non-parametric trait "imputations" in treated people

npsubtreated {GenABEL}R Documentation

non-parametric trait "imputations" in treated people

Description

For people on treatment, the algorithm substitutes the value of the trait using non-parametric algorithm described in Tobin et al., 2005. This algorithm assumes that the measurement in treated subject is a right-censored trait. Essentially, the algorithm substitutes the QT for a person on treatment with the mean of the above-ranked substituted QT value.

Usage

npsubtreated(trait, medication, increase = FALSE)

Arguments

trait vector if trait values
medication medication indicator (0/1)
increase if medication INCREASE the value of the trait (should never be true for e.g. blood pressure, LDL cholesterol, etc.)

Details

Should put the formulas here...

Value

Vector of trait walues, where the values for treated people are substituted with average of the above-ranked substituted trait value.

Author(s)

Yurii Aulchenko

References

Levy D, DeStefano AL, Larson MG, O'Donnell CJ, Lifton RP, Gavras H, Cupples LA, Myers RH. Evidence for a gene influencing blood pressure on chromosome 17. Genome scan linkage results for longitudinal blood pressure phenotypes in subjects from the framingham heart study. Hypertension. 2000 Oct;36(4):477-83.

Tobin MD, Sheehan NA, Scurrah KJ, Burton PR. Adjusting for treatment effects in studies of quantitative traits: antihypertensive therapy and systolic blood pressure. Stat Med. 2005 Oct 15;24(19):2911-35.

Examples

# simulate SBP data
	simmeddat <- function(mu=144,bage=0.5,bsex=4.,bg=2.,pB=0.3,rvar=21^2,N=1000) {
		ageb <- c(25,74)
		pmale <- .5
		htthresh <- 160
		trprob <- .5
		mutreff <- (-15.)
		trvar <- 4^2
		age <- runif(N,min=ageb[1],max=ageb[2])
		sex <- 1*(runif(N)<=pmale)
		gt <- rbinom(N,size=2,prob=pB)
		y.true <- rnorm(N,mu,sqrt(rvar)) + bage*age + bsex*sex + bg*gt
		d.true <- (y.true>=htthresh)
		medication <- 1*d.true
		medication[d.true] <- 1*(runif(sum(d.true))<=trprob)
		treatm <- rnorm(sum(medication),mutreff,sqrt(trvar))
		treatm[treatm>0] <- 0
		treff <- rep(0,N)
		treff[medication==1] <- treatm
		y.obs <- y.true + treff
		out <- data.frame(age,sex,gt,y.true,d.true,medication,treff,y.obs)
		out
}
x <- simmeddat(bg=2.0,N=3000)
x[1:15,]

# substitute value of treated people
imptra <- npsubtreated(x$y.obs,x$medication)
imptra[1:15]

# Almost always, correlation should be higher for the "imputed" trait
cor(x$y.true,x$y.obs)
cor(x$y.true,imptra)

# see what comes out of regression
# analysis of true value
summary(lm(y.true~sex+age+gt,data=x))
# ignore treatment (as a rule, betas are underestimated; on average, power goes down)
summary(lm(y.obs~sex+age+gt,data=x))
# treatment as covariate (as a rule, betas are underestimated; on average, power goes down)
summary(lm(y.obs~sex+age+gt+medication,data=x))
# analyse "imputed" trait (as a rule betas are better; power approaches that of analysis of "true" trait)
summary(lm(imptra~sex+age+gt,data=x))

[Package GenABEL version 1.6-7 Index]