Repeated measurements for outcome depending on covariables

marc1

Allow me to bring a dataset to your attention, which I am struggling to analyze correctly:

Study ID	Alive	Day of measurement	Variable 1	Variable 2	Variable co
A	1	0	210	500	yes
A	1	2	210	250	yes
B	1	0	151	398	no
B	1	1	151	372	no
B	1	2	151	320	yes
B	1	3	151	268	no
B	1	4	151	372	no
C	0	0	33	269	yes
C	0	1	33	203	no

 

As I do have repeated measurements for all my subjects (Study ID), all variables have been stacked for the various time points (day of measurement) . All have a nominal outcome : Alive : 1/0.

1. Variable 1 and 2 are continuous variables and may determine the outcome (alive) for my subject (Study ID)

- can you guide me which test I best used, or refer to the reference

2. Problem1: subjects die or survive at different time periods (up to the latest day of measurement) - how do I account for this ?

3. Problem2: not all subjects have repeated measurements for all days (day of measurement), some are missing (example Study ID:A) - how do I account for those ?

4. Variable Co may influence Variable 2 , which may affect outcome (Alive) - how would I account for this effect ?

 

Thanks a lot for your guidance ! I really appreciate your help in this analysis ! Marc

 

Mark_Bailey

This is not a conventional 'repeated measures' design. This is a 'survival' study. The outcome (response) is not alive or dead. Death is the 'event.' The response is the time to this event (day). The fact that some subjects are still alive means that the last time observed is a minimum life time. Such data is 'censored.' You could use 'right censored' data in JMP (only lower bound). The fact that you do not continuously monitor the subject but only once a day suggests that you should use 'interval censoring' to obtain more accurate estimates (lower and upper bound).

 

Build your table so that you have two columns for the outcome (day). The first column is the time that they were last observed alive and the second column is the time  that they were first observed dead. The fact that you observed a subject for many days is irrelevant. It is only the time to death that matters.

 

You set up your data table with one observation for each subject including the covariates for the last observation. It might look like this for the first few subjects. (If I understand your example.)

 

 

To analyze the data, select Analyze > Fit Model. Click the drop down menu for Personality and select Parametric Survival. Select the Early and Late data columns and click Y. NOTE: it is essential that the earliest time be entered first. Select the covariate data columns and use Add or Macros to make the linear predictor. It should look something like this:

 

 

Click Run.

 

I recommend that you do not use both Variable 2 and Variable co in the model together.