Skip to contents

Cox Proportional Hazards Regression Table

Source: R/MakeCoxRegressionTable.R
MakeCoxRegressionTable.Rd

Fits one Cox proportional hazards model per event-predictor pair and returns report-ready tables plus a tidy results dataframe compatible with PlotForestFromTable().

Usage

MakeCoxRegressionTable(
  data,
  time_var,
  event_vars,
  predictor_vars,
  covariates = NULL,
  design = NULL,
  reference_levels = NULL,
  Standardize = FALSE,
  FDR = TRUE,
  FDRAlpha = 0.05,
  CheckPH = TRUE,
  ReturnModels = FALSE,
  Relabel = TRUE,
  TreatOrdinalAs = "Categorical"
)

Arguments

data

Data frame containing the time, event, predictor, and covariate variables. When design is supplied, data provides the variable-label contract and the design variables provide the modeling data.

time_var

Character scalar naming the follow-up-time variable.

event_vars

Character vector naming event indicators.

predictor_vars

Character vector naming predictors.

covariates

Optional character vector naming adjustment variables.

design

Optional survey.design whose variables contain all model variables. Models are fit with survey::svycoxph() when supplied.

reference_levels

Optional named list or named character vector giving the reference level for categorical predictors. Unspecified predictors retain their first factor level.

Standardize

Logical. If TRUE, standardize numeric predictors and covariates within each model's complete-case analysis set.

FDR

Logical. If TRUE, add FDR, adjusted across all returned rows using ApplyFDRCorrection().

FDRAlpha

Numeric threshold used for FDR-adjusted significance.

CheckPH

Logical. If TRUE, calculate the global survival::cox.zph() p-value for ordinary Cox models. Survey models return NA.

ReturnModels

Logical. If TRUE, retain fitted models in ModelSummaries.

Relabel

Logical. If TRUE, use attached variable labels.

TreatOrdinalAs

How ordered predictors and covariates are handled.

Value

A list containing:

  • FormattedTable: a report-facing gt table.

  • LargeTable: a detailed gt table.

  • Results: one row per predictor coefficient, with the tidy regression fields used by MakeUnivariateRegressionTable() plus Events, Concordance, PH_PValue, and (when requested) FDR.

  • ModelSummaries: fitted models when ReturnModels = TRUE, otherwise NULL.

  • Metadata: event coding, modeling engine, and analysis settings.

Details

Each model has the form Surv(time_var, event) ~ predictor + covariates. Without design, models are fit with survival::coxph(). When design is a survey.design, the corresponding design variables are used and models are fit with survey::svycoxph(). N and Events are unweighted analytic row counts in both cases.

Numeric events must be coded 0/1. Logical events use TRUE as the event. For two-level factors, the second factor level is the event. Character event variables are deliberately rejected so the event ordering is never implicit.

See also

MakeUnivariateRegressionTable(), PlotForestFromTable(), ApplyFDRCorrection()

Examples

lung <- survival::lung
lung$event <- lung$status == 2
attr(lung$age, "label") <- "Age"

cox_results <- MakeCoxRegressionTable(
  data = lung,
  time_var = "time",
  event_vars = "event",
  predictor_vars = c("age", "sex")
)
cox_results$FormattedTable



  

      
        
event

      		
    
HR (95% CI)
      p-value
      FDR
    
Age
1.02 (   1, 1.04)*
0.042
0.042
sex
0.588 (0.424, 0.816)**
0.0015
0.003
PlotForestFromTable(cox_results$Results)