Support for pairwise contrasts from emmeans

[larmarange]

• to be finalised
• update doc
• add test
• update vignette
• update NEWS and README

[ddsjoberg]

hey hey @larmarange !

this all looks great!

I was doing some testing with other types of contrasts. do you think we should document in the help file how various contrasts are handled? for example, if you make a ploy contrast from a factor variable, it is ignored by this new function.

[larmarange]

I need to check. I guess that this is not covered by emmeans. But not sure

[larmarange]

It seems to work indeed. What was your exact test?

library(gtsummary)
library(broom.helpers)
#> 
#> Attachement du package : 'broom.helpers'
#> Les objets suivants sont masqués depuis 'package:gtsummary':
#> 
#>     all_continuous, all_contrasts

mod <- glm(
  response ~ stage + grade + trt,
  gtsummary::trial,
  family = binomial,
  contrasts = list(
    stage = contr.sum,
    grade = contr.helmert,
    trt = contr.poly
  )
)


mod |> 
  tbl_regression(exponentiate = TRUE, add_pairwise_contrasts = TRUE) |> 
  as_kable()

Characteristic	OR	95% CI	p-value
T Stage
T2 / T1	0.61	0.20, 1.86	0.7
T3 / T1	1.10	0.35, 3.45	>0.9
T3 / T2	1.82	0.56, 5.94	0.6
T4 / T1	0.81	0.27, 2.42	>0.9
T4 / T2	1.33	0.42, 4.18	>0.9
T4 / T3	0.73	0.23, 2.34	0.9
Grade
II / I	0.96	0.39, 2.37	>0.9
III / I	1.13	0.47, 2.76	>0.9
III / II	1.19	0.48, 2.95	0.9
Chemotherapy Treatment
Drug B / Drug A	1.24	0.67, 2.29	0.5

^{Created on 2022-11-23 with reprex v2.0.2}

[ddsjoberg]

@larmarange that is great! The contrast I looked at was the one from ordered factor variables:

library(broom.helpers)

mod <- glm(response ~ factor(stage, ordered = TRUE) + trt, gtsummary::trial, family = binomial)

mod %>%
  tidy_and_attach(exponentiate = TRUE) %>%
  tidy_add_pairwise_contrasts() %>%
  dplyr::select(term, variable, var_class, var_type, contrasts, contrasts_type, estimate) %>%
  knitr::kable()

term	variable	var_class	var_type	contrasts	contrasts_type	estimate
(Intercept)	(Intercept)	NA	intercept	NA	NA	0.4153831
factor(stage, ordered = TRUE).L	factor(stage, ordered = TRUE)	ordered	continuous	contr.poly	poly	1.0009112
factor(stage, ordered = TRUE).Q	factor(stage, ordered = TRUE)	ordered	continuous	contr.poly	poly	1.0862980
factor(stage, ordered = TRUE).C	factor(stage, ordered = TRUE)	ordered	continuous	contr.poly	poly	0.6407352
Drug B / Drug A	trt	character	dichotomous	pairwise	pairwise	1.2431465

^{Created on 2022-11-23 with reprex v2.0.2}

[ddsjoberg]

🦁

[codecov]

Codecov Report

Merging #192 (c27d9b3) into main (46e8c6d) will decrease coverage by 0.23%.
The diff coverage is 93.33%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #192      +/-   ##
==========================================
- Coverage   98.03%   97.80%   -0.24%     
==========================================
  Files          38       40       +2     
  Lines        1785     1870      +85     
==========================================
+ Hits         1750     1829      +79     
- Misses         35       41       +6     

Impacted Files	Coverage Δ
R/model_get_pairwise_contrasts.R	88.88% <88.88%> (ø)
R/tidy_add_pairwise_contrasts.R	93.75% <93.75%> (ø)
R/broom.helpers-package.R	100.00% <100.00%> (ø)
R/tidy_and_attach.R	100.00% <100.00%> (ø)
R/tidy_plus_plus.R	100.00% <100.00%> (ø)