RBA with longitudinal data?

[ErinR]

Hello,

Is it possible to use the RBA program with longitudinal data? Although I have used the RBA program with cross-sectional data in the past, I am not sure if the program can be used with longitudinal data. Further, the imaging data I have was collected at baseline which we would like to consider as an explanatory variable, if possible.

Any help is greatly appreciated. Thank you!
Erin

[Gang]

Hi Erin,

I haven’t fully tested the situation with a within-subject factor yet, but try something like

-model ‘1+baseline+session’ -qVars ‘baseline’ -cVars ‘session’

Let me know if it works.

[ErinR]

Thank you for your reply!

More specifically, we have baseline brain data that we want to use to predict follow-up survey data. Is this possible with the RBA program? We weren’t sure if this would present issues with directionality.

[Gang]

we have baseline brain data that we want to use to predict follow-up survey data. Is this possible with the RBA program?

By “survey data” do you mean one value (behavior?) per subject? How is this associated with the longitudinal nature of your data? Do you mind sharing the information in detail: what is the outcome/response variable and what are the predictors?

We weren’t sure if this would present issues with directionality.

What directionality?

[ErinR]

We have fMRI task data from baseline and want to see if it is associated with mental health symptoms (behavioral variable) measured one year later. We’d also like to test an interaction between the fMRI data (baseline) and a measure of stressful events (one year later) on the mental health symptoms (one year later). for this interaction, would it be coded like Example 3 on the RBA page but with value*stress?

Thank you!
Erin

[Gang]

Let me try one more time.

So you have two models. For the first model,

Outcome/response variable (y): mental health symptoms (behavioral variable)
Predictor (x): baseline fMRI effects at a list of ROIs

For the second model:

Outcome/response variable (y): mental health symptoms (behavioral variable)
First predictor (x1): baseline fMRI effects at a list of ROIs
Second predictor (x2): stressful events (behavioral variable)

Is this accurate? If so, you can follow Example 3 for the interaction between the two predictors. Make sure that you properly center the two predictors before creating the interaction predictor (the product of the two predictors).

[ErinR]

Thank you for your reply.

Yes, those are the models we would like to run. This is our original script for model 1. We added the stressful events (epii) and baseline mental health symptoms (dep_1, anx_1) in as covariates.

RBA -prefix decrease_rba_covid -Subj subj -ROI ROI -Y Y
-PDP 6 1 -chains 4 -iterations 10000 -model ‘1+age+sex+epii+dep_1+anx_1+dep_2+anx_2’
-distROI ‘student’ -distSubj ‘student’ -distY ‘student’
-qVars ‘age,epii,dep_1,anx_1,dep_2,anx_2’ -cvars ‘sex’ -EOI ‘dep_2,anx_2’
-stdz ‘age,epii,dep_1,anx_1,dep_2,anx_2’
-dataTable decrease_data_table.txt

However for model 2, I am unsure of how to incorporate an interaction term between the baseline fMRI effects at a list of ROIs (Y) and stressful events (epii) to predict the mental health symptoms (dep_2,anx_2). Would the variable under ‘-Y’ in the script be our mental health variable? If so, how does that correspond to having a list of ROI labels for ‘-ROI,’ and would we enter betas for each ROI as a separate predictor (e.g., amygdalaB, insulaB), each with its own interaction term?

Thank you!
Erin

[Gang]

For your second model, let me use the mental health symptoms dep_2 as an example. Suppose that you have the following header for your data table:

Subj ROI beta epii IA dep_2
…

1. Y in the RBA script is ‘dep_2’
2. Center beta and epii properly and get the product of the two, and add it as the interaction between the two variables (column ‘IA’ above)

Does this make sense?

[ErinR]

Thank you! We were able to run the following script with dep_2 as the Y:

RBA -prefix decrease_rba_covid_intx -Subj subj -ROI ROI -Y dep_2
-PDP 6 1 -chains 4 -iterations 10000 -model ‘1+age+sex+aepii_negc+beta_c+dep_b+anx_b+aepii_beta_c’
-distROI ‘student’ -distSubj ‘student’ -distY ‘student’
-qVars ‘age,aepii_negc,beta_c,dep_b,anx_b,aepii_beta_c’ -cvars ‘sex’ -EOI ‘beta_c,aepii_beta_c’
-stdz ‘age,dep_b,anx_b’
-dataTable decrease_fu2_intx_centered.txt

*beta_c = centered beta; aepii_negc = centered stress variable; aepii_beta_c = interaction term

However, the resulting output and plots (attached) show very odd quantiles. Is there an issue with our model?

[Gang]

The model didn’t converge (Rhat > 1.05). With only 6 ROIs, it is unlikely you would be able to make inferences for 7 predictors. If possible, try something with 20 or more ROIs.