Cross-over RCT and Linear mixed model

[Atnamis Roy]

Let's assume I want to see if the acute effect of smoking on heart rate varies according to flavor (menthol vs. no flavor) in two groups (hypertensive vs. non-hypertensive).
In a within-subject randomized (according to flavor) controlled trial, the change in heart rate of the 20 individuals was measured upon using menthol cigarettes and again with non-menthol cigarettes.

Considering the within and between-subject variability in heart rate and other confounders, I considered a linear mixed model to see if the change is bigger upon using menthol cigarettes in hypertensive individuals.

Is it the correct code for Stata? If not, can you please suggest the correct one? And what other things do I need to consider here?

xtmixed heart_rate hypertension var1 var2 || flavor:||individual_id:

Please note, the data is in long format.
Thanks in advance.

[Joseph Coveney]

Is it the correct code for Stata?

1. flavor isn't a random effect and study participants aren't nested under it.

2. What are var1 var2 and what are they doing in a crossover study design? Is one the participant's preferred cigarette flavor? (I assume that only smokers are eligible for enrollment.)

3. Are you measuring heart_rate or change in heart rate?

4. Why are you issuing the command under an obsolete name?

If not, can you please suggest the correct one? And what other things do I need to consider here?

Take a look at list of options (sequence, period, carryover) in the help file for pkcross for starters. Crossover study designs can get involved.

I'm kinda curious about the biological underpinnings for the research question, and how you came up with a sample size of 20.

[Atnamis Roy]

Thanks for replying.
Sorry for not clarifying. The study design was initially designed to see the effect of flavor on heart rate. However, I am currently interested in a sub-group analysis by examining the between-subject change in two groups ( hypertensive vs non-hypertensive) with data from this study; so, I need to adjust for other factors that might influence the findings since this is between subjects.

1. The random effect is the individual_id. However, I am not sure how to specify flavor. Can you please suggest?

2. The study was primarily designed to see the effect of flavor on heart rate. However, I am interested in seeing the between subject change in two groups ( hypertensive vs non-hypertensive); So, I need to adjust for other factors that might influence the findings since this is between subjects.

3. Change in heart rate.

4. The variable names are hypothetical. I am not sure what you meant. Can you please explain a little?


Menthol can influence the bioavailability of nicotine, which should be reflected in heart rate and other parameters. However, the two groups (hypertensive vs non-hypertensive) are just hypothetical examples.
Sorry for the typo; the sample size is 200.

N.B: Some part of the design in this post is used hypothetically. I am a new Stata user, so I am interested in understanding the codes for these circumstances.

[Joseph Coveney]

The study design was initially designed to see the effect of flavor on heart rate. However, I am currently interested in a sub-group analysis by examining the between-subject change in two groups ( hypertensive vs non-hypertensive) with data from this study . . . I am not sure how to specify flavor. Can you please suggest?

Because this is presumably an opportunistic hypothesis-generating sift through what is now a convenience sample, you could keep the modeling simple. You mention that the dataset is arranged "long", and so something like the following, where heart_rate is actually change of heart rate.

mixed heart_rate i.hypertension##i.flavor || individual_id:
contrast hypertension#flavor

Because it's exploratory, I suggest liberal use of graph commands.

The study was primarily designed to see the effect of flavor on heart rate. However, I am interested in seeing the between subject change in two groups ( hypertensive vs non-hypertensive); So, I need to adjust for other factors that might influence the findings since this is between subjects.

If the data are balanced, then participant characteristics (time-invariant covariates) won't affect the interaction term.

Menthol can influence the bioavailability of nicotine, which should be reflected in heart rate and other parameters.

So what do the results (primary analysis of the primary outcome) of the study as initially designed say?

However, the two groups (hypertensive vs non-hypertensive) are just hypothetical examples. . . .N.B: Some part of the design in this post is used hypothetically.

I don't follow that, but okay.