An introduction to `svycoxme`"
In svycoxme: Mixed-Effects Cox Models for Complex Samples

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Introduction

This document provides an introduction to svycoxme. This package allows users to fit mixed-effect proportional hazards models to data from complex samples. The main function, svycoxme, is similar spirit to survey::svycoxph. Point estimation is handled internally by coxme::coxme, which is passed the relevant parameters from the survey design. Extra steps are then taken for variance estimation, which is either by Taylor linearisation or replicate weighting.

Data generation

Time-to-event data can be generated using the functions one_dataset and draw_event_times. The first of these functions, one_dataset, takes a one-sided formula, plus information about the terms in the equation to generate a dataset with a column of event times. This is data generation in \cite{ripatti_}

library(svycoxme)

For example, if event times were generated

event times could be specified using the

taking one of the parameter combinations from the simulations of simple random cluster sampling without replacement,

Each observation had a vector of three fixed-effect covariates, $\bmX_{ij}={X_{1ij}, X_{2i}, X_{3i}}$ where, \begin{tabular}{ll} $X_{1ij} \sim N(0, 1),$ & \text{varied across clusters and individuals;}\ $X_{2i} \sim N(0, 1),$ & \text{constant within clusters, varied across clusters; and}\ $X_{3i} \sim Bern(0.5),$& \text{constant within clusters, varied across clusters.} \end{tabular}%

The associated fixed effects were set to $\bmbeta = (\beta_1 = 1,\ \beta_2 = -0.7,\ \beta_3 = 0.5)$ and the time-to-event for observation $j$ in cluster $i$ was generate as, \begin{align}T_{ij} = \exp(-\bmX_{ij}\bmbeta - \bmb_i)\epsilon_{ij},\end{align}% where,

% \begin{align}\epsilon_{ij} \sim \Exp(\lambda = 0.1).\end{align} % Censoring time was randomly drawn from, \begin{align} C_{ij} \sim \text{Uniform}(0, 0.8), \end{align} which gave approximately 20\% censoring. The observed time was then, \begin{align} t_{ij} = \min(T_{ij}, C_{ij}). \end{align}

put in example from chapter 2.

This is the sort of data generation from ripatti palmgren.

Example and intent with draw event times

show svycoxme functionality, along with some sampling stuff and calibration

talk about the residuals function