survMTP.ncc: Accuracy measures for a risk prediction marker under nested...

In mdbrown/survMarkerTwoPhase: Evaluate the prognostic accuracy of a marker under two phase designs

Description

This function estimates the AUC, TPR(c), FPR(c), PPV(c), and NPV(c) for for a specific timepoint and marker cutpoint value c with survival data from a nested case-control subcohort design.

Usage

  survMTP.ncc(time, event, marker, 
                           subcoh, 
                           id, 
                           data,
                           risk.sets, 
                           estimation.method = "NP", 
                           predict.time,
                           alpha = 0.05, 
                           Npert = 500, 
                           marker.cutpoint = 'median' )

Arguments

time	time to event
event	indicator for the status of event of interest. event = 0 for censored observations, and event = 1 for event of interest.
marker	marker variable of interest
subcoh	indicator for subjects included in the subcohort sample (1=included, 0=not included)
id	unique numeric identifier for each observation.
data	data frame in which to look for input variables. This should be the full cohort data, including time, event, subcoh, id, and marker values when subcoh = 1.
risk.sets	a data frame or matrix identifying risk sets. Specifically this should have one row for each case (event==1), consisting of the case id, followed by corresponding matched id's. See below for a more concrete example.
estimation.method	should non-parametric ('NP') or semi-parametric ('SP') estimates be calculated. Semi-parametric methods assume a proportional hazards model.
predict.time	numeric value of the timepoint of interest for which to estimate the risk measures
alpha	alpha value for confidence intervals. (1-alpha)*100% confidence intervals are provided. default is alpha = 0.05.
Npert	number of perturbations to use when calculating standard errors. Default is 500.
marker.cutpoint	numeric value indicating the value of the cutpoint 'c' at which to estimate 'FPR', 'TPR', 'NPV' or 'PPV'. default is 'median' which takes cutpoint as the marker median.

Value

a list with components

estimates	point estimates for risk measures
se	standard errors for risk measure estimates
CIbounds	(1-alpha)*100 % confidence bounds for measures.
model.fit	*only returned if estimation.method = 'SP'* object of type 'coxph' from fitting the model coxph(Surv(time, event)~Y)
marker.cutpoint, estimation.method, predict.time	function inputs

References

1. Liu D, Cai T, Zheng Y. Evaluating the predictive value of biomarkers with stratified case-cohort design. *Biometrics* 2012, 4: 1219-1227.

2. Pepe MS, Zheng Y, Jin Y. Evaluating the ROC performance of markers for future events. *Lifetime Data Analysis.* 2008, 14: 86-113.

3. Zheng Y, Cai T, Pepe MS, Levy, W. Time-dependent predictive values of prognostic biomarkers with failure time outcome. *JASA* 2008, 103: 362-368.

4. Cai T. and Zheng Y . Resampling Procedures for Making Inference under Nested Case-control Studies. *JASA* 2013 (in press).

5. Cai T and Zheng Y/*, Evaluating Prognostic Accuracy of Biomarkers under nested case-control studies. *Biostatistics* 2012, 13,1, 89-100.

(* equal contributor and corresponding author).

Examples

#simulated data for illustration 
data(SimData)

# use the function 'ccwc' from the 
# epi package to generate a subcohort. 
require('Epi')

#2 matched controls for each case
nmatch = 2

cohortData_ncc <- SimData
cohortData_ncc$id = 1:dim(cohortData_ncc)[1]

subcohort_ncc <- ccwc(exit = survTime, fail= status, data = cohortData_ncc, controls = nmatch) # match 2 controls for each case. 

#indicator for inclusion in the subcohort
sampleInd = rep(0, nrow(cohortData_ncc)) #initialize all equal to zero
sampleInd[subcohort_ncc$Map] = 1

cohortData_ncc$subcohort = sampleInd  

table(sampleInd) #subcohort sample size of 352

#marker data is unavailable for those not in the subcohort
cohortData_ncc$Y[sampleInd==0] = NA  

#now we need to build the risk set matrix,
#which will be dimension (# of cases) x (nmatch + 1), so 200x3 here
#each row denotes a risk set, with the case id followed by the matched control ids
RiskSets <- matrix(nrow = sum(cohortData_ncc$status), ncol = nmatch +1)

for(i in subcohort_ncc$Set){
  RiskSets[i, ] <- unlist(subset(subcohort_ncc, Set==i, select=Map))
}

#Nonparametric estimates 
survMTP.ncc(time = survTime, 
         event = status, 
         marker=Y, 
         subcoh=subcohort, 
         id = id, 
         data = cohortData_ncc,
         risk.sets = RiskSets, 
         estimation.method = "NP", 
         predict.time = 2 , 
         marker.cutpoint = 0)

#Semiparametric estimates 
survMTP.ncc(time = survTime, 
         event = status, 
         marker=Y, 
         subcoh=subcohort, 
         id = id, 
         data = cohortData_ncc,
         risk.sets = RiskSets, 
         estimation.method = "SP", 
         predict.time = 2 , 
         marker.cutpoint = 0)

mdbrown/survMarkerTwoPhase documentation built on May 22, 2019, 3:23 p.m.