Multinomial logistic regression question

[William Dudley-2]

I have a small data set  (< 100) with data on what drug (there are 5 drugs) was prescribed across multiple clinical visits (we have data 5 visits) and the year in which the drug was prescribed.  The data set has about 10 years of data from 2007 to 2016.  We anticipate that prescribing practices may change over clinic visit as pts disease changes AND over time (years) as new drugs come onto the market.

My client finds stacked bar charts of the percent of each drug at each clinic visit (we are looking at 5 consecutive visits) to be interesting BUT correctly points out that this only represents one variable (clinic visit).  I am wondering how to arrive at percent prescribed at each visit after controlling for the year of the visit.

Can I arrive at this with a multinomial Log regression (Drug as DV and Year as IV) at each year (Thus 5 analyses)?  If so how might I go about extracting the percent of each drug prescribed at each clinic visit after controlling for the effect of year?

Thanks in advance

Bill

--

===================== 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

[Maguin, Eugene]

I wonder if I’m misunderstanding your dataset. Is it approximately 100 people who were seen (clinic visit) 5 times in the approximate 10 year period? So, in long format approximately 500 records? So everybody was recruited in year 1? Or is this more like a continuous recruitment so that maybe 10 people per year were seen for the first time?

 

How are your data organized; what does a record look like? Like this:

Id clinic1date clinic1drug …. Clinic5date clinic5drug

 

If so, what if you extracted year from clinic(i)date so that you now have

Id clinic1date clinic1year clinic1drug …. Clinic5date clinic5year clinic5drug

Then restructure your file wide to long using varstocases.

Varstocases make year from clinic1year …. Clinic5year/

   make drug from clinic1drug … clinic5drug/

   drop clinic1date … clinic5date.

Then sort cases by id and year.

Then casestovars

Casestovars id=id/index=year.

So now you have 11 variables: id plus one drug variable for each year in the original study.

 

This will work for plotting. You can produce a set of stacked bar charts, one for each study year.

 

Gene Maguin

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of William Dudley
Sent: Saturday, November 23, 2019 1:21 PM
To: [hidden email]
Subject: Multinomial logistic regression question

 

I have a small data set  (< 100) with data on what drug (there are 5 drugs) was prescribed across multiple clinical visits (we have data 5 visits) and the year in which the drug was prescribed.  The data set has about 10 years of data from 2007 to 2016.  We anticipate that prescribing practices may change over clinic visit as pts disease changes AND over time (years) as new drugs come onto the market.

 

My client finds stacked bar charts of the percent of each drug at each clinic visit (we are looking at 5 consecutive visits) to be interesting BUT correctly points out that this only represents one variable (clinic visit).  I am wondering how to arrive at percent prescribed at each visit after controlling for the year of the visit.

 

Can I arrive at this with a multinomial Log regression (Drug as DV and Year as IV) at each year (Thus 5 analyses)?  If so how might I go about extracting the percent of each drug prescribed at each clinic visit after controlling for the effect of year?

 

 

 

Thanks in advance

 

Bill

 

 

 

 

===================== 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