CODEWITHSUNDEEP
gtsummary
Amina Gaye
Amina Gaye

Reputation: 13

How do I exponentiate beta coefficients of a log-transformed outcome of a linear regression using tbl_regression()?

I ran a linear regression with a natural log-transformed outcome which I wanted to un-transform in tbl_regression() by setting exponentiate = T. Instead, an error was produced. The regression equation without exponentiating,

ln(y) = B0 + B1x,

is not easy to interpret for a 1 unit change in x. Exponentiating would result in a more interpretable form where a 1 unit change in x would result in a percent change in y:

exp(ln(y)) = exp(B0 + B1x) y = exp(B0)*exp(B1x)

Is there a way to force tbl_regression() to provide exponentiated betas to a linear regression? If not, is there an alternative?

lm_cmv <-
  lm(
    log(cmv) ~ edu + covariates,
    data = analytic_sample
  ) %>%
  tbl_regression(
    exponentiate = T
  )

Error:

Error in `tidy_and_attach()`:
! `exponentiate = TRUE` is not valid for this type of model.

Upvotes: 1

Views: 396

Answers (1)

Daniel D. Sjoberg
Daniel D. Sjoberg

Reputation: 11754

The way you're approaching the problem would usually work. But the lm() tidier specifically includes this error for the exponentiate argument.

You can still get the table you're after with a bit of work.

  1. First you need to remove the column with the unepxonentiated confidence limits.
  2. Then exponentiate the regression slope estimate, along with the unformatted confidence limits.
  3. Lastly, you re-construct the merge lower and upper bounds of the CI into a single column and set the column header.
library(gtsummary)
library(dplyr, warn.conflicts = FALSE)

tbl <-
  lm(log(mpg) ~ am, mtcars) |> 
  tbl_regression()

# run this function to see the underlying column names in `tbl$table_body`
# show_header_names(tbl)


tbl |>
  # remove character version of 95% CI
  modify_column_hide(ci) |> 
  # exponentiate the regression estimates
  modify_table_body(
    \(x) x |> mutate(across(c(estimate, conf.low, conf.high), exp))
  ) |> 
  # merge numeric LB and UB together to display in table
  modify_column_merge(pattern = "{conf.low}, {conf.high}", rows = !is.na(estimate)) |> 
  modify_header(conf.low = "**95% CI**") |> 
  as_kable() # convert to kable to display on stackoverflow
Characteristic Beta 95% CI p-value
am 1.4 1.2, 1.7 <0.001

Created on 2023-11-06 with reprex v2.0.2

Upvotes: 0

Related Questions