Do we need functions for working with our hierarchical results?

[ddsjoberg]

• Filtering based on the prevalence?
• Sorting them by prevalence?

[ddsjoberg]

Make separate functions we can pipe into to filter/sort.

Make this generalizable for ard_categorical() as well.

[ddsjoberg]

Question for the devs: Do we ever need to filter on anything other than the last grouping variable?

[ddsjoberg]

@bzkrouse @jtalboys take a look at your shells and list the types of filtering we may need.

[ddsjoberg]

@bzkrouse @jtalboys Would an API like this work for your needs? I think it meet our needs at Roche...

Can you let us know by Wednesday, so we can get started on the implementaton?

ard <- 
  cards::ard_stack_hierarchical(
    cards::ADAE,
    variables = c(AESOC, AEDECOD),
    by = TRTA,
    denominator = cards::ADSL |> dplyr::rename(TRTA = ARM),
    id = USUBJID,
    filter = <>, 
  )

# `filter = p > 0.05`: we keep all rows with at least one AE with a prev above 5%
# `filter = n > 10`: we keep all rows with at least one AE with a count above 10
# `filter = sum(n) > 10`: we keep all rows with where the sum of the counts in the row is above 10

here, the rows are across by groups.

[ddsjoberg]

I know we casually mentioned using followup functions, but we lost the detailed information on the grouping variables vs variables from the original call which is why they are now proposed to be a part of the primary function. Let us know by EOD if you'd like to discuss further. If not, we'll get started on the implementation tomorrow!

[jtalboys]

@bzkrouse @jtalboys Would an API like this work for your needs? I think it meet our needs at Roche...

Can you let us know by Wednesday, so we can get started on the implementaton?

ard <- 
  cards::ard_stack_hierarchical(
    cards::ADAE,
    variables = c(AESOC, AEDECOD),
    by = TRTA,
    denominator = cards::ADSL |> dplyr::rename(TRTA = ARM),
    id = USUBJID,
    filter = <>, 
  )

# `filter = p > 0.05`: we keep all rows with at least one AE with a prev above 5%
# `filter = n > 10`: we keep all rows with at least one AE with a count above 10
# `filter = sum(n) > 10`: we keep all rows with where the sum of the counts in the row is above 10
here, the rows are across `by` groups.

Hi @ddsjoberg, sorry for the delay in replying but this looks great! Would the p's and n's in your example come from the stat_label? So the user would have the flexibility to filter by any stat types in the table?

[ddsjoberg]

@jtalboys yes, users can filter on any of the statistics. The n and p would come from the stat_name, rather than the stat_label, however.