Maternal vaccines protect not only the pregnant person from infections but also the developing fetus. Pregnant people typically receive shots for the flu; tetanus, diphtheria, and pertussis (Tdap); and COVID-19 (1). Just last year, the Food and Drug Administration (FDA) approved the first maternal vaccine for respiratory syncytial virus (RSV).

Like Tdap or the flu, Group B Streptococcus (GBS) bacteria can cause deadly infections in infants, but there is currently no preventative vaccine. GBS bacteria colonize the lower gastrointestinal or genital tracts of more than one third of women in the United States, and during birth the infection can spread to the infant (2). Clinicians typically screen for a GBS infection during the third trimester and can administer antibiotics right before birth, but this treatment only protects infants during the first week of life; they can still develop the infection later.

“It causes around 90,000 infant deaths annually and around 50,000 stillbirths,” said Rebecca Kahn, an infectious disease epidemiologist and an Epidemic Intelligence Service (EIS) fellow at the Centers for Disease Control and Prevention (CDC). “A vaccine would be such an amazing development.”

The problem is no maternal GBS vaccines have reached Phase 3 clinical trials due to multiple reasons, including a reluctance to perform research on pregnant people. Recently, the FDA decided that rather than measure vaccine efficacy based on whether it prevented GBS disease itself, clinical trials could measure whether a maternal vaccine led to the production of enough fetal antibodies to protect against disease. But how much antibody production is enough? With no method available to answer that question yet, Kahn and her colleagues at the CDC were up for the challenge.

Using data from a large case-control study, the researchers developed a model to predict the antibody levels necessary to prevent GBS while accounting for confounding factors in the data. This new model will allow Phase 3 maternal GBS vaccine trials to begin, saving the lives of countless babies.

How did you first get interested in building statistical models related to infectious diseases?

My background is in infectious disease epidemiology and modeling. A lot of my research during my PhD involved thinking about vaccine trials during epidemics, somewhat inspired by some of the trials people conducted during the Ebola outbreak in West Africa in 2014. I was thinking through how we could be most prepared to conduct vaccine trials in emergency settings when there's a lot of uncertainty and urgency to get a vaccine out. I finished my PhD in 2020, so I was mostly working on COVID-19 at the time.

Then, I joined the EIS fellowship, and I matched with the bacterial respiratory disease group. Stephanie Schrag, who is one of our team leads, has been working on GBS for a long time and knew about my interest in infectious disease modeling and vaccine development. Since most of what I'd been working on were vaccines that were already in Phase 3 trials, she thought this could be a good opportunity for me to learn about an earlier part of the vaccine process and apply some of my modeling experience to a new question. It sounded like a really interesting project and a great opportunity to help bring this vaccine to reality.

Why has it been so challenging to develop a maternal vaccine for GBS?

People have been working on maternal vaccines for GBS for decades. There's always some concern about conducting research in pregnant people. Additionally, even though GBS does cause a large burden of disease, it is relatively rare, so to do a traditional Phase 3 trial where we’re looking at disease endpoints that would require tens of thousands of pregnant people to be enrolled isn't feasible and would be really expensive. 

A few years ago, regulators said that instead of licensing a vaccine based on comparing disease in the vaccinated and unvaccinated groups in the trial, they'd consider licensing based on an immunologic correlative protection — measuring whether the vaccine can help people create enough antibodies to prevent disease. That's really where our team has come in; we’re trying to figure out what that correlate threshold should be.

How did you figure out what that immunologic correlate threshold should be?

Traditionally, the data informing how much antibody is needed to protect against infection have come from randomized controlled trials, for example in the case of the first pneumococcal conjugate vaccines trials. But because there hasn't been a randomized controlled trial for maternal GBS vaccines, we get the antibody data from observational studies. 

A lot of people think of statistics as just sitting at the computer and working on math and methods, but we're really motivated to develop these statistics to improve public health and save lives. 
– Rebecca Kahn, Centers for Disease Control and Prevention

Our team at CDC has been running a case-control study that looks at infants with and without GBS disease who were born from pregnant people with GBS and compares the antibody levels in the children. We have a few thousand infants in the study, which allows us to estimate thresholds for both early and late onset disease and also for different serotypes of GBS. 

Because the study is observational, there are potential variables that can distort the relationship between antibody and disease. For example, preterm infants might have lower antibody levels because they have less time to receive the antibodies while they're in utero. They also might be at risk for disease for other reasons, so we can't just take the antibody data on its own. We have to think about those potential confounders. Because FDA and other regulators will be using data from these observational studies, they require that the analyses include those covariates to make sure that the estimates are unbiased and that we're getting the most accurate correlate of protection possible.

So, Nong Shang, who is the lead statistician in our team at CDC, has been working on developing this new model called the Covariate Adjusted Logit Model or CALM (3). I've been helping him test it out in simulations and then applying it to the case-control study. The model allows us to account for the confounders and covariates in the observational data and get a curve that shows an unbiased relationship between antibody levels and disease.

Did you encounter any surprising results while testing CALM on the case-control data?

Due to the careful design of the case-control study, we have detailed antibody data on thousands of infants as well as their clinical and demographic covariates that allow us to adjust for potential confounders. We looked at early versus late onset disease and looked at the different bacterial serotypes. In our results so far, we found some heterogeneity in the immunologic thresholds across serotypes and between early and late onset disease. That was not that surprising given that we've seen similar things with pneumococcal conjugate vaccines, but that was also not something we necessarily expected.

How did you feel when you presented your model to the FDA?

While there are other studies around the world evaluating GBS case-control data, ours at CDC is by far the largest one, so FDA was really relying on our study. We felt a lot of pressure to make sure we had developed our method and presented the results to the FDA in a way that could really inform their decision making. I was also nine months pregnant at the time, so I was very tired too!

What was it like being pregnant while working on this project?

Being pregnant while working on a maternal vaccine project definitely made it more real! I was waiting for the results of my GBS test, and if it was positive, I would have to make sure I got to the hospital early enough to get the antibiotics to protect my baby. I was thinking about GBS a lot, both at work and in my personal life. It definitely provided extra motivation for working on the project.

What is the current status of the project?

We shared the interim results in the fall with FDA and then in the winter with some other regulators globally. Our final sample size is larger than the interim analysis, so we're working on estimating the thresholds and putting together the results to share with FDA. We're hoping that we'll be able to share those and that they'll be able to determine correlates for the vaccine trials soon.

Could researchers use CALM to evaluate other maternal vaccines?

The great thing about CALM is that we can apply it to any dose-response relationship. We're using antibody level as the dose and GBS specific disease as the response, but it could be antibodies against any other disease and the rates of that disease. People could also use it for things like sunlight exposure and cancer risk or air pollution and mortality.

What excites you most about developing statistical models like CALM?

A lot of people think of statistics as just sitting at the computer and working on math and methods, but we're really motivated to develop these statistics to improve public health and save lives. As observational studies become more common for vaccines, ensuring that we're analyzing the studies in accurate ways to get the most accurate results is critical for improving public health preparedness.

What have you found most rewarding about having worked on this maternal GBS vaccine project?

The opportunity to hopefully save infants’ lives is by far the most rewarding piece of it. Having been pregnant and having gone through the GBS testing experience, I could see the great potential advantages of having a vaccine. It just really brought it home for me, thinking about my baby, and if we could save the lives of babies around the world, that would be amazing.

This interview has been condensed and edited for clarity.

References

1. Centers for Disease Control and Prevention. Vaccine Safety for Moms-To-Be. (2021). https://www.cdc.gov/vaccines/pregnancy/vacc-safety.html 
2. Hanna, M. and Noor, A. Streptococcus Group B. StatPearls (StatPearls Publishing, 2024).
3. Kahn, R., Shang, N., Rhodes, J., and Schrag, S. The Covariate Adjusted Logit Model: A Novel Statistical Method for Generating Immunologic Protection Thresholds and an Application to a Group B Streptococcus Case-Control Study — United States, 2010–2022. 2024 Epidemic Intelligence Service (EIS) Conference,  60 (2024).