Skip to contents

Time-dependent Cox Proportional Hazards Analysis with Splines and Diagnostics

Source: R/cox_analysis.R
analysis_cox.Rd

Fits a Cox proportional hazards model with time-dependent exposure, optional spline-modeled continuous covariates, and categorical covariates. The function constructs start-stop (counting process) survival data, applies reference level releveling, fits the model, and returns diagnostic plots and proportional hazards tests.

Usage

analysis_cox(
  dpop,
  data_dates,
  data_socioeconomic,
  normal_vars = c("edu"),
  spline_vars = c("age_bs", "bmi"),
  reference_values = list(bmi_cat1 = "Healthy Weight", bmi_cat2 = "Healthy Weight", edu =
    "1 - Low"),
  censoring_date = as.Date("2024-12-31")
)

Arguments

dpop

A data frame containing population-level variables including at minimum: `ID`, `DATE_BIRTH`, `DATE_DEATH`, `DATE_MIGRATION`, `exp.DATE`, and `resp.DATE`.

data_dates

A data frame containing baseline date information. Must include `ID` and `vpvmbl` (baseline date).

data_socioeconomic

A data frame containing socioeconomic or additional covariates (e.g., education).

normal_vars

Character vector of covariates to include in the model as standard (non-spline) terms.

spline_vars

Character vector of continuous variables to be modeled using cubic B-splines (`splines::bs()`).

reference_values

Named list specifying reference levels for categorical variables. Used internally for releveling.

censoring_date

Administrative censoring date (Date).

Value

A list containing:

model

Fitted `coxph` model object.

test_residual

Result of `cox.zph()` proportional hazards test.

plot_diagnostics

Schoenfeld residual diagnostic plot.

plot_splines

Named list of spline effect plots.

plot_forest

Forest plot of model estimates.

Details

Designed for registry-based longitudinal data with:

  • Baseline date (`vpvmbl`)

  • Exposure date (`exp.DATE`)

  • Response/event date (`resp.DATE`)

  • Migration and death censoring dates

The function performs the following steps:

  1. Merges baseline, population, and socioeconomic data.

  2. Calculates age at baseline and follow-up time variables.

  3. Handles exposure occurring before baseline by shifting exposure to baseline.

  4. Constructs counting-process format survival data (`Surv(tstart, tstop, event)`).

  5. Creates a time-dependent exposure indicator.

  6. Fits a Cox model using `survival::coxph()`.

  7. Produces:

    • Forest plot (`survminer::ggforest`)

    • Spline effect plots (`ggeffects::ggpredict`)

    • Proportional hazards test (`cox.zph`)

    • Schoenfeld residual plots (`ggcoxzph`)

Tied event times are handled using default `coxph` tie handling.

See also

coxph, ggforest, ggcoxzph