AIPWmeasures: Estimate standard measures of predictive accuracy for...

In mdbrown/AIPWmeasures: Estimate measures of predictive accuracy using augmented inverse probability weights for two-phase biomarker validation studies

Description

Estimate measures of predictive accuracy using augmented inverse probability weights (ipw) or true ipw for two-phase biomarker validation studies

Usage

AIPWmeasures(time, event, X, subcohort, aug.weights.x = NULL, risk.threshold,
  landmark.time, weight.method = c("Aug", "True"), design = c("CCH", "NCC"),
  calculate.sd = TRUE, smoothing.par = 0.7, pnf.threshold = 0.5,
  pcf.threshold = 0.2, ncc.nmatch = NULL)

Arguments

time	vector of time to event
event	status indicator (1 for observed failure, 0 for censoring)
X	matrix or data frame of covariates
subcohort	indicator for selection into the subcohort (1 for selected, 0 for not selected)
aug.weights.x	vector of variable to be used to derive augmented inverse probability weights.
risk.threshold	vector of risk thresholds on the absolute risk scale used to calcuate cutoff-based summary measures.
landmark.time	numeric value specifying landmark time at which to estimate the performance measures.
weight.method	either "Aug" for augmented ipw or "True" for true ipw.
design	either "CCH" for case-cohort or "NCC" for nested case control specifying the subcohort design used.
calculate.sd	should analytic standard errors be calculated.
smoothing.par	nearest neighbor smoothing parameter used in locfit (default = 0.7)
pnf.threshold	value used to calculate proportion needed to follow (PNF) such that pnf.threshold percent of the cases are classified high risk.
pcf.threshold	value used to calculate proportion of cases followed (PCF) if pcf.threshold percent of the population are classified high risk.
ncc.nmatch	For design = "NCC", specify the number of controls matched per case.

Value

data.frame with estimates of AUC, IDI, ITPR, IFPR, TPR, FPR, PPV, NPV and net benefit (NB) and standard errors.

Note

variance calculations are unavailable for the NB measure.

Examples

data(CCHsimdata)

predict.time <- 0.75

## augmented ipw
AIPWmeasures( time = CCHsimdata$xi, event = CCHsimdata$di,
             X = cbind(CCHsimdata$y1, CCHsimdata$y2),
            subcohort = CCHsimdata$vi,
             aug.weights.x =   CCHsimdata$y1,
             risk.threshold  =  c(.05, .3),
             landmark.time = predict.time,
             weight.method = 'Aug',
             design = "CCH",
             smoothing.par = 0.9,
             calculate.sd  = TRUE,
             pnf.threshold = 0.85,
             pcf.threshold = 0.2)


#simulated data from a ncc design with nmatch = 2
data("NCCsimdata")
AIPWmeasures( time = NCCsimdata$xi, event = NCCsimdata$di,
             X = cbind(NCCsimdata$y1, NCCsimdata$y2),
             subcohort = NCCsimdata$vi,
             aug.weights.x =   NCCsimdata$y1,
             risk.threshold  =  c(.01, .03),
             landmark.time = predict.time,
             weight.method = 'Aug',
             design = "NCC",
             smoothing.par = 0.9,
             calculate.sd  = TRUE,
             pnf.threshold = 0.85,
             pcf.threshold = 0.2,
             ncc.nmatch = 2)

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