Negative binomial regression

[AndreyMyslivets]

Dear Sirs or madams, could you please help me with the NBRM analysis.

I have following data (the dataset in attchment):

num_of_resp="Number of positive responses"
offs="Offset"
chis="Standartizied baseline val"
site="Site"
treatment="Treatment Arm"

And I have following description of the analysis in the SAP:

NBRM is a negative binomial regression model for the number of events for all observation periods after the start of treatment. The Treatment Arm will be used as a fixed factor, the baseline will be used as covariates, the period as a factor with repetitions, and the interaction group “therapy arm× visit” will be included in the model. The model will use an unstructured covariance matrix [UN]. If there is no convergence, the 1st order autoregression matrix [AR (1)] and the composite symmetry covariance matrix [CS] will be used successively. If it is impossible to build a model or it is impossible to interpret a model with a period as a factor with repetitions, separate models will be constructed for each period after the start of therapy. For the supporting analysis, the same model will be built (or individual models by periods after the start of therapy) with the center as an additional fixed factor.

Could you please help with the code for the anlysis?

Thanks in advanse,

Best Regards, Andrey!

[StatDave]

The description suggests that there are repeated measurements taken on subjects. The specification for various types of covariance matrices might mean that a Generalized Estimating Equations (GEE) model using the negative binomial distribution is intended. This can be done in PROC GEE. See the example titled "Log-Linear Model for Count Data" in the PROC GEE documentation for your release of SAS. While that example uses the Poisson dist, you can easily specify the negative binomial distribution with DIST=NEGBIN instead of DIST=POISSON.