Hi all,

I'm working on the impact of blood pressure variability on cognitive function over time. I runned the following model:

mmse : cognitive function (MMSE test)
zcv_sbp : CV% of systolic blood pressure variability (visit-to-visit variability). Measured every 6 months.
time : from 1 to 7 (visit 1 to visit 7 every 6 months)

xi:xtmixed mmse zcv_sbp age sexe education [other confounders] time || id : time

In this model, the beta associated with mmse is -0.6, p=0.01 so I can say that whatever the time, per 1-SD increase in systolic blood pressure variability, cognitive performances are lower (-0.6).

I checked the interaction with time :

xi:xtmixed mmse c.zcv_sbp##c.time || ctrpat : time

The coefficient of c.zcv_sbp#c.time is -0.004 but p=0.78 so not at all significant.

I just wanted to be sure that I'm allowed to say that the negative effect of systolic blood pressure variability is the same over time. So patients with an elevated variability have lower cognitive performances but they don't have a greater cognitive decline over time compared to patients with a lower variability.

I was expecting a cognitive decline in this population because I've also done a cox model looking at incident dementia and patients with a high blood pressure variability have a higher risk of developing dementia.

It's weird not to be able to show that they have a greater cognitive decline over time.

Basically, I just wanted to be sure, that if the interaction like I've done is not significant, I can conclude that the effect of variability is the same over time. Just based on the non significant p value of the interaction at 0.8?

Thank you so much +++ for your valuable help.

I'm not very familiar with liner mixed models at all...

Javier