Share your ideas for the JMP Scripting Unsession at Discovery Summit by September 17th. We hope to see you there!
Choose Language Hide Translation Bar
Highlighted
Level III

## Comparing individual groups to a baseline group

Hello,

a medical statistic problem for me:

- a discrete outcome variable - alive - 1/0

- multiple groups, which I created by binning a continuos variable per percentiles (0-25,25-50,50-75,75-100)

- for each group, I created a specific column (0-25, 25-50, 50-75, 75-100) which is again 1/0

my problem:

I want to set all the patients in group 1 (0-25) as a baseline group and compare each individual (25-50, 50-75,75-100) to the baseline group (0-25) and subsequently calculate OR in relation to the baseline group (0-25) - example attached.

 Patient Outcome 0-25% 25-50% 50-75% 75-100% A 1 0 1 0 0 B 1 1 0 0 0 C 1 0 0 1 0 D 0 0 0 0 1 E 0 1 0 0 0

Many thanks for your help and ideas ! Marc

2 ACCEPTED SOLUTIONS

Accepted Solutions
Highlighted
Super User

## Re: Comparing individual groups to a baseline group

I would setup you data table differently. 3 columns, Patient, Outcome, Percentile. Percentile would have the values "0-25", "26-50" etc. Then, make sure all of your columns are set to a Modeling Type of "Nominal". Then you can run the Fit Y by X platform, and by using a Local Data Filter, that you would set to select on Percentile, you can then select the grouping you want to compare, and JMP would perform a Chi Square analysis on the groups.
Jim
Highlighted
Level VI

## Re: Comparing individual groups to a baseline group

I echo txnelson's suggestion and would add two things.  First, consider using the percentiles ungrouped - you can always group them later, but there is no reason to lose detail in the data. Second, this type of data is often censored data - in other words, the people who have not died have not died yet.  Only if you have a context where the treatment period (or whatever time period you are using) is over, is the 0/1 response variable the correct way to view this.  Instead, you might try the survival platform (e.g., fit prorportional hazard model) - but you would need to have a continous response variable which is usually the time at which death occurs.

5 REPLIES 5
Highlighted
Level VI

## Re: Comparing individual groups to a baseline group

Hi,
I'm not sure we can help you with this level of information. Could you clarify a couple of points?
Is it correct that for each groups A - E, you have the actual frequency of death which you expressed as 1 or 0 depending on the quartile these belong to?
What is the relationship between the Outcome column and the other columns (if any)?
Most importantly, what is the scientific / medical question you need to address?
Thierry R. Sornasse
Highlighted
Level III

## Re: Comparing individual groups to a baseline group

Hi - for clarity

Column A represents individual patients - about 3000 - I just listed here just as example (A-E)

Column B - outcome - my outcome variable (Y)

Column C, D, E, F - group distribution related to their percentile distribution and binned for percentiles

Column C would be my reference group

What is the outcome (B) in regards to OR for patients in group D to C, group E to C and group F to C.

Thanks a lot for looking into this. Marc

Highlighted
Super User

## Re: Comparing individual groups to a baseline group

I would setup you data table differently. 3 columns, Patient, Outcome, Percentile. Percentile would have the values "0-25", "26-50" etc. Then, make sure all of your columns are set to a Modeling Type of "Nominal". Then you can run the Fit Y by X platform, and by using a Local Data Filter, that you would set to select on Percentile, you can then select the grouping you want to compare, and JMP would perform a Chi Square analysis on the groups.
Jim
Highlighted
Level VI

## Re: Comparing individual groups to a baseline group

I echo txnelson's suggestion and would add two things.  First, consider using the percentiles ungrouped - you can always group them later, but there is no reason to lose detail in the data. Second, this type of data is often censored data - in other words, the people who have not died have not died yet.  Only if you have a context where the treatment period (or whatever time period you are using) is over, is the 0/1 response variable the correct way to view this.  Instead, you might try the survival platform (e.g., fit prorportional hazard model) - but you would need to have a continous response variable which is usually the time at which death occurs.

Highlighted
Level III

## Re: Comparing individual groups to a baseline group

Thank you both - the local data filter made it easy for me to compare those groups.

Indeed I first run the mixed model with continuos variable (time to death), however as many even high impact medical journals tend more to binning of variables (which I personally dont support) I added this as a new feature for my article.

I also run cox proportional hazard, but I restricted my analysis now to multivariate models , which seem also better to fit using nominal variables only , as mixed (continuos and nominal variables) give me a lot of times a positive lack of fit test.

Thanks again for your help !

Article Labels

There are no labels assigned to this post.