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

knitr::opts_chunk$set(echo = TRUE)

AIPWmeasures

This package includes a function AIPWmeasures that calculates standard measures of predictive accuracy commonly used in biomarker validation studies. More specifically, this function calculates estimates using data from two-phase sampling designs ('case-cohort' and 'nested case-control') using two different methods, a standard ipw estimator (true ipw) and a novel method (augmented ipw) that has been shown to be more efficient in some contexts. See the manuscript "Improving Efficiency in Biomarker Incremental Value Evaluation under Two-phase Study Designs" by Zheng et. al. for more details.

Install the package

if (!require("devtools")) install.packages("devtools")
devtools::install_github("AIPWmeasures", "mdbrown")

Quick Example

library(AIPWmeasures)
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 = 0.8)

# true 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 = 'True',
              design = "CCH",
              smoothing.par = 0.9,
              calculate.sd  = TRUE,
              pnf.threshold = 0.85,
              pcf.threshold = 0.8,
              ncc.nmatch = 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 = 0.8,
                    ncc.nmatch = 2)

#true ipw
AIPWmeasures( time = NCCsimdata$xi, event = NCCsimdata$di,
                    X = cbind(NCCsimdata$y1, NCCsimdata$y2),
                    subcohort = NCCsimdata$vi,
                    risk.threshold  =  c(.01, .03),
                    landmark.time = predict.time,
                    weight.method = 'True',
                    design = "NCC",
                    smoothing.par = 0.9,
                    calculate.sd  = TRUE,
                    pnf.threshold = 0.85,
                    pcf.threshold = 0.8,
                    ncc.nmatch = 2)

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