Pertussis model: imperfect vaccines and silent infections can sustain disease, but fitted parameters are not identifiable

The model can show how pertussis persists, but not which parameters are true
The fitted transmission rate and reporting fraction cannot be uniquely recovered from eleven years of US case counts.
Mark

Why does a model like this matter if it can't even reproduce recent US case counts?

Mimi

Because its main contribution is structural. It puts imperfect vaccines, breakthrough infections, silent transmission and fading immunity into one framework, and shows how disease can persist even when vaccination is high.

Luke

But persistence is shown inside the model. It doesn't prove those mechanisms explain what actually happened in the US.

Mark

What does 'not identifiable' mean here?

Mimi

The transmission rate and reporting fraction can't be uniquely recovered from eleven yearly counts. Different parameter sets fit about equally well, including ones where the vaccination-adjusted number is above 1.

Luke

And the 0.80 baseline is just one local optimum. Nothing in the paper says it is the true value.

Mark

Does the model miss the recent trends?

Mimi

Yes. It doesn't capture the 2017–2019 plateau or the 2023–2024 resurgence, so the authors treat it as a reference scenario only.

Mark

And the intervention rankings?

Mimi

Strategies including transmission reduction scored best, and that held when weights were halved or doubled.

Luke

Hold that loosely. The size of the benefit depended on assumed control bounds and the calibration scenario, and the costs are assumptions. It's a ranking under stated conditions, not evidence about real programs.

  • Whooping cough keeps finding footholds in communities where vaccination is widespread, and the team set out to explain why.
  • Their six-group model traces how silent carriers and waning immunity can keep the disease circulating despite high coverage.
  • When the researchers tried to fit the model to eleven years of US case counts, several different parameter sets matched the data equally well, leaving the calibration unreliable.
  • The calibrated model misses both the 2017–2019 plateau and the 2023–2024 resurgence, so it is kept as a reference rather than a forecast.
  • Simulated control strategies that cut transmission scored best, but those rankings rest on assumed costs and limits and await age, seasonal, and economic evidence before informing policy.

Pertussis, the highly contagious whooping cough, has never fully left populations that vaccinate widely, and a team of modelers has tried to explain why by building a mathematical picture of the disease in six linked groups, including people who carry it without showing symptoms. Their work is careful about its own limits: the picture can show how the illness might linger among the vaccinated, but the data available cannot fix its inner settings with confidence. The result is offered less as a map toward action than as a lens for understanding persistence, with its conclusions held open until age, seasons, and richer records are brought in.

A team of modelers set out to understand why pertussis, the respiratory illness known as whooping cough, refuses to disappear even where vaccines are common. They built a mathematical model that divides a population into six groups, among them the vaccinated, the infected who show symptoms, and those who carry the bacterium silently. The model accounts for four realities: vaccines are imperfect, vaccinated people can still become infected, people without symptoms can still spread the disease, and immunity fades over time.

The mathematics reveals when the disease can take hold and when it dies out, and sensitivity tests point to vaccination, recovery rates, and the share of silent infections as the strongest levers. Yet the authors are candid about the model's weak point. When they tried to fit it to eleven years of reported US cases, two key parameters could not be reliably determined, since many different settings matched the data equally well. The calibrated version also fails to capture the plateau of 2017 to 2019 or the resurgence of 2023 and 2024.

For that reason the model serves as a reference scenario, not a forecasting tool. When the team explored intervention strategies, those that reduced transmission consistently produced the best outcomes, though the scale of benefit depended on assumptions about costs and limits that were chosen rather than measured. The researchers say age structure, seasonal patterns, richer surveillance data, and economic evaluation are needed before any of this could guide policy. For now, the work is a framework for thinking about persistence, not a guide to action.

A team of modelers has built a mathematical picture of pertussis, a highly contagious respiratory disease caused by Bordetella pertussis, and then tested how far that picture can be trusted. Their answer, stated openly in the paper's own title, is that the model can show how the disease might persist in a vaccinated population, but that its fitted parameters cannot be pinned down from the data.

The model is deterministic and divides a population into six groups: susceptible, vaccinated, exposed, symptomatic infectious, asymptomatic infectious, and recovered. The authors call it SVEIAR. It builds in four features meant to capture why whooping cough has not disappeared where vaccination is widespread: imperfect vaccines, breakthrough infections, transmission by people who show no symptoms, and immunity that does not last.

Using the next-generation matrix method, the researchers derived two thresholds: the basic reproduction number without vaccination, and a vaccination-adjusted number at the vaccinated disease-free equilibrium. The disease-free state is locally stable if and only if the vaccination-adjusted number is below 1, and globally stable if the basic number is below 1. When the adjusted number exceeds 1, a unique endemic equilibrium exists. An exact quadratic criterion shows the transition at the threshold is a forward bifurcation for every parameter value drawn from literature ranges, though an imperfect vaccine can produce a backward bifurcation outside those ranges. Stability of the endemic state was established near the threshold using centre-manifold analysis and checked numerically further away.

Sensitivity analysis found that vaccination, recovery rates, how infectious asymptomatic people are, and the proportion of infections that are asymptomatic were the main drivers of the vaccination-adjusted reproduction number.

The weak point is calibration. The authors fitted the transmission rate and a reporting fraction to eleven annual counts of reported US cases from the CDC, covering 2014 to 2024. Those two parameters turned out not to be identifiable. The fitting objective has several local minima. The reported baseline, with a vaccination-adjusted number of 0.80, is only a local optimum, and parameter sets with that number above 1 fit the data at least as well. The calibrated model also fails to reproduce the 2017–2019 plateau or the 2023–2024 resurgence, with R-squared below zero. The authors therefore use it as a reference scenario and explicitly not for forecasting.

They then extended the model into an optimal control framework with four time-dependent interventions, using cost weights, control bounds and a time horizon that are stated and assumed rather than measured. Strategies that included reducing transmission produced the lowest objective values, and that ranking held when each weight was halved or doubled. The size of the benefit, however, depended mainly on the assumed control bounds and on which calibration scenario was used.

The authors stress that these rankings are findings conditional on their assumptions, not predictions of how interventions would work in the real world. They say age structure, seasonality, stochasticity, richer surveillance data and economic evaluation are needed before results like these could inform policy. For now the work is a framework for exploring why pertussis persists in highly vaccinated populations, not a guide to action.

The rankings are model-based findings under stated assumptions, not predictions of real-world intervention effectiveness.
— Study authors
Quieres la nota completa? Lee el original en Nature ↗
Contáctanos FAQ