topic hazard ratio from regression coefficients in an interaction term in Statistical Procedures

hazard ratio from regression coefficients in an interaction term

psh23 — Wed, 13 Jun 2018 04:16:47 GMT

I am using proc phreg to estimate hazard ratios, my model is as follows:

time*event(0) = week|avg education race;

random zipcode;

I would like to get the hazard ratio for 'week'. I was originally doing this by hand by summing the estimate from week and the interaction term, but assume there is a way to do this using estimate/contrast?

Thank you!

Re: hazard ratio from regression coefficients in an interaction term

Reeza — Wed, 13 Jun 2018 00:00:31 GMT

Try a HAZARDRATIO statement instead. Much simpler syntax.

Re: hazard ratio from regression coefficients in an interaction term

pau13rown — Wed, 13 Jun 2018 00:54:45 GMT

youdont have week as a main factor in the model, just in the interaction? you'll need to use 'at' in the hazardratio statement @Reeza suggested, see here: https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_phreg_sect014.htm

Re: hazard ratio from regression coefficients in an interaction term

psh23 — Wed, 13 Jun 2018 04:21:30 GMT

When I use the hazardratio statement I cannot recreate what I get when I am summing the coefficients by hand. This is why I wasn't using it initially - is there a specification that needs to be made?

Example:

	Estimate	Exp summed estimate	Hazard Ratio from SAS
average	0.000196	0.999763628	0.995
week	-0.0002121
week*average	-0.0000243

Re: hazard ratio from regression coefficients in an interaction term

pau13rown — Wed, 13 Jun 2018 04:42:08 GMT

are you doing it by hand or in a data step? it could be rounding error, calculate the hr in a data step as exp(estimate). Also did you use 'at' in the hazard ratio statement? if you just sum them up then it's at average=1?

Re: hazard ratio from regression coefficients in an interaction term

Reeza — Wed, 13 Jun 2018 14:42:15 GMT

99% of the time the error is with the hand calculation.

Do it with a data step so you can't get any typo's or rounding errors and post your code if it still doesn't match.