rm_uvsum: Output several univariate models nicely in a single table

Description

A table with the model parameters from running separate univariate models on each covariate. For factors with more than two levels a Global p-value is returned.

Usage

rm_uvsum(
  response,
  covs,
  data,
  digits = getOption("reportRmd.digits", 2),
  covTitle = "",
  caption = NULL,
  tableOnly = FALSE,
  removeInf = FALSE,
  p.adjust = "none",
  unformattedp = FALSE,
  chunk_label,
  gee = FALSE,
  id = NULL,
  corstr = NULL,
  family = NULL,
  type = NULL,
  offset,
  strata = 1,
  nicenames = TRUE,
  showN = TRUE,
  showEvent = TRUE,
  CIwidth = 0.95,
  reflevel = NULL,
  returnModels = FALSE,
  fontsize,
  forceWald
)

Value

A character vector of the table source code, unless tableOnly=TRUE in which case a data frame is returned

Arguments

response: string vector with name of response
covs: character vector with the names of columns to fit univariate models to
data: dataframe containing data
digits: number of digits to round estimates and CI to. Does not affect p-values.
covTitle: character with the names of the covariate (predictor) column. The default is to leave this empty for output or, for table only output to use the column name 'Covariate'.
caption: character containing table caption (default is no caption)
tableOnly: boolean indicating if unformatted table should be returned
removeInf: boolean indicating if infinite estimates should be removed from the table
p.adjust: p-adjustments to be performed (Global p-values only)
unformattedp: boolean indicating if you would like the p-value to be returned unformatted (ie not rounded or prefixed with '<'). Should be used in conjunction with the digits argument.
chunk_label: only used if output is to Word to allow cross-referencing
gee: boolean indicating if gee models should be fit to account for correlated observations. If TRUE then the id argument must specify the column in the data which indicates the correlated clusters.
id: character vector which identifies clusters. Only used for geeglm
corstr: character string specifying the correlation structure. Only used for geeglm. The following are permitted: '"independence"', '"exchangeable"', '"ar1"', '"unstructured"' and '"userdefined"'
family: description of the error distribution and link function to be used in the model. Only used for geeglm
type: string indicating the type of univariate model to fit. The function will try and guess what type you want based on your response. If you want to override this you can manually specify the type. Options include "linear", "logistic", "poisson",coxph", "crr", "boxcox", "ordinal", "geeglm"
offset: string specifying the offset term to be used for Poisson or negative binomial regression. Example: offset="log(follow_up)"
strata: character vector of covariates to stratify by. Only used for coxph and crr
nicenames: boolean indicating if you want to replace . and _ in strings with a space
showN: boolean indicating if you want to show sample sizes
showEvent: boolean indicating if you want to show number of events. Only available for logistic.
CIwidth: width of confidence interval, default is 0.95
reflevel: manual specification of the reference level. Only used for ordinal regression This will allow you to see which model is not fitting if the function throws an error
returnModels: boolean indicating if a list of fitted models should be returned. If this is TRUE then the models will be returned, but the output will be suppressed. In addition to the model elements a data element will be appended to each model so that the fitted data can be examined, if necessary. See Details
fontsize: PDF/HTML output only, manually set the table fontsize
forceWald: boolean indicating if Wald confidence intervals should be used instead of profile likelihood. This is not recommended, but can speed up computations. To use throughout a document use options(reportRmd.forceWald=TRUE)

Details

Global p-values are likelihood ratio tests for lm, glm and polr models. For lme models an attempt is made to re-fit the model using ML and if,successful LRT is used to obtain a global p-value. For coxph models the model is re-run without robust variances with and without each variable and a LRT is presented. If unsuccessful a Wald p-value is returned. For GEE and CRR models Wald global p-values are returned.

The number of decimals places to display the statistics can be changed with digits, but this will not change the display of p-values. If more significant digits are required for p-values then use tableOnly=TRUE and format as desired.

Examples

Run this code

# Examples are for demonstration and are not meaningful
# Coxph model with 90% CI
data("pembrolizumab")
rm_uvsum(response = c('os_time','os_status'),
covs=c('age','sex','baseline_ctdna','l_size','change_ctdna_group'),
data=pembrolizumab,CIwidth=.9)

# Linear model with default 95% CI
rm_uvsum(response = 'baseline_ctdna',
covs=c('age','sex','l_size','pdl1','tmb'),
data=pembrolizumab)

# Logistic model with default 95% CI
rm_uvsum(response = 'os_status',
covs=c('age','sex','l_size','pdl1','tmb'),
data=pembrolizumab,family = binomial)
# Poisson models returned as model list
mList <- rm_uvsum(response = 'baseline_ctdna',
covs=c('age','sex','l_size','pdl1','tmb'),
data=pembrolizumab, returnModels=TRUE)
#'
# GEE on correlated outcomes
data("ctDNA")
rm_uvsum(response = 'size_change',
covs=c('time','ctdna_status'),
gee=TRUE,
id='id', corstr="exchangeable",
family=gaussian("identity"),
data=ctDNA,showN=TRUE)