matched logistic regression

["Biggerstaff, Brad J." <bkb5@cdc.gov>]

 

Dear S-news list folks:

 

A few years ago I posted a function, mlogitreg, that is a wrapper function to coxph for doing matched logistic regression.  In that function, the matching variable was specified using a character string, e.g., matching.var = "M".  I have written a new version of this function, now called mlr, to use the within-formula specification of strata/matching variable as with coxph and other functions, just to be consistent with current S use.  Here the matching variable is specified using matching(M) in the formula, as one specifies strata(M) in coxph.  The function is, as before, just a wrapper for coxph; it's just updated to have a more modern S flavor.

 

I include this updated version (written for S-Plus 6.x) in case you might find it useful.  While the returned object has the coxph class, I changed the call element in the return object to be the call in the mlr function. (But update doesn't work yet.) The arguments are as with coxph.

 

I welcome comments, problems.

 

Cheers,

Brad

 

"mlr"<-

function(formula=formula(data), data = "" subset, na.action,

   control, singular.ok = T, robust = F, model = F, x = F, y = T, ...)

{

   # Function to do matched logistic regression by calling coxph

   #

   # Required arguments

   #

   # formula:

   #

   #  A formula of the form Response ~ Covariates + matching(Var), where

   #  Covariates indicates model terms and Var is the variable indicating

   #  the matching.

   #

   #  All other arguments as with coxph().

   #

   #  Author:

   #

   #    Brad Biggerstaff, Ph.D.

   #    Division of Vector-Borne Infectious Diseases

   #    National Center for Infectious Diseases

   #    Centers for Disease Control and Prevention

   #    P.O. Box 2087, Fort Collins, CO, 80522-2987 USA

   #    +1 970.221.6473  ---  BBiggerstaff@cdc.gov

   #  

   # really just use the coxph function strata for the matching function

   if(!exists("matching"))

      assign("matching",strata,frame=1)

   m <- match.call(expand = F)

   call <- match.call()

   temp <- c("", "formula", "data", "subset", "na.action")

   m <- m[match(temp, names(m), nomatch = 0)]

   special <- c("matching")

   Terms <- if(missing(data)) terms(formula, special) else terms(formula,

         special, data = "">

   m$formula <- Terms

   m[[1]] <- as.name("model.frame")

   m <- eval(m, sys.parent())

   m$PseudoTime <- rep(1, nrow(m))

   #

   # manipulate the formula to make it look like a call to coxph

   #

   CharFormula <- as.character(formula(formula))

   ResponseName <- CharFormula[2]

   CharFormula[2] <- paste("Surv(PseudoTime,", ResponseName, ")", sep = "",

      collapse = "")

   rhs <- CharFormula[3]

   matching.ind <- as.vector(regMatchPos(rhs, "matching"))

   substring(rhs, matching.ind[1], matching.ind[2]) <- "strata"

   CharFormula[3] <- rhs

   CharFormula <- paste(CharFormula[c(2, 1, 3)], collapse = "")

   if(missing(control))

      control <- coxph.control(...)

   if(missing(na.action))

      na.action <- get("na.exclude")

   names.m <- names(m)

   MatchingVarNamePos <- seq(along = names.m)[regMatch(names.m, "matching"

      )]

   MatchingVarName <- names.m[MatchingVarNamePos]

   MatchingVarName <- unpaste(unpaste(MatchingVarName, "(")[[2]], ")")[[

      1]]

   names(m)[MatchingVarNamePos] <- MatchingVarName

   fit <- coxph(formula(CharFormula), data = "" = "exact", control = control,

      singular.ok = singular.ok, robust = robust, model = model, x = x, y = y)

   fit$call <- call

   fit

}

 

 

 

 

Brad Biggerstaff, Ph.D.
Mathematical Statistician
Division of Vector-Borne Infectious Diseases
National Center for Infectious Diseases
Centers for Disease Control and Prevention
P.O. Box 2087
Fort Collins, Colorado  80522-2087
(970) 221-6473 ... BBiggerstaff@cdc.gov