully non-parametric absolute risk in the presence of multiple competing events.

Usage

survfit.risk(begin, end, survfit1, ...)

Arguments

begin	vector with each row specifying the beginning time in the [begin, end) of the projection interval
end	vector with each row specifying the ending time in the [begin, end) of the projection interval
survfit1	a survfit.object for the primary event
...	additional survfit.objects for competing events

Details

The non-parametric approch assumes that the time-to-events of all individuals in the sample in which the event times were observed follow the same distribution. This is equivalent to assuming that the hazard ratio for any two individuals of the sample is 1 for all time points.

Value

A vector of the absolute risk of the primary event occurring within [begin, end).

Author(s)

Stephanie Kovalchik <[e-mail]>

See Also

coxph.risk

Examples

data(mgus)

# Mayo Clinic 20-35 year follow-up of patients with 
# monoclonal gammopathy of undetermined significance (MGUS)

# Hazard models of multiple myeloma, death, other plasma malignancy
# Time scale is days from MGUS diagnosis

surv1 <- survfit(Surv(time, status)~1,data=mgus2,subset=event=="myeloma")
surv2 <- survfit(Surv(time, status)~1,data=mgus2,subset=event=="death")
surv3 <- survfit(Surv(time, status)~1,data=mgus2,subset=event=="other")

# Absolute risk predictions for multiple myeloma consecutive biyearly intervals
bytwo <- seq(0,9,by=2)
intervals <- cbind(bytwo, bytwo+2)*365.25
risk <- survfit.risk(intervals[,1],intervals[,2], surv1, surv2, surv3)
risk

# How much do competing risks of death and other plasma malignancy reduce net risk?
netrisk <- survfit.risk(intervals[,1], intervals[,2], surv1)
netrisk

names(risk) <- paste("[",bytwo,", ",bytwo+2,")",sep="")

dotchart(risk*100, xlim = range(c(risk, netrisk)*100), 
                    xlab = "risk of multiple myeloma progression (%)",
                    ylab = "projection interval")
                    
points(x=netrisk*100, y=1:length(risk), col="red")

legend("topleft",bty="n",c("Net Risk", "Abs. Risk"), pch=1,col = c("red", "black"))


# How does this compare with cumulative risk?
intervals <- cbind(rep(0, length(bytwo)), bytwo+2)*365.25
risk <- survfit.risk(intervals[,1], intervals[,2], surv1, surv2, surv3)
netrisk <- survfit.risk(intervals[,1], intervals[,2], surv1)

names(risk) <- paste("[0, ",bytwo+2,")",sep="")

dotchart(risk*100, xlim = range(c(risk, netrisk)*100), 
                    xlab = "risk of multiple myeloma progression (%)",
                    ylab = "projection interval")
                    
points(x=netrisk*100, y=1:length(risk), col="red")

legend("topleft",bty="n",c("Net Risk", "Abs. Risk"), pch=1,col = c("red", "black"))

### STRATIFIED BY GENDER

surv1 <- survfit(Surv(time, status)~1,data=mgus2,subset=event=="myeloma"&sex=="male")
surv2 <- survfit(Surv(time, status)~1,data=mgus2,subset=event=="death"&sex=="male")
surv3 <- survfit(Surv(time, status)~1,data=mgus2,subset=event=="other"&sex=="male")

# 5-year, 10-year RISK OF MULTIPLE MYELOMA
intervals <- cbind(c(0,0), c(5,10))*365.25
male.risk <- survfit.risk(intervals[,1], intervals[,2], surv1, surv2, surv3)

male.risk

skoval/coxph.risk documentation built on May 30, 2019, 1:07 a.m.