How to use parameters of different measurement times?

data analysis modeling strategy
1

Hi, I am currently working at a project with different blood chemistry parameters. We are looking at risk prediction and want to know if our parameter of interest - lets call it BVN - is independent and improves risk prediction (including it in a multiv. model).
The problem is, that BVN is only available for timepoint 2, which is 2 years after all other assessments. I do have age, sex, and other “fixed characteristics”, but the main question would be if it is independent and “better” than other biomarkers.

Is there any way to model this time-difference in a cox-model?

Another way would be to try to interpolate the blood values, by using their timepoint 0 and timepoint 1 values, but this feels kind of wrong, and the same feeling is if I just ignore the timeproblem and just argue it in the limitations.

Just for clearness of my problem:
timepoint 0: “fixed parameters” (like age, sex, smoking,…), and blood values
timepoint 1 (4 years later): updated “fixed parameters”, and blood values
timepoint 2 (another 2 years later): updated “fixed parameters”, and BVN (but not blood values)

2

The only analysis I can think of that would be interpretablee would be a landmark analysis where you start the clock over at 2y, recharacterize the patients on all changeable covariates and use the biomarker as a baseline in that analysis.

3 Likes
3

Thanks for your fast reply :blush:
To be honest, I only heard of landmark analysis in the context of cutting the kaplan-meier curve and “starting” it again but never in the context of models.
How would this work? Its not like that, that I just build a cox-model with the “updated” covariates, include BVN and beginning the time at timepoint 2, right? How does the recharacterization works?

4

Yes it uses standard models. van Houwelingen has a nice paper about landmark analysis. Recharacterization = the data collection plan re-measured all important covariates that are capable of changing very much during the landmark period. You know how to update age, but if say blood pressure is a big predictor you’d need data collection that updates bp near the landmark time so that you can properly adjust for the biomarker at that time.

2 Likes