clinpubr (version 1.1.1)

regression_scan: Scan for significant regression predictors

Description

Scan for significant regression predictors and output results. Both logistic and Cox proportional hazards regression models are supported. The predictor variables in the model are can be used both in linear form or in restricted cubic spline form.

Usage

regression_scan(
  data,
  y,
  time = NULL,
  time2 = NULL,
  predictors = NULL,
  covars = NULL,
  cluster = NULL,
  num_to_factor = 5,
  p_adjust_method = "BH",
  save_table = FALSE,
  filename = NULL
)

Value

A data frame containing the results of the regression analysis.

Arguments

data

A data frame.

y

A character string of the outcome variable.

time

A character string of the time variable. If NULL, logistic regression is used. Otherwise, Cox proportional hazards regression is used.

time2

A character string of the ending time of the interval for interval censored or counting process data only.

predictors

The predictor variables to be scanned for relationships. If NULL, all variables except y and time are taken as predictors.

covars

A character vector of covariate names.

cluster

A character string of the cluster variable. If set, correct for heteroscedasticity and for correlated responses from cluster samples using rms::robcov().

num_to_factor

An integer. Numerical variables with number of unique values below or equal to this value would be considered a factor.

p_adjust_method

The method to use for p-value adjustment for pairwise comparison. Default is "BH". See ?p.adjust.methods. Note that the p-value adjustment is only applied column wise, not applied among all available p-values in the table.

save_table

A logical value indicating whether to save the results as a table.

filename

The name of the file to save the results. File will be saved in .csv format.

The available transforms for each predictor type are

• numerical: original, logarithm, categorized, rcs

• num_factor: original, categorized

• factor: original

• other: none

The transforms are applied as follows

• original: Fit the regression model with the original variable. Provide HR/OR and p-values in results.

• logarithm: If the numerical variable is all greater than 0, fit the regression model with the log-transformed variable. Provide HR/OR and p-values in results.

• categorized: For numerical variables, fit the regression model with the binarized variable split at the median value. For num_factor variables, fit the regression model with the variable after as.factor(). Provide HR/OR and p-values in results. If the number of levels is greater than 2, no single HR/OR is provided, but the p-value of the overall test can be provided with TYPE-2 ANOVA from car::Anova().

• rcs: Fit the regression model with the restricted cubic spline variable. The overall and nonlinear p-values are provided in results. These p-vals are calculated by anova() of rms::cph() or rms::Glm.

Details

The function first determines the type of each predictor variable (numerical, factor, num_factor (numerical but with less unique values than or equal to num_to_factor), or other). Then, it performs regression analysis for available transforms of each predictor variable and saves the results.

Examples

data(cancer, package = "survival")
regression_scan(cancer, y = "status", time = "time", save_table = FALSE)