exposureMatch: Exact matching on time-dependent exposure

In tagteam/heaven: Data Preparation Routines for Medical Registry Data

Exact matching on time-dependent exposure

Description

A case is a person who initiates treatment at index date. A control is a person who has not yet initiated treatment at the index date of the case. The index date of the case can also be the onset of a disease (or other exposure), in which case controls are persons who do not yet have the disease (are not yet exposed). See incidenceMatch for situations where a case has a time to event outcome at the index date and controls do not have the outcome at the case index date.

Within a cohort or registry study, a number of controls are matched to each case on demographics, comorbidity, concomitant medication and other pre-treatment person characteristics measured at the index date of the case.

Matching may be useful in non-randomized studies when the aim is to estimate a

• causal parameter: The average effect of treatment in a hypothetical study which randomizes treatment allocation. E.g., the difference or ratio of the absolute risk of developing an outcome within the first 6 months after the index date.

• regression parameter: The conditional effect of treatment (exposure) on an outcome given pre-treatment (pre-exposure) covariates. E.g., a hazard ratio in a Cox regression model, or an odds ratio in a logistic regression model. In particular, when it is difficult or expensive or time-consuming to measure pre-exposure variables or to remove model dependency. See Details and references.

Usage

exposureMatch(
  ptid,
  event,
  terms,
  data,
  n.controls,
  case.index = NULL,
  end.followup = NULL,
  date.terms = NULL,
  duration.terms = NULL,
  output.count.controls = TRUE,
  cores = 1,
  seed = 0,
  progressbar = TRUE
)

Arguments

ptid	Personal ID variable defining participant
event	Name of variable that defines cases. MUST be numeric 0/1 where 0 codes for never-case, and 1 for case.
terms	Vector of variable names specifying the variables that should be matched on. Make sure that appropriate classification is in place for truly continuous variables, such as age. This is to ensure a sufficient number of controls for each case. For example it may be difficult to find controls for cases of very high and very low ages and extreme ages should therefor further aggregated.
data	The single dataset with all information - coerced to data.table if data.frame
n.controls	Number of controls for each case
case.index	Name of the variable which contains the case index dates. This can be a calendar date variable or a numeric variable, i.e., the time to outcome event from a well defined baseline date. Missing values are interpreted as no event at the end of followup.
end.followup	Name of the variable which defines the date (as date or time) from which a control can no longer be selected as a control due in general to. death: Nothing happens thereafter exposure of interest: a control can be selected among those that later develop condition of interest, but not after the date of exposure other competing risks: Event after which we are not interested in the subject anymore. E.g., the date of an outcome event. censoring: Event after which we do not observe anymore. E.g., emmigration, end of study period, end of registry, drop-out The end.followup must be larger or equal to the case.index date.
date.terms	A vector of variable names (character) in data specifying dates of conditions such as comorbidities. Controls are considered as matched only when the date.terms of the control are on the same chronological side of case index as for the case. Missing values are interpreted as absence of comorbidity at end.followup. Many datasets have comorbidities as time dependent variables. Matching on these requires that the comorbidity date is not (yet) reached for a corresponding variables for cases if the case does not have the comorbidity and similarly that the date has been reached when the case does have that co-morbidity.
duration.terms	A list where each element defines a time duration term with two elements: startName of a variable which defines a date or time which defines the duration as the difference between this variable and the case.index. minNumeric value given in days when case.index is a date and in the same time unit as case.index otherwise. Cases and controls have either both a duration as least as long as min or both a duration shorter than min. Useful to prepare to summarize the history of exposure for cases and controls in an equally long period looking back in time from the case.index.
output.count.controls	Logical. If TRUE add number of found controls to each case/control set.
cores	number of cores to use in the calculation.
seed	Random seed to make results reproducible
progressbar	set to FALSE to avoid progressbar

Details

The function performs exact matching and hence all matching variables must be factor variables or character.

In some cases there is a necessity to account for delayed entry. An example could be to ensure that people are not selected as controls prior to immigration to the country. This can be handled with "date.terms". Cases and controls can as an example be provided with a date.term that is maximum of the date of birth and the date of immigration.

In some cases it makes sense to think about the study population of a hypothetical study with randomized treatment allocation. Then, the aim of the function is similar to that of propensity score analysis and exact matching may be preferred as it does not dependent on a propensity score model (see King & Nielsen 2019).

It may appear tempting always to use multiple cores, but this comes at the cost of copying the data to the cores.

The function matchReport may afterwards be used to provide simple summaries of use of cases and controls

Value

data.table with cases and controls. After matching, a the variable "case.id" identifies sets which include 1 case and x matched controls.

Variables in the original dataset are preserved. The final dataset includes all original cases but only the controls that were selected.

Author(s)

Christian Torp-Pedersen & Thomas Alexander Gerds

References

Paul R Rosenbaum. Risk set matching. Design of observational studies, Springer, 2010.

Yunfei Paul Li, Kathleen J Propert, and Paul R Rosenbaum. Balanced risk set matching. Journal of the American Statistical Association, 96(455):870– 882, 2001.

Mohammad A Mansournia, Miguel A Herna'n, and Sander Greenland. Matched designs and causal diagrams. International journal of epidemiology, 42(3): 860–869, 2013.

King, G., & Nielsen, R. Why Propensity Scores Should Not Be Used for Matching. Political Analysis, 1-20. doi:10.1017/pan.2019.11

See Also

riskSetMatch matchReport Matchit

Examples

require(data.table)
case <- c(rep(0,40),rep(1,15))
ptid <- paste0("P",1:55)
sex <- c(rep("fem",20),rep("mal",20),rep("fem",8),rep("mal",7))
byear <- c(rep(c(2020,2030),20),rep(2020,7),rep(2030,8))
case.Index <- c(seq(1,40,1),seq(5,47,3))
startDisease <- rep(10,55)
control.Index <- case.Index
diabetes <- seq(2,110,2)
heartdis <- seq(110,2,-2)
diabetes <- c(rep(1,55))
heartdis <- c(rep(100,55))
library(data.table)
dat <- data.table(case,ptid,sex,byear,diabetes,heartdis,case.Index,
control.Index,startDisease)
# Exposure density matching
matchdat <- exposureMatch(ptid="ptid",event="case",
            terms=c("byear","sex"),data=dat,n.controls=2,
            case.index="case.Index",end.followup="control.Index")
matchdat
# Same with 2 cores
## Not run: 
library(parallel)
library(foreach)
matchdat2 <- exposureMatch("ptid","case",c("byear","sex"),data=dat,2,case.index=
  "case.Index",end.followup="control.Index"
  ,cores=2)
matchdat2
all.equal(matchdat,matchdat2)

## End(Not run)
#Time dependent matching. In addition to fixed matching parameters there are
#two other sets of dates where it is required that if a case has that condi-
#tion prior to index, then controls also need to have the condition prior to
#the case index to be eligible - and if the control does not have the condi-
#tion prior to index then the same is required for the control.
matchdat <- exposureMatch("ptid","case",c("byear","sex"),data=dat,n.controls=2,
  case.index="case.Index",end.followup="control.Index",cores=1,
  date.terms=c("diabetes","heartdis"))
matchdat

tagteam/heaven documentation built on May 28, 2024, 9:22 p.m.