I'm trying to model correctly the data of a research were doing.
We have many indicators measured along a day in n patients. For example, we have heart rate recorded once per minute for every patient. The patients belong to 4 groups.
We want to estimate the effect of the group they belong to on the mean value and on the variability of the single patient heart rate.
What would be a statistical correct way to approach this? Should i treat each patient as a point value of mean and variance of their data or should I use each measurement in a complex model?
I would like to use a glm model, simple multivariate or more complex mixed effects, but i'm not sure how to handle them.
Would it make sense to model the dependent variable as the single patient measurement in time or I should model directly the mean and variance for every patient?
In the first case I should take in account the intra person variability, and separate it from it the effect of the group. From reading around I thought I should use a mixed effect model with the person within group as random effect and the group as fixed, but how to treat the fact that the measurement are linked IN the single person (we could say paired)? and then how can I analyze the variance in a person with this model?
The second model is easier (one less dimension), since I would already have mean and variance calculated for every person, then I would just do a regression to asses how group influence mean and variance. Very simple but I would loose a lot of power (instead of hundreds of measurements per person I would just have 40 patients really unevenly distributed).
What would you suggest?
To be noted, I excluded ANOVA because the numerosity of the values for every patient varies and even more the number of patients per group. Secondly, every patient has a different distribution of values especially between groups, and this distribution could not be said normal (sometimes are even bi o multimodal)
Ex, two subjects frm different groups, with values in time and distribution:
