Including frameworks of public health ethics in computational modelling of infectious disease interventions

TL;DR

This work integrates public health ethics into quantitative infectious-disease analyses by embedding ethical values into a loss function that balances aggregate clinical burden and equity across groups. The authors implement a two-group SIR with all-or-none vaccination and optimize vaccination allocations over $p_1$ and $p_2$ under Melbourne-analogous COVID-19 parameters, exploring unlimited and limited vaccine supply. They demonstrate how different ethical framings, encoded by weights $w_{EI}$ and $w_{EV}$, yield distinct optimal strategies and reveal potential trade-offs between reducing total burden and achieving equitable outcomes, including scenarios where equity in adverse effects drives different allocations. The approach provides a formal, extensible link between normative public-health principles and policy-relevant modelling, offering a principled way to assess robustness of vaccination strategies to normative choices and to illuminate value-driven policy trade-offs across outbreak contexts.

Abstract

Decisions on public health interventions to control infectious disease are often informed by computational models. Interpreting the predicted outcomes of a public health decision requires not only high-quality modelling, but also an ethical framework for assessing the benefits and harms associated with different options. The design and specification of ethical frameworks matured independently of computational modelling, so many values recognised as important for ethical decision-making are missing from computational models. We demonstrate a proof-of-concept approach to incorporate multiple public health values into the evaluation of a simple computational model for vaccination against a pathogen such as SARS-CoV-2. By examining a bounded space of alternative prioritisations of values (outcome equity and aggregate benefit) we identify value trade-offs, where the outcomes of optimal strategies differ depending on the ethical framework. This work demonstrates an approach to incorporating diverse values into decision criteria used to evaluate outcomes of models of infectious disease interventions.

Figures (6)

• Figure 1: Compartmental model diagram for the two subpopulations with imperfect "all-or-none" vaccination indicating which compartments contribute to the components of the loss function. Each group is stratified into susceptible $S$, infectious $I$ and removed $R$ (i.e. immune due to recovery.) Subscripts represent group membership and vaccination status: vaccinated but unprotected $U$, or vaccinated and protected $P$. Dashed green lines indicate the proportion initially vaccinated. The purple and orange boxes indicate the compartments contributing to adverse effects of vaccination and clinical burden.
• Figure 2: The space of loss functions used to define different ethical frameworks. The two weight coefficients $w_{\text{EI}}$ and $w_{\text{EV}}$ determine the prioritisation of three values we consider: 1) aggregate clinical burden, 2) equity of infection-induced clinical burden, and 3) equity of clinical burden from vaccine-induced adverse reactions. The three extreme ethical frameworks we use as examples in our results are indicated by the teal, red, and yellow circles at the corners of the triangle.
• Figure 3: Examples of optimal vaccination strategies when vaccine supply is unlimited. Three different ethical frameworks are shown, one with emphasis on avoiding clinical burden (a, d), one placing priority on equity of infection-induced burden (b, e), and one with emphasis on equity of burden from adverse vaccination impacts (c, f). Trajectories generated under the optimal vaccination strategy for each framework are shown in (a), (b), and (c), while the corresponding loss surfaces are shown below, in (d), (e), and (f). In (d, e, f), red dots correspond to the optimal vaccination strategy.
• Figure 4: Outcomes of optimal vaccination strategies (with unlimited vaccine supply) as functions of the ethical frameworks used to define the loss function. Total clinical burden (days in hospital) is shown in (a). The proportion of the total population vaccinated is shown in (b). The proportion of individuals vaccinated in groups 1 and 2 are shown in (c) and (d), respectively.
• Figure 5: Examples of optimal vaccination strategies when vaccine supply is limited to $20\%$ of the population. Three different ethical frameworks are shown, one with emphasis on avoiding clinical burden (a, d), one placing priority on equity of infection-induced burden (b, e), and one with emphasis on equity of burden from adverse vaccination impacts (c, f). Trajectories generated under the optimal vaccination strategy for each framework are shown in (a), (b), and (c), while the corresponding loss surfaces are shown below, in (d), (e), and (f). In (d, e, f), red dots correspond to the optimal vaccination strategy. White space in (d, e, f) corresponds to strategies that exceed the vaccine supply.