Rethinking Mortality: A State-Based Dynamic Probabilistic Modelling Approach Using National-Scale Health Data

Oct 30, 2025·
Pramo Samarasinghe
Honours Thesis
· 1 min read
PDF
Mortality dynamics and state transitions
Abstract
Australia’s retirement income system is shifting from defined benefit to defined contribution schemes, placing longevity risk on individuals. This thesis develops a health-informed framework for mortality modelling by incorporating health-related variables from the Personal Level Integrated Data Asset (PLIDA). Using machine learning approaches like K-means clustering, survival trees, and a non-standard Markov-chain model, the research captures mortality dynamics and transitions between health cohorts. The results demonstrate that health-informed models provide more accurate and equitable survival estimates than traditional age-sex models like the Australian Life Tables (ALT).
Type
Publication
The Australian National University

Overview

[cite_start]This thesis argues that disease-centred mortality models—incorporating health conditions, medical procedures, and medication histories—offer significantly enhanced predictive accuracy for retiree mortality compared to conventional models relying solely on age and gender[cite: 102].

Key Findings

  • [cite_start]Improved Accuracy: The proposed models outperformed the Australian Life Tables (ALT) benchmark in both individual-level predictions and cohort-level forecasts[cite: 667].
  • [cite_start]Financial Impact: For a cohort of 100,000 retirees, the proportion expected to outlive their planning horizon falls from 9.9% (ALT) to 6.7% with this model[cite: 73].
  • [cite_start]National Policy: Implementing these models could decrease Age Pension expenditure by approximately $83.2 million per year by improving retirement drawdown strategies[cite: 74].

Full Thesis

The full thesis PDF is embedded below for convenient reading.

Full thesis (PDF): Download the thesis (PDF)