Support specifying which variables to vary by group in fit_stan_survival_model

[e5c]

The pem_survival_model_varying_coefs model allows us to test the grouping of one continuous feature (or collection of continuous features) by a categorical feature. The former is specified by formula and the latter by group_col in fit_stan_survival_model. We'd like to be able to specify which continuous features we want to group and those we do not, in just one fit_stan_survival_model instance.

[jburos]

Thanks @e5c ..

One thing I've been thinking about is how best to support this in the API for fit_stan_survival_model.

• Option 1: update patsy formula syntax to support this, perhaps analogous to lmer-style-syntax in R. Using lmer-style-syntax, the formula would read something like ~ age + bmi + (age|sex) to allow age but not bmi to vary by sex.
  • In the long run, going in this direction will allow greater flexibility (e.g. we would use a similar approach when selecting which covariates are allowed to vary over time).
• Option 2: create a new argument to capture the "varying" formula separate from the "non-varying" formula. this would require that we give two formula objects to support the above scenario, e.g.: formula_varying = 'age', formula = 'bmi'.
  • this is the simplest to implement, and possibly the easiest to understand
  • could get complicated if we have many grouping-levels, etc.

Noting here to get your feedback on these competing approaches & also noting here for benefit of others.