re: Re: st: replicating 2 X 2 data from a paper

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

Thanks Adam and David. It's nice to know that I am not crazy. I also tried
cci and iri and obviously got the same results as Adam. I am somewhat
surprised that a paper in the Archives of Internal Medicine would not be
scrutinized more closely by statistical reviewers. 

Date: Tue, 2 Apr 2013 21:04:43 -0400
From: Adam Olszewski <adam.olszewski@gmail.com>
Subject: Re: st: replicating 2 X 2 data from a paper

Hi Ariel,
The footnote in the table says "Percent differences were calculated by
dividing the absolute difference between groups by the usual care
group value."
So, the "Treatment difference" = (158-152)/152 = .03947368

How they got the P-value of 0.45 I have no idea.
cci 152 158 95 96   gives 0.87
iri  152 158 95 96   gives  0.80

AO

Date: Tue, 2 Apr 2013 22:21:47 -0400
From: David Hoaglin <dchoaglin@gmail.com>
Subject: Re: st: replicating 2 X 2 data from a paper

Hi, Ariel.

As presented, those data would not be compatible with a 2 X 2 table
format.  If one were classifying the subjects in each group according
to whether they had a hospitalization at (prior to?) baseline, the
result would form a 2 X 2 table, to which one could apply a
chi-squared test or an exact test.

If I'm interpreting the numbers correctly (it's not convenient for me
to look up the article), the 96 subjects in the treatment group had a
total of 152 baseline hospitalizations among them, and the 95 subjects
in the control group had a total of 158 baseline hospitalizations.
Those data yield rates (as you calculated), not proportions.  It seems
that the authors did not document their statistical analysis
adequately.  A comparison of two rates may involve a chi-squared test.
 I don't recall.  Treating it as Poisson regression would seem to be
overkill.

David Hoaglin

On Tue, Apr 2, 2013 at 8:16 PM, Ariel Linden, DrPH
<ariel.linden@gmail.com> wrote:
> Hi Fellow Statalisters,
>
> I am having a problem conceptualizing how to replicate data from a paper I
> am reading  (Bourbeau et al. 2003).* They have two groups (treatment = 96,
> control= 95) who had 152 and 158 baseline hospitalizations, respectively.
>
> They describe in the text: " the comparison of the proportion of hospital
> admissions or emergency department and medical visits was based on the
chi2
> test. The Fisher exact test was used when the frequencies were small.
> Percent difference effects of the intervention were calculated by dividing
> the absolute difference between the intervention and usual care group
> values."
>
> The table (table 2) shows the difference as 3.9% and a p-value of 0.45.
>
> So with this minimal information, I'd like to replicate the results,
perhaps
> using a 2 X 2 table format and test with chi2.
>
> My basic math (see below) does not seem to replicate the difference they
> present. I get close using (treat-usual)/treat, but that is not consistent
> with their text. Using a rate difference, I get something completely
> different. In any case, I am not sure if there is a simple way to
reproduce
> these results consistent with how they claim to get there...
>
>                 usual   treat           (treat-usual)/treat
> admits          152     158             0.038
> N               95      96
> admits/N        1.6     1.65            0.046
>
> * Bourbeau JB, et al. Reduction of hospital utilization in patients with
> chronic obstructive pulmonary disease. Arch Intern Med 2003: 163:585-591.
>
> Any help would be appreciated!
>
> Ariel


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/