Estimating treatment effects, 2 Group Pre-Post Matched Analysis

[bzagor1]

Greetings-

I am trying to estimate the treatment effect for a pre/post design, subjects were measured on healthcare utilization 1yr before and after the intervention (non-random assignment).   The dependent vars are  cost/utilization measures that are right skewed, some zero-inflated.   To control for the vastly different distribution of covariates, case subjects are matched to controls 1:1.  For the dependent variables that are not zero-inflated I am using GEE in GENMOD with either Gamma or Negative Binomial distribution, however, with the zero-inflated dep vars, GEE is not available and am seeking advice on which SAS statistical procS are available that would accommodate the zero-inflated and matched design.

Thanks in advance, Brandon Zagorski

Pre-Post Matched Analysis

[Ksharp]

You could take "Matched Analysis" as Mixed model(a.k.a repeated measured model).
Here is something you can refer to:

https://support.sas.com/kb/46/997.html

Also you could try PROC GLIMMIX to model Mixed model.

[bzagor1]

Thanks for the response.

Unfortunately, my dependent var is not dichotomous (its continuous, right skewed, zero inflated) so that reference do not apply.   And Proc Glimmix does not have zero inflated distributions available.

Any other suggestions would be greatly appreciated.

Regards, Brandon

[Ksharp]

proc genmod can fit zero-inflated Poisson distribution.
also you could try tweedie distribution which also could handle many zeros in Y.

[bzagor1]

Thanks, @Ksharp .  I have a matched cohort and GENMOD does not accommodate this design in the zero-inflated models.  Any other suggestions would be greatly appreciated.

Regards, Brandon

[Ksharp]

Calling @SteveDenham  @StatDave @lvm 

[StatDave]

See the "Using PROC NLMIXED" sections of this note which show how you can fit a zero-inflated model using PROC NLMIXED. You could include a RANDOM statement to add a random effect for your matched pairs.

But the zero-inflated distributions, Poisson or negative binomial, are for a count response. For a continuous response you might consider using a modified gamma or Tweedie distribution, again fit using NLMIXED to which a RANDOM statement could be added. The modified gamma is shown in this note. 

[lvm · Accepted solution]

As indicated by StatDave, you need to use NLMIXED. THis is a very powerful procedure, but it requires programming. In addition to the link give, I like the following for very good instruction:

https://stats.oarc.ucla.edu/sas/faq/how-do-i-run-a-random-effect-zero-inflated-poisson-model-using-n...

[Ksharp]

@lvm,
You got it, well done!

[bzagor1]

Great, thanks.  I will look into this.  Again, thanks for your help!

Regards, Brandon

[Ksharp]

This credit should give to lvm. He deserve it .
Plz mark lvm's answer correct ,not mine .