Forest tables

[shah-in-boots]

Forest tables will help to document usually subgroup or comparison analyses. It seems like they can vary in the following ways:

• Can have multiple terms displayed from a single model, with each term being shown in different effect sizes by subgroup --> single subgroup repeated throughout
• Can have single exposure variable displayed across multiple subgroups, very typical in the larger journals (e.g. Vaccarino et al. 2021 JAMA)
• Could have areas in-between of multiple terms and multiple subgroups

Will need to come-up with a "methodology" for how to approach these on the spectrum. It should be somewhat flexible to modify as well because stylistic differences are crucial. Best option would be to let ggplot2 handle the plotting differences, and gt handle the font/color differences.

1. Multiple Terms/Exposures x Single Subgroup
2. Single Term/Exposure x Multiple Subgroup

Sub-issues that have arisen thus far:

• [ ] Convert level and level names for relabeling (e.g. whether it is for multiple levels at once or individualized)
• [ ] Spacing and sizing of the forest plot, and giving appropriate options to the end-user for customization
• [ ] Consideration of how to include or work with survival data (e.g. columns for rate differences)
• [ ] Management of confidence interval line ranges on different scales (such as log or continuous)
• [ ] Need to add flexibility for p-value
• [ ] How to create more effective formulas for this (to limit rehashing formulas)

[shah-in-boots]

Potential solution to level issues, which remains a problem in commit c1a42428, may be resolved by ignoring level labels altogether in the function. Instead, the levels may better be renamed at the dataset level? However this requires the naming and labeling to come from original dataset (instead of the mdl_tbl object).

[shah-in-boots]

Starting to review this again, as it had fallen off of development. Have created a new branch forest_by_group to work on this.

[shah-in-boots]

One consideration as we move towards forest plots is that there are multiple subtypes of forest plots. Strata-based, interaction term based, etc. Will need to consider the role of these complex plots and how they can be combined. E.g. should a survival plot be able to be combined with a forest plot?

If so, how can we do this safely? It would require a lot of checks/balances to ensure consistency of the data between two tables to merge them....

[shah-in-boots]

Interaction forest plots should work if given all the right arguments.

May need to push these into new issues

• Need to add flexibility for p-value
• How to create more effective formulas for this (to limit rehashing formulas)

Example below

Local .Rprofile detected at /Users/asshah4/projects/vlndr/.Rprofile

library(vlndr)
#> Loading required package: vctrs
#> Loading required package: tibble
#> 
#> Attaching package: 'tibble'
#> The following object is masked from 'package:vctrs':
#> 
#>     data_frame
f <- vec_c(fmls(hp + mpg ~ .x(wt) + .i(am) + cyl),
           fmls(hp + mpg ~ .x(wt) + .i(vs) + cyl)) |>
  suppressMessages()
m <- fit(f, .fn = lm, data = mtcars, raw = FALSE)
object <- model_table(linear = m, data = mtcars)

# Variables of interest for filtering are function arguments

gtbl <-
  tbl_interaction_forest(
    object,
        outcomes = list(hp ~ "Horsepower"),
        exposures = list(wt ~ "Weight"),
        interactions = list(am ~ "Transmission", vs ~ "Engine"),
        level_labels = list(am ~ c("Automatic", "Manual"),
                           vs ~ c("V8", "V6")),
        columns = list(beta ~ "Estimate", conf ~ "95% CI", n ~ "No.", p ~ "Interaction p-value"),
        axis = list(scale ~ "continuous", title ~ "Forest"),
        width = list(),
        forest = list()
  )

^{Created on 2024-01-28 with reprex v2.1.0}

[shah-in-boots]

Now technically functioning for interaction.