# Modeling unobserved heterogeneity in multistate event history data using frailty and weighted survival approaches

## TL;DR

This paper explores how frailty models can improve survival analysis by accounting for hidden differences among individuals in multistate event data.

## Contribution

The study introduces individual-specific survival weights to better handle unobserved heterogeneity in multistate survival models.

## Key findings

- Frailty models effectively reduce bias in regression coefficients for multistate transitions.
- Weighted survival times improve the accuracy of survival analysis when unmeasured factors are present.
- Simulation results show reduced bias in age estimates when using weighted survival times.

## Abstract

Conventional survival analysis models typically assume that the hazard function depends solely on the baseline hazard and covariate values, overlooking unobserved factors that influence survival outcomes. In practice, however, unmeasured variables often contribute to heterogeneity among seemingly similar individuals. Frailty models offer an effective approach to account for such unobserved heterogeneity, providing a robust framework for analyzing naturally clustered survival data. This study applies frailty models to multistate event history data, emphasizing their ability to handle unobserved heterogeneity. We introduce individual-specific survival weights to adjust survival times, better reflecting the impact of unmeasured factors. These weighted survival times are critical when data exhibit bias or when standard models fail to fully capture the influence of investigated variables. Through a simulation study, we evaluate the effectiveness and performance of frailty models in a multistate framework, comparing mean, mean squared error (MSE), and bias of regression coefficients with and without frailty. For example, in the simulated dataset for age bias has reduced from -0.01 in unweighted survival time to -0.03 in weighted survival time for transition \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\tau _{12}$$\end{document}, similarly for \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\tau _{23}$$\end{document} bias has reduced from 0.01 to -0.05. Our findings underscore the importance of addressing unobserved heterogeneity in survival analysis, particularly in multistate models with weighted survival times.