Time-to-Event Landmark Analysis using an Array of Longitudinal and Survival Sub-Models

Provides a modular end-to-end framework for dynamic risk prediction based on time-to-event and longitudinal data. This allows flexible specifications for the longitudinal and survival sub-models. The 'landmaRk' package enables reproducible benchmarks of different model choices, including cross-validation to assess out-of-sample predictive performance. Methods are described in Velasco-Pardo, Constantine-Cooke, Lees and Vallejos (2026, manuscript under preparation) 'Landmarking with Latent Class Mixed Models for Dynamic Prediction of Time-to-event Data with Heterogeneous Biomarker Trajectories'.

landmaRk

Time-to-event analysis using a wide array of longitudinal and survival sub-models.

Usage	Release	Development

Introduction

Time-to-event, or survival analysis, is used to analyse the time until an event of interest occurs. Common events include hospitalisation, equipment failure, or a prisoner reoffending. Whilst classic survival methods assume model covariates are static, it is often the case that longitudinal data related to the outcome of interest are collected. Two main forms of survival analysis incorporating time-dependent covariates exist, joint models and landmarking [^1]. This package focuses on the latter.

For a set of landmark times, a survival model is fitted up to specified horizon times. At landmark times, any time-dependent covariates must be summarised. Most commonly, the last observation carried forward (LOCF) approach is used. However, a more modern approach is to instead fit a linear mixed effects model which accounts for observations being measured with error [^2]. However, any method which summarises longitudinal observations can be used. Moreover, whilst landmarking methods typically reply on Cox proportional hazards models, nearly any survival model can also be used.

Whilst packages already exist which implement landmarking, these packages implement specific longitudinal and survival models. The aim of landmaRk is to support a wide array of longitudinal and survival sub-models whilst providing a modular system allowing others to incorporate their own models.

Installation

The easiest way to install the package is from CRAN, which can be done in R via

install.packages("landmaRk")

The development version of the package can be installed via our r-universe

install.packages("landmaRk",
                 repos = c("https://vallejosgroup.r-universe.dev",
                           "https://cloud.r-project.org"))

Alternatively, the package can be built from source using remotes

# install.packages("remotes")
remotes::install_github("vallejosgroup/landmaRk", build_vignettes = TRUE)

Getting started

We recommend starting with the landmaRk vignette, which provides an overview of the package and how to use it. You can access the vignette in R by calling

vignette("landmaRk")

Alternatively, you can view the vignette online.

Contributing to landmaRk

If you are interested in contributing to landmaRk, please read our contributing guide.

References

[^1]: Rizopoulos D, Molenberghs G, Lesaffre EMEH. Dynamic predictions with time-dependent covariates in survival analysis using joint modeling and landmarking. Biometrical Journal. 2017;59(6):1261-1276. doi: 10.1002/bimj.201600238 [^2]: Paige E, Barrett J, Stevens D, et al. Landmark models for optimizing the use of repeated measurements of risk factors in electronic health records to predict future disease risk. American Journal of Epidemiology. 2018;187(7):1530-1538. doi: 10.1093/aje/kwy018