How to conduct Cox PH stratified by matched pair in R I am wanting to calculate hazard ratio in a matched cohort design.
For such I am using the coxph() function in R.
But I have been recommended to stratify by matched pair.
My matching has been simulated using the match() and MatchBalance() commands.
But I am confused as to what 'stratify by matched pair' means.
Can someone explain in simple terms please?
ie. in the simulation does this mean that each matched pair is given a unique identity
OR 
are the all the case with exposure given an identity, (say 1), and all case without exposure given a different identity (say 2).....
I have been unable to find any literature in relation to R, or even a tutorial to follow.
The ‘RcmdrPlugin.EZR can be used for the simulation, but it does explain the Stratified Cox proportional hazard regression for matched-pair analysis
 A: The answer to this question might be best provided by a statician but I'll give it a shot.
The stratified Cox model can be used to perform Cox regression on matched designs by using stratification but it can also be done by modeling with frailties. These are two models which I often see in published articles with matched data. It is, however, not necessary to account for the matching. Some believe that accounting for the matching isn't necessary at all, since it doesn't affect beta coefficients materially and the variables which you have matched on can simply be adjusted for as covariates in the model; this is sufficient in most cases.
The most commonly seen procedures are:
(1) Stratification, if you have pairs matched exactly on a few variables (e.g age, sex, region). You can have as many matched units as you'd like, and thus have countless of strata, thats OK. The Cox regression will pool the results from all strata and provide you with beta estimates of desired predictors. Stratification is done with the "strata" argument in the coxph package; you just provide the variable indicating the matches. This can be extracted from the Matching package, but its a bit tricky.
(2) Modeling with frailty: Frailty models are like mixed effect models for Cox regression (I assume). These are commonly used in propensity score matched studies to improve model prediction. The whole idea with frailty is that there might be some factors that are of importance for the outcome, we do not know of the factors, but we believe that this uncertainty can be accounted for by allowing each match have its own hazard function.
So: obtain the variable indicating matched pairs from the Matching package and add that variable to the strata-argument in the coxph function.
Here is a review: http://meetings.sis-statistica.org/index.php/sis2013/ALV/paper/viewFile/2542/328
