Fit a Cox proportional hazards model

survtab() fits a Cox model to time-to-event data and returns a publication-style table of hazard ratios with confidence intervals and p-values. Give the follow-up time, the event indicator, and the predictors; the proportional hazards assumption is checked automatically. For outcomes without follow-up time use regtab().

Usage

survtab(
  data,
  time,
  event,
  predictors,
  design = NULL,
  conf.level = 0.95,
  d = 2,
  labels = NULL,
  style = "default"
)

Arguments

data: A data frame.
time: Follow-up time column: a bare name or string.
event: Event indicator column: a bare name or string. May be 0/1, logical, or a two-level factor whose second level is the event.
predictors: The right-hand side of the model, as a one-sided formula (~ age + sex) or a character vector of column names.
design: String. Optional study design recorded on the result.
conf.level: Number between 0 and 1. Confidence level for intervals.
d: Integer. Decimal places for hazard ratios and intervals.
labels: Named character vector of display labels for model terms, e.g. c(age = "Age (years)", sexMale = "Male sex"). Variable labels already stored in data are used by default.
style: String. A journal preset name (see list_journals()) that controls formatting such as p-values.

Value

A simtab_result of class simtab_cox. Print it to see the formatted table, convert it with as.data.frame(), or save it with export_docx(), export_pptx(), or export_xlsx(). Unrounded results are stored in $data.

Details

Statistical methods

The model is fitted with survival::coxph() using the Efron method for ties. Hazard ratios have Wald confidence intervals. The proportional hazards assumption is tested with the Grambsch-Therneau test on Schoenfeld residuals (survival::cox.zph()); when any term fails, a warning is shown with the result. Concordance and the likelihood-ratio test p-value are available from generics::glance().

Missing data

Rows with a missing time, event, or predictor are dropped. The number of rows and events analysed is shown in the table header and in model_info().

Modifying the result

coef(), confint(), vcov(), formula(), and nobs() work on the result. model_info() reports convergence, generics::tidy() returns one row per term, and ggplot2::autoplot() draws a forest plot.

Limitations

The result is a reporting table, not a fitted model: it keeps no residuals or fitted values and cannot predict. For stratified or time-varying models, survival curves, or residual diagnostics, fit survival::coxph() directly.

References

Cox, D. R. (1972). Regression models and life-tables. Journal of the Royal Statistical Society, Series B, 34(2), 187–202. doi:10.1111/j.2517-6161.1972.tb00899.x .

Grambsch, P. M., & Therneau, T. M. (1994). Proportional hazards tests and diagnostics based on weighted residuals. Biometrika, 81(3), 515–526. doi:10.1093/biomet/81.3.515 .

Examples

if (requireNamespace("survival", quietly = TRUE)) {
  # Hazard ratios for major adverse cardiovascular events
  fit <- survtab(
    epitabl,
    time = mace_time_days, event = mace_event,
    predictors = ~ age + sex + diabetes
  )
  fit

  # Concordance, likelihood-ratio test, and events analysed
  generics::glance(fit)
}
#>      n events concordance         lr_p  ties
#> 1 1500    220   0.6048674 1.443295e-06 efron