jenniferthompson/jthelpers

A package of helper functions for my own frequent use.

jthelpers's Issues

Streamline dependencies

• create_countprocess_data shouldn't import anything from dplyr, tidyr, magrittr

Update formatp() to NEJM style

"P values larger than 0.01 should be reported to two decimal places, and those between 0.01 and 0.001 to three decimal places; P values smaller than 0.001 should be reported as P<0.001. Notable exceptions to this policy include P values arising from the application of stopping rules to the analysis of clinical trials and from genetic-screening studies."

Account for data sets with no complete rows for a given ID

create_countprocess_data() breaks if

• there are no rows with complete data for a given ID
• data.set[, time.var] is missing for a given ID

Might be good to fix that. Or at least give better warnings so people know to take care of it on the front end.

Also, let people name the status variable something other than died. Confusing when the event isn't death.

Add function to make rms_model_results() into a huxtable

Originally written for DelDepLTCI

## -- Function to make huxtable from rms_model_results() df + col headers ------
make_modresults_ht <- function(resultsdf){
  ht <- resultsdf %>%
    hux() %>%
    ## Align variable name to left, all other columns to right
    set_align(row = everywhere, col = everywhere, byrow = TRUE,
              value = c("left", rep("right", ncol(resultsdf) - 1))) %>%
    ## Set p-value column to three decimal places by default
    set_number_format(row = everywhere, col = 7, value = 3) %>%
    ## Bold column headers, add line after first row
    set_bold(row = 1, col = everywhere, TRUE) %>%
    set_bottom_border(row = 1, col = everywhere, value = 1) %>%
    ## Gray out all info except the row for history of depression, our main
    ## exposure of interest
    set_text_color(
      row = grep("^hx\\.depression$", resultsdf$label, invert = TRUE),
      col = everywhere,
      value = "#838383"
    ) %>%
    ## Table should take up 90% of document width
    set_width(0.9)
}

Add clear.labels

https://stackoverflow.com/questions/2394902/remove-variable-labels-attached-with-foreign-hmisc-spss-import-functions

rms_model_results: add functionality for orm()

Should work pretty much like lrm()

rms_po_assume should take rms::orm() model objects as well as lrm

rms_model_results(): Bug fix for Gls models

Many thanks to Yonghao Pua for catching this and creating a reprex.

Currently, rms_model_results() works beautifully with ols and lrm objects but I'm wondering if you could very kindly extend its reach to Gls objects. With orm objects, rms_model_results() works reasonably well; with Gls objects, it balks with an error message: "Error in label == "d.f." : comparison (1) is possible only for atomic and list types". Below, I provide a minimal example.

rm(list=ls())
library(plyr)
library(dplyr)
library(tidyr)
library(rms)
library(JTHelpers)

# create tall-and-thin longitudinal dataframe 
longdf <- data.frame(
    expand.grid(id = seq(1:10), time = 1:5), outcome = round(runif(50, 2, 12), 0)
)

longdf <- longdf %>%
  arrange(id, time) %>%
  mutate( age = rep(round(runif(10, 20, 60),0), each = 5 ))

longdf.dd <- datadist(longdf); options(datadist = "longdf.dd")

glsmod <-Gls(outcome~ time + rcs(age,3), correlation = corCAR1 (form=~time|id), data=longdf)

glsmod; summary(glsmod); anova(glsmod)

rms_model_results(glsmod)  # throws error mesage

Error in label == "d.f." : comparison (1) is possible only for atomic and list types

shortlabels argument in rms_model_results

It would be nice if shortlabels could be extended to apply to all terms (not just interactions) if the user chooses. If that's not doable, clarify the docs.

Option to use `comparison vs reference` format instead of separate columns

eg,

comp.string
     B vs A

is more intuitive than

low    high
  A       B

create_countprocess_data() crashes when all rows are the same

😢

It's in the stop.day calculation.

rms_calc_comparisons should default to `getRatios = TRUE` for orm() fits

Make it possible for rms_model_results() to work with named vector of labels

Currently requires a data.frame with labelled variables to use variable labels instead of variable names. This creates issues with tibbles or data.frames that we want to use with tidyverse functions. Make it more flexible, to take either a data.frame or a named vector, eg c(age = "Age at consent").

rms_po_assume docs

Argument name is lrmObj but Arguments includes rmsObj.

Add Rmd template

Ideas:

• Session info/R version
• Author + date info
• ggplot theme?

Merging summary(), anova() results not working as expected, esp with categorical variables

When no labeldf is passed, unexpected results for categorical variables (this is from fit.mult.impute() with fitter = lrm, not sure if it affects others):

12                    race.comb                                      0.3    1
13  race.comb...Non.white.White   1.00  2.00 0.83 (0.43, 1.62)           <NA>