topic Re: Relative risk per a unit decrease instead of increase? in Statistical Procedures

Relative risk per a unit decrease instead of increase?

Gen — Sun, 26 Oct 2014 03:21:17 GMT

Hi,

I obtained relative risk (RR) using PROC PHREG for a continuous predictor. By default, the RR was calculated for a unit INCREASE of the predictor, However, what I want to obtain is the ratio per a unit DECREASE. In other words, SAS calculates "(risk in a higher level) / (risk in a lower level)" but I want to make SAS to calculated "(risk in a lower level) / (risk in a higher level)." If you know how to obtain relative risk or odds ratio per a unit decrease of a continuous predictor with a command other than PROC PHREG (i.e., PROC REG, PROC LOGISTICS, etc...), please share your knowledge.

Following are the codes that I used.

proc phreg data=please;

model time*event(0)=continuous_predictor;

run;

Thanks!

Re: Relative risk per a unit decrease instead of increase?

Reeza — Sun, 26 Oct 2014 04:40:01 GMT

Try HazardRatio statement with unit = -1.

Re: Relative risk per a unit decrease instead of increase?

Gen — Sun, 26 Oct 2014 15:24:40 GMT

Thank you for your comment, Reeza. Would you specify what you mentioned above?

Re: Relative risk per a unit decrease instead of increase?

Reeza — Mon, 27 Oct 2014 02:57:37 GMT

Hmm...I thought the following would work, but I guess not since unit only accepts positive values.

Not sure

proc phreg data=please;

model time*event(0)=continuous_predictor;

hazardratio continuous_predictor/unit=-1;

run;

Re: Relative risk per a unit decrease instead of increase?

StatDave — Mon, 03 Nov 2014 19:39:54 GMT

PROC LOGISTIC will also report the odds ratio for a unit increase in the continuous predictor. To get the reciprocal of that odds ratio, you would have to model the other level of the binary response. For example, if your response levels are coded 0 and 1 and you model the probability of 0:

model y(event="0")=x;

then modeling the probability of 1 will give an odds ratio that is the reciprocal of the above:

model y(event="1")=x;