Skip to contents

Proportional-Hazards Audit of a Cox Model

Source: R/tbl_schoenfeld_audit.R
tbl_schoenfeld_audit.Rd

Tests the proportional-hazards assumption of a fitted Cox model with scaled Schoenfeld residuals ([survival::cox.zph()]) and returns one row per model term plus the global test, with a flag for the terms whose hazard ratio appears to change over follow-up.

Usage

tbl_schoenfeld_audit(fit, transform = "km", alpha = 0.05)

Arguments

fit

A model fitted by [survival::coxph()], or a tidymodels `workflow` / parsnip `model_fit` wrapping one (unwrapped to the engine fit).

transform

Time transform passed to [survival::cox.zph()]: `"km"` (default), `"rank"`, `"identity"` or `"log"`.

alpha

p-value below which a term is flagged.

Value

A tibble with `term`, `chisq`, `df`, `p.value` and `non_proportional` (logical). The `GLOBAL` row is the joint test and is flagged like a term.

Reading the result

A small p-value means the term's effect is not constant over time. With many terms or a large sample some small departures will be flagged, so look at the Schoenfeld plot (`plot(survival::cox.zph(fit))`) before changing the model. If a term is flagged, stratify on it, interact it with a function of time (`survival::coxph(..., tt = )`) or split follow-up (`survival::survSplit()`) and report a hazard ratio per period.

See also

[cox_check()]

Examples

fit <- survival::coxph(survival::Surv(time, status) ~ age + sex + ph.ecog, data = survival::lung)
tbl_schoenfeld_audit(fit)
#> # A tibble: 4 × 5
#>   term    chisq    df p.value non_proportional
#>   <chr>   <dbl> <dbl>   <dbl> <lgl>           
#> 1 age     0.188     1   0.665 FALSE           
#> 2 sex     2.31      1   0.129 FALSE           
#> 3 ph.ecog 2.05      1   0.152 FALSE           
#> 4 GLOBAL  4.46      3   0.216 FALSE