Unexpected outputs using `add_global_p()` with `crr()`

[fdehrich]

Hello,

Re-posting this issue from {gtsummary} since it was identified as a better fit for {tidycmprsk}.

When using add_global_p() with crr(), I noticed that, in some cases, releveling a 4-level categorical variable (keeping the reference category the same) causes the global p-value to change. More broadly, in these instances, the global p-value appears to consistently match the individual p-value corresponding to the third level for some reason.

I was able to recreate this issue using the tidycmprsk::trial dataset (I had to add some extra variables and used the death variable as a predictor which doesn't make sense, but helped me recreate for some reason). Below, the global p-value for new_categorical changes from >0.9 to 0.2 when the variable is releveled (but keeping A as the reference category).

Thank you!

  library(tidycmprsk)

# Update trial data to include additional variables
trial_updated =
  trial %>% 
  dplyr::mutate(
    new_binary = sample(c(0, 1), 
                        replace = TRUE, 
                        size = nrow(trial)),
    new_categorical = sample(c("A", "B", "C", "D"), 
                             replace = TRUE, 
                             size = nrow(trial))
  )

# Generate table
crr(Surv(ttdeath, death_cr) ~ ., 
    failcode = "death from cancer",
    cencode = "censor",
    trial_updated) %>%
  gtsummary::tbl_regression(exp = TRUE) %>%
  gtsummary::add_global_p(keep = TRUE) %>% 
  gtsummary::as_kable()
#> 27 cases omitted due to missing values

Characteristic	HR	95% CI	p-value
Chemotherapy Treatment			0.7
Drug A	—	—
Drug B	1.31	0.70, 2.45	0.4
Age	1.00	0.98, 1.03	0.5
Marker Level (ng/mL)	1.14	0.78, 1.67	0.5
T Stage			0.12
T1	—	—
T2	0.67	0.23, 1.96	0.5
T3	2.03	0.83, 4.92	0.12
T4	0.93	0.37, 2.37	0.9
Grade			0.5
I	—	—
II	1.80	0.82, 3.95	0.14
III	1.14	0.59, 2.21	0.7
Tumor Response	1.08	0.49, 2.37	\<0.001
Patient Died	110,310	65,336, 186,241	0.3
new_binary	1.49	0.73, 3.07	0.2
new_categorical			>0.9
A	—	—
B	1.72	0.72, 4.06	0.2
C	1.06	0.38, 2.96	>0.9
D	1.83	0.72, 4.64	0.2

# Relevel `new_categorical` variable
trial_updated = 
  trial_updated %>% 
  dplyr::mutate(
    new_categorical = forcats::fct_relevel(
      new_categorical,
      "A", "C", "B", "D"
    )
  )

# Regenerate table
crr(Surv(ttdeath, death_cr) ~ ., 
    failcode = "death from cancer",
    cencode = "censor",
    trial_updated) %>%
  gtsummary::tbl_regression(exp = TRUE) %>%
  gtsummary::add_global_p(keep = TRUE) %>% 
  gtsummary::as_kable()
#> 27 cases omitted due to missing values

Characteristic	HR	95% CI	p-value
Chemotherapy Treatment			0.7
Drug A	—	—
Drug B	1.31	0.70, 2.45	0.4
Age	1.00	0.98, 1.03	0.5
Marker Level (ng/mL)	1.14	0.78, 1.67	0.5
T Stage			0.12
T1	—	—
T2	0.67	0.23, 1.96	0.5
T3	2.03	0.83, 4.92	0.12
T4	0.93	0.37, 2.37	0.9
Grade			0.5
I	—	—
II	1.80	0.82, 3.95	0.14
III	1.14	0.59, 2.21	0.7
Tumor Response	1.08	0.49, 2.37	\<0.001
Patient Died	110,310	65,336, 186,241	0.3
new_binary	1.49	0.73, 3.07	>0.9
new_categorical			0.2
A	—	—
C	1.06	0.38, 2.96	>0.9
B	1.72	0.72, 4.06	0.2
D	1.83	0.72, 4.64	0.2

^{Created on 2023-04-25 with reprex v2.0.2}

[gaborcsardi]

Hi, this is a read only mirror of CRAN, please see the package authors in the DESCRIPTION file. Look for Maintainer, BugReports and URL. Thanks!

[fdehrich]

Sorry about that! Just posted there now