Skip to contents

survivalMPL

Penalised maximum likelihood fits for Cox proportional hazards models with right, left, and interval censoring. Baseline hazards are smoothed via flexible bases (step, Gaussian, M-splines, Epanechnikov) with non-negativity enforced during optimisation.

Features

  • Cox PH models with right, left, and interval censoring
  • Smoothed baseline hazard with multiple basis choices
  • Penalised likelihood with automatic (marginal likelihood) or fixed smoothing
  • Inference for coefficients, baseline hazard/survival, predictions, and residuals

Installation 

install.packages("survivalMPL")

Quick start

A right-censored fit on survival::lung:

library(survivalMPL)
library(survival)

lung <- na.omit(lung[, c("time", "status", "age", "sex", "ph.karno", "wt.loss")])

fit_lung <- coxph_mpl(
  Surv(time, status == 2) ~ age + sex + ph.karno + wt.loss,
  data = lung,
  tol  = 1e-05
)

summary(fit_lung)
plot(predict(fit_lung, type = "survival"))

Because coxph_mpl() estimates the baseline hazard rather than profiling it out, predict() returns absolute survival and hazard estimates with standard errors.

See the Getting Started article for installation notes, the coxph_mpl() interface, how each censoring scheme is encoded, and the control arguments. The remaining articles work through one censoring scheme at a time and compare the modelling choices.

Resources