Statistical procedure for velocity as a predictor

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

Posts: 3
Joined: Sat Dec 12, 2009 9:30 am

Statistical procedure for velocity as a predictor

Postby ejoffe » Sun Jul 08, 2012 12:12 pm


I am trying to evaluate the association between the velocity a biomarker rises in the blood and the length of stay in the hospital.
I have:
1) Value #1 at time 0
2) Value #2 at time #2
3) Length of stay in days.

*** Time #2 is different between the subject (between 4-24 hrs after time 0)
Velocity = (Value2-Value1) / (time2-0)

My question:
The velocity is dependent on the value #1 (i.e., a patient arriving with a very high value will have a low velocity even though his condition is bad, because the biomarker has a maximal value. On the other hand, a patient with a very low value #1 may have a high velocity even though his condition is good, because the marker can be produced but will never get to very high levels).

How do I estimate the association between velocity and my outcome variable ?

One of the options I thought of was to divide the value #1 into categories and then evaluate the velocity in each category (maybe with a Linear Mixed Model) however, I am not sure how to create the categories (there is no reason to assume that clinical meaningful cutpoints are equally distributed in the data).

Many thanks

Who is online

Users browsing this forum: No registered users and 2 guests