mercRel: mercRel

In vanessaxiaofan/merc: Regression calibration methods for validation study and reliability study

mercRel

Description

This funciton calculates regression coefficients, their standard errors, and odds ratios, when relevant, and 95 biologically meaningful difference specified by model covariates. A reliability study is required to empirically charaterize the measurement error model. Details are given in Rosner et al.(1989), Rosner et al.(1990), and Rosner et al.(1992) including "real data" examples

Usage

mercRel(
  supplyEstimates = FALSE,
  relib,
  pointEstimates = NA,
  vcovEstimates = NA,
  sur,
  woe = NA,
  event = NA,
  time = NA,
  outcome = NA,
  weri,
  rr,
  ms,
  weights,
  r1 = 1,
  method = "lm",
  family = NA,
  link = NA
)

Arguments

supplyEstimates	Indicates whether uncorrected estimates will be supplied by the user.
relib	Name of the reliability dataset
pointEstimates	A numeric vector of point estimates from uncorrected standard regression analysis. It must be in the order of the variables indicated in sur and woe Intercept estimate must be removed. Only binary and numeric variables are accepted, all other classes of variables must be transformed (this is automatically done by most regression softwares). Must include names for each point estimates corresponding to the (transformed) names from 'covCalib' followed by 'covOutcome'. Must be supplied if supplyEstimates = TRUE.
vcovEstimates	A p by p Variance-covariance matrix estimates from uncorrected standard regression analysis. Intercept estimates must be removed. Only binary and numeric variables are accepted, all other classes of variables must be transformed (this is automatically done by most regression softwares). Must include column names (excluding intercept) for the estimates corresponding to the (transformed) names from 'covCalib' followed by 'covOutcome'. Must be supplied if supplyEstimates = TRUE.
sur	character vector of mismeasured exposure and covariates (i.e. surrogates) in the main study dataset
woe	Character vector of names of perfectly measured covariates
event	Required if method is cox. The event status indicator, normally 0=alive/censored, 1=dead/event. For interval censored data, the status indicator is 0=right censored, 1= event at time, 2=left censored, 3=interval censored.
time	Required if method is cox. For right censored data, this is the follow up time. For interval data, the first argument is the starting time for the interval.
outcome	Outcome variable, required if method is lm or glm.
weri	Character vector of names of variables which have reliability measures. These should have a set of variable names for each reliability measure (e.g. x1 y1 z1 x2 y2 z2 for measures). The variables must be in the same order as the names in the names and increments dataset specified in 'pointEstimates'
rr	Number of reliability measures taken (e.g. 2).
ms	name of main study data set
weights	Name of the dataset containing the variable names and increments for the odds ratio or regression slopes.Including mismeasured and perfectly measured variables
r1	Number of replicates in main study
method	Methods for modeling, currently only 'lm', 'glm' and 'cox' methods are available. Required.
family	Supply family parameter to pass to glm function. Not a character. Required if method="glm".
link	Supply link parameter to pass to glm function. Should be character. Required if method="glm".

Value

printable dataframe from standard regression results (when supplyEstimates==FALSE) as well as corrected results

Author(s)

Xiaofan Liu and Xin Zhou

References

Rosner B, Spiegelman D, Willett WC. Correction of logistic regression relative risk estimates and confidence intervals for measurement error: the case of multiple covariates measured with error. Am J Epidemiol. 1990 Oct;132(4):734-45. doi: 10.1093/oxfordjournals.aje.a115715. PMID: 2403114.

Examples

# # Only one mismeasured covariate, linear model
y <- c(1,1,2,3,2,3)
x <- c(2,2.1,3,4,2.9,4.1)
s <- c(2,5,4,3,3,5)
case <- c(0,1,0,1,0,0)
age <- c(10,10,11,12,13,13)
test <- data.frame(y,x,s,case,age)

x <- c(1.1,1.2,0.8)
x2 <- c(1.2,1.2,0.9)
x3 <- c(0.9,1.0,1.3)
relib <- data.frame(x,x2,x3)
wts <- data.frame(x=1,s=1)

### Fit main study linear model
outModel <- lm(y ~ x + s , data = test)
outcomeParam=coef(outModel)
outcomeParamVCOV=vcov(outModel)
outcomeModelResults<-(list(outcomeParam,outcomeParamVCOV))
Bstar<-outcomeParam[2:length(outcomeParam)] #p' x 1
VBstar<-outcomeParamVCOV[2:length(outcomeParam),2:length(outcomeParam)]
mercRel(supplyEstimates=TRUE, relib=relib, pointEstimates = Bstar,
         vcovEstimates = VBstar,sur = c("x"), woe = c("s"), weri = c("x","x2","x3"),
         rr=3, ms=test, weights=wts, method = "lm" )
mercRel(supplyEstimates=FALSE, relib = relib, sur = c("x"), woe = c("s"),
         weri = c("x","x2","x3"), outcome = c("y"), rr=3,
         ms=test,method = "lm",weights=wts)

# # Only one mismeasured covariate, logistic model
### Fit main study logistic model
outModel <- glm(case ~ x + s , family = binomial(link="logit"), data = test)
outcomeParam=coef(outModel)
outcomeParamVCOV=vcov(outModel)
outcomeModelResults<-(list(outcomeParam,outcomeParamVCOV))
Bstar<-outcomeParam[2:length(outcomeParam)] #p' x 1
VBstar<-outcomeParamVCOV[2:length(outcomeParam),2:length(outcomeParam)] # p' x p'
mercRel(supplyEstimates=TRUE, relib=relib, pointEstimates = Bstar, vcovEstimates = VBstar,
         sur = c("x"), woe = c("s"), weri = c("x","x2","x3"), rr=3, ms=test, weights=wts,
         link = "logit",method = "glm" )
mercRel(supplyEstimates=FALSE, relib = relib, sur = c("x"), woe = c("s"),
         weri = c("x","x2","x3"),outcome = c("case"), rr=3, ms=test,
         method = "glm", family = binomial, link = "logit", weights=wts)

vanessaxiaofan/merc documentation built on May 7, 2023, 6:17 p.m.