s-news
[Top] [All Lists]

Re: Calculate CI of rate (person time)

To: blasarmien@yahoo.com.mx
Subject: Re: Calculate CI of rate (person time)
From: Terry Therneau <therneau@mayo.edu>
Date: Mon, 19 May 2008 10:27:04 -0500 (CDT)
Cc: s-news@wubios.wustl.edu
Reply-to: Terry Therneau <therneau@mayo.edu>
> Hi. I would like to know if you have a command to calculate the confidence
> intervals of the rate (Person Time) and to compare and to determine
> statistical differents between two rates.

A simple way is to use poisson regression via glm.  Here is a dummy data set:

> tdata
  event time group 
1    12  100     1
2    28  200     2
3    29  100     3

> fit <- glm(event ~ factor(group) + offset(log(time)), data=tdata,             
>         
                        family=poisson)
> summary(fit)

Coefficients:
                    Value Std. Error t value Pr(>|t|) 
   (Intercept) -2.1202635  0.2886751 -7.3448       NA
factor(group)2  0.1541507  0.3450328  0.4468       NA
factor(group)3  0.8823892  0.3432435  2.5707       NA

(Dispersion Parameter for Poisson family taken to be 1 )

    Null Deviance: 9.730052 on 2 degrees of freedom

Residual Deviance: 0 on 0 degrees of freedom

Number of Fisher Scoring Iterations: 1 

-------------

  Since my dummy data has only 1 subject per group S doesn't know what to do 
for 
a p-value, but don't worry about it.  It will fix itself when you have multiple 
subjects. The test for group 1 vs group 3 has
        relative risk = exp(.88) = 2.41 
        with 95% CI of  exp(.88 - 1.96*.343) to exp(.88 + 1.96*.343)
        
  For a single group use the intercept and confidence interval for the 
intercept.
  
        Terry T.
        
        


<Prev in Thread] Current Thread [Next in Thread>