Diagnose regression models

Universal diagnostic function that performs appropriate diagnostics based on the model type. For Cox models, tests proportional hazards assumption using Schoenfeld residuals and provides comprehensive Cox diagnostics. For other models, provides general diagnostic information.

Usage

br_diagnose(breg, idx = NULL, transform = "km", ...)

Arguments

breg

A regression object with results (must pass assert_breg_obj_with_results()).

idx

Index or name (focal variable) of the model(s) to diagnose. If NULL, diagnoses all models in the breg object.

transform

Character string specifying how to transform time for Cox PH tests. Options are "km" (Kaplan-Meier), "rank", "identity", or a function.

...

Additional arguments passed to specific diagnostic functions.

Value

A list containing diagnostic results for each model.

See also

Other accessors: accessors, br_predict()

Examples

# Create models
mds <- br_pipeline(
  survival::lung,
  y = c("time", "status"),
  x = colnames(survival::lung)[6:10],
  x2 = c("age", "sex"),
  method = "coxph"
)
#> exponentiate estimates of model(s) constructed from coxph method at default

# Diagnose models (includes PH testing for Cox models)
diagnostics <- br_diagnose(mds)
print(diagnostics)
#> 
#> ── Model Diagnostics Summary ───────────────────────────────────────────────────
#> 
#> ── Model: "ph.ecog" (coxph) ──
#> 
#> Sample size: 227
#> Events: 164
#> Log-likelihood: -729.23
#> Concordance: 12544, 7117, 126, 28, 0, 0.637, and 0.025
#> 
#> Proportional Hazards Test (Schoenfeld Residuals):
#> + ph.ecog: χ² = 2.054, df = 1, p = 0.152
#> + age: χ² = 0.188, df = 1, p = 0.665
#> + sex: χ² = 2.305, df = 1, p = 0.129
#> Global test: p = 0.216 - Assumption + SATISFIED
#> 
#> 
#> ── Model: "ph.karno" (coxph) ──
#> 
#> Sample size: 227
#> Events: 164
#> Log-likelihood: -735.078
#> Concordance: 12578, 7145, 65, 28, 0, 0.637, and 0.025
#> 
#> Proportional Hazards Test (Schoenfeld Residuals):
#> x ph.karno: χ² = 8.017, df = 1, p = 0.00463
#> + age: χ² = 0.478, df = 1, p = 0.489
#> + sex: χ² = 3.085, df = 1, p = 0.079
#> Global test: p = 0.0157 - Assumption x VIOLATED
#> 
#> 
#> ── Model: "pat.karno" (coxph) ──
#> 
#> Sample size: 225
#> Events: 162
#> Log-likelihood: -721.587
#> Concordance: 12343, 6957, 57, 26, 0, 0.639, and 0.025
#> 
#> Proportional Hazards Test (Schoenfeld Residuals):
#> x pat.karno: χ² = 4.226, df = 1, p = 0.0398
#> + age: χ² = 0.054, df = 1, p = 0.817
#> + sex: χ² = 2.752, df = 1, p = 0.0971
#> Global test: p = 0.0819 - Assumption + SATISFIED
#> 
#> 
#> ── Model: "meal.cal" (coxph) ──
#> 
#> Sample size: 181
#> Events: 134
#> Log-likelihood: -573.568
#> Concordance: 7761, 5080, 7, 17, 0, 0.604, and 0.029
#> 
#> Proportional Hazards Test (Schoenfeld Residuals):
#> x meal.cal: χ² = 4.65, df = 1, p = 0.031
#> + age: χ² = 0.622, df = 1, p = 0.43
#> + sex: χ² = 1.481, df = 1, p = 0.224
#> Global test: p = 0.0942 - Assumption + SATISFIED
#> 
#> 
#> ── Model: "wt.loss" (coxph) ──
#> 
#> Sample size: 214
#> Events: 152
#> Log-likelihood: -673.056
#> Concordance: 10531, 6672, 10, 22, 0, 0.612, and 0.027
#> 
#> Proportional Hazards Test (Schoenfeld Residuals):
#> + wt.loss: χ² = 0.014, df = 1, p = 0.904
#> + age: χ² = 0.508, df = 1, p = 0.476
#> + sex: χ² = 2.549, df = 1, p = 0.11
#> Global test: p = 0.391 - Assumption + SATISFIED
#>