Listeria kills 14% of those it infects - and pregnancy risk models are only now catching up

One in four. That is the stillbirth rate when Listeria monocytogenes crosses the placenta and reaches a fetus. The bacterium hospitalizes 86% of those it sickens, kills roughly 14%, and between 2021 and 2023 alone, contaminated ice cream, queso fresco, and enoki mushrooms were linked to five stillbirths in the United States.

Yet until now, the dose-response models guiding food safety regulators - the mathematical tools that predict how much of a pathogen causes how much illness - were never built with pregnant people in mind. A study soon to appear in Risk Analysis changes that, offering the first biologically plausible models that specifically quantify how Listeria doses translate into maternal infection and stillbirth risk.

A blind spot in food safety math

About 1,250 Americans develop listeriosis each year. Pregnancy-associated cases make up 14% of that total, but the consequences are vastly disproportionate. Pregnant individuals often experience only mild flu-like symptoms - or none at all - while the bacterium silently invades fetal tissue. Previous risk models lumped pregnant people into broader immunocompromised categories, ignoring the unique physiology, behavior, and clinical trajectory of pregnancy.

Tyler Stump, Carly Gomez, and Jade Mitchell at Michigan State University set out to close the gap. They turned to animal studies - specifically guinea pig and gerbil data - because these species share key biological traits with humans when it comes to Listeria pathogenesis, including placental structure and immune response patterns during pregnancy.

Fetal brain infection as a surrogate marker

The team's critical insight was methodological. Instead of relying solely on direct stillbirth outcomes, which are noisy and hard to quantify at lower doses, they identified fetal brain infection as a more precise indicator. In the animal data, fetal brain infection was present in every observed stillbirth and absent in every case that did not result in stillbirth. That perfect binary separation made it a powerful surrogate endpoint.

By pooling fetal brain infection data with broader stillbirth datasets, the researchers produced two models: one predicting the probability of maternal infection at a given Listeria dose, and another predicting stillbirth risk conditional on infection. Both outperformed existing models in goodness of fit.

Why generic models fall short

The distinction matters for policy. Generic immunocompromised models treat all vulnerable groups - organ transplant recipients, cancer patients, elderly individuals, pregnant people - as interchangeable in their susceptibility. But pregnancy involves a specific kind of immune modulation: the body deliberately suppresses certain immune responses to avoid rejecting the fetus, which simultaneously opens a door for intracellular pathogens like Listeria.

"Public health agencies should use population-specific models like these when developing food safety guidance rather than applying generic population estimates," Mitchell said. The team argues that as outbreaks continue to occur, sharper risk tools can inform more targeted and protective regulations.

What pregnant people can do now

Current FDA guidance already recommends that pregnant individuals avoid unpasteurized cheeses, raw sprouts, deli meats, hot dogs, and smoked seafood unless heated thoroughly. Listeria is unusual among foodborne pathogens because it multiplies under refrigeration - a cold fridge slows most bacteria but not this one. Symptoms, when they appear, can range from fever and muscle aches to nausea and diarrhea, showing up anywhere from one day to several weeks after exposure.

But the gap between individual dietary advice and population-level policy is where this study aims to make its mark. If regulators adopt pregnancy-specific dose-response curves, they could set different tolerance thresholds for foods commonly consumed during pregnancy, or adjust inspection protocols for products implicated in past outbreaks.

Limitations worth noting

The models are built on animal data - guinea pigs and gerbils, not humans. While these species were chosen for their biological relevance to human pregnancy and Listeria infection, extrapolating across species always carries uncertainty. The sample sizes in animal dose-response studies are also inherently small compared to epidemiological datasets. And the models address stillbirth specifically; other adverse pregnancy outcomes like preterm birth or neonatal infection were not modeled separately.

Still, the researchers argue that biologically grounded animal models with a clear mechanistic basis are preferable to the alternative: continuing to apply generic curves that were never designed for this population.

The study was conducted at Michigan State University's Department of Biosystems and Agricultural Engineering and supported by the Society for Risk Analysis.