Using proc genmod for log-binomial regression; failing to converge with continuous variable in model

[mlensing]

I am using this code to calculate RR (outcome: dpneum=1, primary predictor: pathogen) for a retrospective cohort study using log-binomial regression:

proc genmod data = clean.stacked;
	class	pathogen			(REF = 'RSV')
			cld					(REF = '99')
			race_eth			(REF = 'White')
			;
	model dpneum2 (event="1")= pathogen age_years cld race_eth/dist=bin link=log;
run;

I recently introduced "age_years" (continuous variable) into the model and since doing so, the model is failing to converge with the log results as follows: 

NOTE: PROC GENMOD is modeling the probability that dpneum2='1'. One way to change this to model the probability that dpneum2='99' 
      is to specify the DESCENDING option in the PROC statement.
WARNING: The specified model did not converge.
NOTE: The Pearson chi-square and deviance are not computed since the AGGREGATE option is not specified.
ERROR: The mean parameter is either invalid or at a limit of its range for some observations.
NOTE: The scale parameter was held fixed.
NOTE: The SAS System stopped processing this step because of errors.
NOTE: PROCEDURE GENMOD used (Total process time):
      real time           6.78 seconds
      cpu time            0.18 seconds

I am not sure of next steps. I have been able to successfully run the model as both a Poisson regression and logistic regression (model converges), however each of these is giving vastly different parameter estimates (i.e. completing changing the direction of the effect between models).

I have opted to use log-binomial regression to assess the association between pathogen and dpneum2 because the overall prevalence of dpneum2 in the sample is >10% (16.5%), however the prevalence per pathogen does dip below 10% for some (16.1%, 10.1%, 7.8%) so I am wondering if logistic regression is the better choice if this is criteria for suggesting the outcome is "rare."

Ideally I would like to run a log-binomial regression here, so any suggestions for data restructuring or otherwise are appreciated.

[sbxkoenk]

Hello,

Please see this usage note which addresses the error message

and this note that discusses such problems with log-linked binomial models.

The answers to many questions can be found in the Samples and SAS Notes in our searchable knowledgebase.  

You can use the search engine there to find the answers you need.  
See also our list of Frequently Asked for Statistics, FASTats.

(answer stolen from @StatDave 20-OCT-2016)

Koen