[Paper Review] On the age-, time- and migration dependent dynamics of diseases

Ralph Brinks|arXiv (Cornell University)|Sep 6, 2012

COVID-19 epidemiological studies8 references3 citations

TL;DR

This paper extends a classical illness-death model to include age-, time-, and migration-dependent dynamics by deriving a partial differential equation (PDE) that governs the evolution of disease prevalence. The PDE enables both forward and inverse problems: predicting prevalence from incidence/mortality/remission rates, or estimating incidence from observed prevalence, with applications to chronic disease modeling under real-world demographic complexities like migration and calendar time trends.

ABSTRACT

This paper generalizes a previously published differential equation that describes the relation between the age-specific incidence, remission, and mortality of a disease with its prevalence. The underlying model is a simple compartment model with three states (illness-death model). In contrast to the former work, migration- and calendar time-effects are included. As an application of the theoretical findings, a hypothetical example of an irreversible disease is treated.

Motivation & Objective

• To generalize a classical age-structured illness-death model by incorporating calendar time and migration effects.
• To address the limitations of stationary population assumptions in traditional epidemiological models.
• To develop a practical framework for estimating disease incidence from cross-sectional prevalence data using inverse problem solutions.
• To analyze the impact of migration on disease prevalence dynamics, particularly the 'healthy immigrant effect' and data scarcity on emigrants.
• To validate the model using Monte Carlo-simulated register data under controlled conditions for both forward and inverse problems.

Proposed method

• Derives a first-order nonlinear PDE (Eq. 5) governing age- and time-dependent disease prevalence by combining partial derivatives in age and calendar time.
• Introduces a migration impact term μ = γ(1−p) − pσ to account for net migration of healthy and diseased individuals.
• Applies the method of characteristics to solve the PDE for both forward and inverse problems.
• Uses affine-linear interpolation for incidence and mortality rates between time points to approximate trends in the simulated data.
• Employs a proportional hazards assumption (Eq. 9) in the inverse problem to estimate incidence from observed prevalence.
• Validates the model using a simulated register generated via Monte Carlo methods, with analysis strictly separated from simulation inputs.

Experimental results

Research questions

• RQ1 How can a standard illness-death model be extended to include time- and migration-dependent dynamics in disease prevalence?
• RQ2 What is the mathematical form of the PDE that governs age-, time-, and migration-dependent prevalence dynamics?
• RQ3 Can the inverse problem—estimating incidence from observed prevalence—be reliably solved under realistic assumptions?
• RQ4 How sensitive is the inverse problem solution to inaccuracies in rate interpolation, particularly during trend reversals?
• RQ5 What are the practical limitations of applying the model when migration health data are incomplete or biased?

Key findings

• The derived PDE (Eq. 5) successfully models the joint dynamics of age, calendar time, and migration on disease prevalence, with a migration term μ capturing net effects.
• The forward problem accurately predicts prevalence 20 years ahead in the simulated register using only initial prevalence and interpolated rates.
• The inverse problem reconstructs incidence from observed prevalence with a relative error of approximately 1.4× when using affine-linear interpolation, indicating sensitivity to rate approximation.
• The inverse problem is ill-posed, as shown by sensitivity to small errors in input data, particularly when incidence trends reverse between time points.
• The model reveals that lack of emigrant health data—especially in systems focused only on immigrants—limits accurate migration impact estimation.
• Despite these limitations, the model remains a viable tool for incidence estimation from cross-sectional data, especially when trends are smooth and migration data are reasonably available.