Dear all,
I am currently trying to fit a bivariate longitudinal model with autocorrelation for several clinical trials (for the validation of surrogate endpoints). Repeated measures are available for the two outcomes (True clinical endpoint (TE) and Surrogate endpoint (SE)) I want to look at. Now I want to define the autocorrelation study specific and outcome specific. From the data I know I have slight correlations between the two outcomes. Within the outcomes, I have a strong correlation between the repeated measurements. For my purposes, I think I can define the correlation with fcor and then provide the correlation matrix with the data2 argument of the brms formula function. My question is how do I use fcor and data2 correctly so that I can pass study specific and outcome specific correlation matrices.
formula_TE <- bf(TE ~ treat + time1 + time2 + (1 + treat|corr|study) + (1|id))
formula_SE <- bf(SE ~ treat + time1 + time2 + (1 + treat|corr|study) + (1|id))
model <- brm(formula_TE + formula_SE, iter = 4000, chains = 4, cors = 4, data = data),
where:
- TE, SE are the outcomes
- treat: if patient is treated or not
- time1; time2: Time points of the measuremet. Since the trajectories are not linear, fractional polynomials are used.
- study: indicator to identify the clinical study
- id: identicator to identify the patient
I scipped the autocorrelation term.
Unfortunately, I am not allowed to share the data.
- Operating System: Windows Server
- brms Version: latest version
Thanks in advance for your help.
best,
Stefan