1

Can I use a Cox regression for the following...?
I have a longitudinal dataset with five waves of data. In two of the waves, wave 3 and wave 4, I have a health risk profile variable I constructed from biomarkers (allostatic load score). I want to know the likelihood of transitioning from a low risk health profile in wave 3 to a higher risk health profile in wave 4 (my event). Is it appropriate to use a cox regression analysis with only one time interval (wave 3 to wave 4) in which the event can happen? Thanks...

Steve
  • 13
  • 4

1 Answers1

0

A cox discrete-time model is itself equivalent to a series of conditional logistic regressions (check this and this links).

Therefore, if you have only one time point, or one transition, then a standard binary logistic regression is what you want:

  1. Select all those that have "a low risk health profile in wave 3" as your sample.
  2. Code your dependent variable the "higher risk health profile in wave 4" as 1 (and zero otherwise).
  3. Specify and run a logistic regression.
LuizZ
  • 325
  • 3
  • 10
  • Ah...thank you! – Steve Nov 18 '20 at 03:13
  • @Steve, you're welcome. Please accept the answer if it worked for you. – LuizZ Nov 18 '20 at 03:29
  • My quandary is accounting for those who transitioned from medium risk to high risk (if my sample is low risk)... I was just thinking of conditional statements for coding (ifelse in R...). I would like to account for the transition from low to med or high. And from med to high. – Steve Nov 18 '20 at 03:35
  • You may check if a ordinal logistic regression fits to your data, following this tutorial: https://stats.idre.ucla.edu/r/dae/ordinal-logistic-regression/ You will see that in a ordinal logistic you have the proportional odds assumption. That is: the distance from low to medium risk is the same distance from medium to high risk. You need to check for this assumption and the above link explains to you how to do so. If the assumption does not hold, you have two options: 1. a multinomial logistic regression (https://stats.idre.ucla.edu/r/dae/multinomial-logistic-regression/), or.... -> – LuizZ Nov 18 '20 at 03:58
  • ... 2. aggregating two of the categories that are more similar and specifying a binary logistic regression (https://stats.idre.ucla.edu/r/dae/logit-regression/) – LuizZ Nov 18 '20 at 04:00
  • For conditional recoding of variables in R, you may follow this good tutorial: https://rpubs.com/prlicari13/541675 – LuizZ Nov 18 '20 at 04:03
  • 1
    I use R and have run separate logistical analysis on AL scores (health risk profile) from wave 3 and wave 4. But my real interest is who is the most likely to transition from a lower risk AL score in adolescence (wave 3) to a higher risk AL score in adulthood (wave 4). The Cox model seems perfect, but is it theoretically grounded? The study is in the context of physical deterioration over time. Is there another event history analysis that is more appropriate for one time interval? I don't need the exact time of the transition, just if there was a transition during the time interval. – Steve Nov 18 '20 at 17:35
  • I am not sure about this one. I guess that if you have just one time interval, then all your predictors will be referred in t0. This is because if the predictors are explaining outcome they should be logically prior to outcome. So what I was thinking is to run just one logistic regression, with the predictors from wave 3 (t0) and the outcome on wave 4 (t1). But you may try to run a cox model and compare the results. – LuizZ Nov 18 '20 at 21:38
  • 1
    I looked at an old study where I used a Cox regression to see how I coded the transition variable. I am going with your first suggestion and code the outcome variable (AL score) as binary. I will code a 1 if you experience a transition from a lower risk category in adolescence (wave3) to a higher risk category in adulthood (wave 4), and I'll code a 0 if you do not experience a transition from a lower risk to a higher risk category. Then I'll use a logistic regression. Thanks! – Steve Nov 18 '20 at 22:40
  • I am glad that I could help! – LuizZ Nov 18 '20 at 23:37