Compare years

[Øivind Hundal]

Hello to you all
 
I calculate odds ratio for concurrent use of antidepressants when you also receive antidiabetic medication. Utilizing logistic regression. How can I compare and test difference between 2 years in these ORs? Appreciate syntax.
 
E.g.: OR for concurrent use of drug A when you use drug B in 2005 is X, and OR for concurrent use of drug A when you use drug B in 2012 is X.
 
How can I test if OR(2005) is significantly different from OR(2012)? 
 
In both cases we use data from the complete Norwegian population. Our ORs are a "snapshot" from each year, and as some patients use drug B in both years, there is a partial overlap between the patient groups in these two years.
However, I am not interested in any development from 2005 to 2012. Just to compare two snapshots, point prevalences.
 
Øivind

[Bruce Weaver]

Do you have unique identifiers for individuals so that you can match in cases where people have data for both years?

Øivind Hundal wrote

Hello to you all

I calculate odds ratio for concurrent use of antidepressants when you also
receive antidiabetic medication. Utilizing logistic regression. How can I
compare and test difference between 2 years in these ORs? Appreciate syntax.

E.g.: OR for concurrent use of drug A when you use drug B in 2005 is X, and
OR for concurrent use of drug A when you use drug B in 2012 is X.

How can I test if OR(2005) is significantly different from OR(2012)?

In both cases we use data from the complete Norwegian population. Our ORs
are a "snapshot" from each year, and as some patients use drug B in both
years, there is a partial overlap between the patient groups in these two
years.
However, I am not interested in any development from 2005 to 2012. Just to
compare two snapshots, point prevalences.

Øivind

[Øivind Hundal]

Yes, all patients have unique identifiers.

 

Øivind

---------- Forwarded message ----------
From: Bruce Weaver <[hidden email]>
Date: 2013/3/18
Subject: Re: Compare years
To: [hidden email]


Do you have unique identifiers for individuals so that you can match in cases
where people have data for both years?



Øivind Hundal wrote

> Hello to you all
>
> I calculate odds ratio for concurrent use of antidepressants when you also
> receive antidiabetic medication. Utilizing logistic regression. How can I
> compare and test difference between 2 years in these ORs? Appreciate
> syntax.
>
> E.g.: OR for concurrent use of drug A when you use drug B in 2005 is X,
> and
> OR for concurrent use of drug A when you use drug B in 2012 is X.
>
> How can I test if OR(2005) is significantly different from OR(2012)?
>
> In both cases we use data from the complete Norwegian population. Our ORs
> are a "snapshot" from each year, and as some patients use drug B in both
> years, there is a partial overlap between the patient groups in these two
> years.
> However, I am not interested in any development from 2005 to 2012. Just to
> compare two snapshots, point prevalences.
>
> Øivind

-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Compare-years-tp5718788p5718794.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

Thanks for clarifying.  Here are a few points to consider.  

"Just to compare two snapshots, point prevalences."

1. Given that you have a mix of people with data for both years and others with data for only one of the years, I don't think there is any straightforward way, unproblematic to make the comparison with a single test.  See point 3, however.

2. Prevalence is a proportion, so it's not clear to me why you want to compare odds ratios.  For patients with data for both years, why not just use the McNemar chi-square?  For the others with data for only one of the years, use Pearson's chi-square.  

3. The McNemar and Pearson chi-squares should be independent of each other.  The sum of independent chi-squares is also a chi-square with df = the sum of the df values for each of the independent chi-squares.  In this case, the McNemar and Pearson chi-squares each have df=1, so the sum has df=2.

4. Before trying to publish anything using this method, I'd suggest you wait to hear what other list members have to say about it -- or even what I have to say about it after I've thought about it a little more!  ;-)

5. Given that you are likely dealing with huge numbers of patients, isn't the more important question whether the difference is practically important?  (With huge samples, even tiny, inconsequential differences will be statistically significant.)

HTH.

Øivind Hundal wrote

Yes, all patients have unique identifiers.

Øivind

---------- Forwarded message ----------
From: Bruce Weaver <[hidden email]>
Date: 2013/3/18
Subject: Re: Compare years
To: [hidden email]


Do you have unique identifiers for individuals so that you can match in
cases
where people have data for both years?



Øivind Hundal wrote
 > Hello to you all
>
> I calculate odds ratio for concurrent use of antidepressants when you also
> receive antidiabetic medication. Utilizing logistic regression. How can I
> compare and test difference between 2 years in these ORs? Appreciate
> syntax.
>
> E.g.: OR for concurrent use of drug A when you use drug B in 2005 is X,
> and
> OR for concurrent use of drug A when you use drug B in 2012 is X.
>
> How can I test if OR(2005) is significantly different from OR(2012)?
>
> In both cases we use data from the complete Norwegian population. Our ORs
> are a "snapshot" from each year, and as some patients use drug B in both
> years, there is a partial overlap between the patient groups in these two
> years.
> However, I am not interested in any development from 2005 to 2012. Just to
> compare two snapshots, point prevalences.
>
> Øivind





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
View this message in context:
http://spssx-discussion.1045642.n5.nabble.com/Compare-years-tp5718788p5718794.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Rich Ulrich]

I think that "utilizing logistic regression" is a red herring,
or a mistake.  You are looking at treated diabetics who are or
are not using antidepressants.

"Ignoring any development from 2005 to 2012" seems to
say that you are not interested in the matches between time.

In that case, you have a 2x2  contingency table, where the total
N is the sum of "receiving antidiabetics" at 2005 and 2012.  No
one else is in the sample.

That is the comparison of two snapshots.  This is a 2x2
Crosstab.  If the N (diabetic) for each year is about the same, the
OR will be the ratio of the N's who are taking antidepressants.

More sophisticated testing would pull out the matches and look
at the "changes" (McNemar's test) separately from the people who
are only in one year of data.  Or do something with similar effect.

--
Rich Ulrich

---

Date: Mon, 18 Mar 2013 09:49:20 +0100
From: [hidden email]
Subject: Compare years
To: [hidden email]

Hello to you all
 
I calculate odds ratio for concurrent use of antidepressants when you also receive antidiabetic medication. Utilizing logistic regression. How can I compare and test difference between 2 years in these ORs? Appreciate syntax.
 
E.g.: OR for concurrent use of drug A when you use drug B in 2005 is X, and OR for concurrent use of drug A when you use drug B in 2012 is X.
 
How can I test if OR(2005) is significantly different from OR(2012)? 
 
In both cases we use data from the complete Norwegian population. Our ORs are a "snapshot" from each year, and as some patients use drug B in both years, there is a partial overlap between the patient groups in these two years.
However, I am not interested in any development from 2005 to 2012. Just to compare two snapshots, point prevalences.
 
Øivind