Predicting survival with newdata in model without PH

[danjamstat]

Hi @dmphillippo,

Thanks very much for all your work on this package. I'm having an issue with trying to predict survival using new data with a model where we don't assume proportional hazards.

I think the problem is that I need to specify a study and a treatment in the aux argument, but I think the function only expects a study name as a string. I'm not sure if it's possible / how to specify a treatment. If I only specify a study then I get the following error:

Error in aux_array[, , (s - 1) * n_aux + (1:n_aux)] <- aux_temp[, , grep(paste0("\\[",  : 
  number of items to replace is not a multiple of replacement length

I've tried specifying a study as e.g., Jackson2019:Len but that gives the error:

Error in `predict()`:
! All elements of `aux` must match the name of an IPD or AgD (arm-based) study in the network, or be a list of distr() distributions.

As an example with the vignette data:

ndmm_fit_nph  <- nma(
    ndmm_net,
    regression = ~(age + iss_stage3 + response_cr_vgpr + male)*.trt,
    likelihood = "mspline",
    prior_intercept = normal(0, 100),
    prior_trt = normal(0, 100),
    prior_reg = normal(0, 100),
    prior_aux = half_normal(1),
    aux_by = c(.study, .trt),
    QR = TRUE
)

predict(
   ndmm_fit_nph,
   newdata = ndmm_agd_covs[1,] %>% mutate(age = age_mean, .trt = trt),
   baseline = "Jackson2019", 
   aux = "Jackson2019",
   times = 1:3
)