Optimization of vaccination for COVID-19 in the midst of a pandemic

Qi Luo, Ryan Weightman, Sean T. McQuade, Mateo Díaz, Emmanuel Trélat, William Barbour, Dan Work, Samitha Samaranayake, Benedetto PiccoliNetworks Heterog. Media 2022 · DOI

TL;DR

This paper develops an age-structured epidemiological model coupled with an optimization framework to determine the optimal COVID-19 vaccination strategy for minimizing deaths. Their findings suggest that prioritizing older age groups first, while also considering essential workers, is the most effective strategy for reducing mortality.

What Problem Does This Paper Address?

The COVID-19 pandemic presented a massive challenge in determining the most effective strategies for vaccination. With limited vaccine supplies, policymakers faced the critical question of who to vaccinate first. Various criteria could be used for optimization, such as minimizing deaths, achieving herd immunity quickly, or ensuring the functioning of the healthcare system. This paper tackles the problem of designing an optimal vaccination schedule specifically to minimize the total number of deaths in a population, considering the complex interplay of age, occupation, and disease transmission dynamics.

Methods

The authors employ a sophisticated approach that combines epidemiological modeling with optimal control theory. They build an augmented age-based SEIRV (Susceptible, Exposed, Infected, Removed, Vaccinated) compartmental model. Key methodological innovations include:

  • Age and Occupation Stratification: The population is divided into seven age groups, with working-age individuals further categorized into "essential" and "non-essential" workers. This allows for a more nuanced understanding of interaction patterns.
  • Realistic Infection Dynamics: The model accounts for the possibility of vaccinated individuals becoming infected (though with reduced severity) and transmitting the virus, reflecting real-world observations.
  • Optimal Control Framework: The problem of finding the best vaccination strategy is formulated as an optimal control problem. This involves minimizing a cost function (total deaths) over a defined time horizon (180 days) by adjusting control variables (vaccination rates for each group).
  • Numerical Optimization: A direct transcription method is used to discretize the optimal control problem, which is then solved using an interior-point optimization routine (Ip0pt) with automatic differentiation. This approach is chosen for its ability to handle state constraints effectively.

Datasets

The model is tuned and tested using real-life data from:

  • US Census: Used to obtain raw population information and break it down by age group for New Jersey and Florida.
  • CDC Sources: Used for estimating epidemiological constants like infection rates, latent periods, and infectious periods.
  • Bureau of Labor Statistics: Used to determine the proportion of essential workers within different age groups.
  • rt.live: A live COVID-19 tracking website used to estimate the basic reproduction number (R0) for each state.

Key Insights and Ideas

  • Age-Based Prioritization is Robust: The most striking and comforting result is that the optimal strategy consistently favors vaccinating the oldest population first, then progressing down through age groups. This aligns with the intuitive approach adopted by many countries.
  • Essential Workers Matter: While age is a primary factor, the model also highlights the importance of considering essential workers, especially those in younger age groups who may have higher interaction rates and thus act as super-spreaders if not vaccinated.
  • Vaccine Efficacy is Paramount: Regardless of the vaccination schedule, the introduction of vaccines dramatically reduces the number of deaths. Even a mild disease scenario shows significant life-saving potential with vaccination.
  • R0 is a Major Driver of Severity: The basic reproduction number (R0) has a profound impact on the number of deaths. A small increase in R0 can lead to a disproportionately large increase in fatalities in an unvaccinated population, underscoring the urgency of vaccination.
  • Model Flexibility: The framework is designed to be adaptable to different states and scenarios by adjusting parameters, making it a valuable tool for future public health planning.

Key Results

The study found that the optimal vaccination strategy, aimed at minimizing deaths, consistently prioritized older age groups. For instance, in New Jersey with R0=1.2 and 44% of workers deemed essential, the optimal solution resulted in significantly fewer deaths compared to "no vaccine" or a strategy prioritizing essential workers first. The paper presents figures and tables (e.g., Figure 3, Table 3) demonstrating that vaccination, irrespective of the specific schedule, leads to substantial reductions in mortality. For example, with R0=1.0, vaccination can save approximately 2000 lives even in a non-optimal scenario. When R0 increases to 1.2, the number of lives saved jumps to over 23,500, emphasizing the vaccine's critical role.

Limitations and Future Work

The authors acknowledge several limitations:

  • Realistic Vaccination Rollout: The model assumes a municipality can vaccinate an entire sub-population with ease, which is a less realistic assumption than the actual, often slower, vaccination rollout.
  • Definition of "Essential Worker": The definition of essential workers can be ambiguous and vary by state, requiring careful justification and sensitivity analysis.
  • Vaccine Efficacy Data: While the model incorporates reduced transmission from vaccinated individuals, there is still a lack of extensive reliable data on how vaccines affect transmission. The chosen factor of 0.3 for vaccinated-vaccinated transmission is an estimate.

Future work could involve:

  • Updating the Model: The program can be updated and adjusted to fit a multitude of situations by modifying the model's equations.
  • Exploring Other Optimization Criteria: While this paper focuses on minimizing deaths, future work could explore optimizing for herd immunity or healthcare system functioning.
  • Incorporating More Granular Data: Further refinement could involve more detailed data on individual behavior, vaccine hesitancy, and the impact of specific non-pharmaceutical interventions.

Summary updated

BibTeX
@article{luo2022optimization,
  author = {Qi Luo and Ryan Weightman and Sean T. McQuade and Mateo Díaz and Emmanuel Trélat and William Barbour and Dan Work and Samitha Samaranayake and Benedetto Piccoli},
  title = {{Optimization of vaccination for COVID-19 in the midst of a pandemic}},
  journal = {Networks Heterog. Media},
  year = {2022},
  doi = {10.3934/nhm.2022016},
}
Figure