Pregnancy complications, such as preterm births and stillbirths, are among the most frightening and potentially devastating health challenges an expectant family can experience. However, not all families face the same risks—new research shows that certain groups are more vulnerable than others. This growing awareness is reshaping the epidemiological approach to maternal health, with teams of clinicians and researchers nationwide now collaborating to understand why these disparities exist—and what can be done to address them.
Hidden Influences
When we think about health, most of us first focus on what happens in the doctor’s office: prescriptions, treatments, surgeries. However, much of our health is shaped by factors that exist outside of medical care. These are the host factors—the social and environmental conditions that influence our well-being, from socioeconomic status to lived experiences of systemic discrimination. Together, these factors can either support or undermine health.
Recent epidemiological research is shedding light on how these social determinants of health may be tied to pregnancy outcomes. For expectant mothers, factors like low income, poor access to healthcare, and racial discrimination have been linked to a higher risk of experiencing challenges with pregnancy or delivery.
One promising area of research is the microbiome—the complex community of bacteria and microorganisms that live in and on our bodies, coexisting with our cells. Among these, the highly sensitive community of microorganisms residing in vaginas, making up the vaginal microbiome, is one target for exploration because of its sensitivity to external factors and involvement in reproductive health.
To investigate the relationship between host factors and child health, the National Institutes of Health created the Environmental Influences on Child Health Outcomes (ECHO) Consortium in 2016. The program brought together researchers from across the country, including Kimberly McKee, an associate professor of family medicine and epidemiologist at the University of Michigan. At the time of reporting, the ECHO Consortium has recruited over 107,000 participants from diverse backgrounds across the United States. Among them are 683 pregnant women who participated in a recent study connecting vaginal microbiomes and host factors, forming four cohorts to represent three distinct regions of the United States: Michigan, Atlanta, and rural Wisconsin. By continually analyzing participants from varying socioeconomic, racial, and educational backgrounds, the researchers hope to uncover how social factors may influence the vaginal microbiome and, ultimately, impact maternal and child health outcomes.
Linking the Vaginal Microbiome to Host Factors
The vaginal microbiome has previously been studied in relation to various health conditions, including pelvic inflammatory disease, fertility issues, and susceptibility to sexually transmitted infections like chlamydia, human papillomavirus, and human immunodeficiency virus. Several studies suggest that the composition of the microbiome itself may have a causal effect on patient outcomes.
The idea of using the vaginal microbiome to predict outcomes was initially celebrated because it appeared to capture random variations between patients. While previous studies have demonstrated that the vaginal microbiome can effectively be used as a screening tool to determine individual patients’ likelihood of developing disease, these studies have not yet explored how variations in the microbiome may interact with patients’ host factors.
Kimberly McKee and the researchers leading the ECHO Consortium wanted to take a deeper dive into this relationship, hoping to link the two factors and their relationship to pregnancy outcomes. “The reason that I was interested in [this project] was bringing an epidemiological lens to this microbiome work. […] When you look at the metadata or host characteristics, they’re very limited. Even race and ethnicity can often be missing,” McKee said. McKee’s concern about the lack of host-factor annotation in sequencing datasets highlights a critical question in clinical epidemiology: Are different microbiome distributions shaped by certain host factors? In other words, are variations in microbiome diversity tied to factors such as race, education, or socioeconomic status, rather than being simply random or unique to each individual?
The researchers gathered a range of data, including race, public insurance status, maternal education, age, body mass index (BMI), smoking habits, and antibiotic use, from their sample of 683 pregnant women across the United States. They also collected vaginal samples for microbial sequencing.
To study the vaginal microbiome, clinicians swabbed a sample from the region and analyzed it using a sequencing tool called 16S ribosomal RNA (rRNA) sequencing. 16S rRNA is a component of bacterial ribosomes, the cellular organelles responsible for synthesizing bacterial proteins that aren’t found in human cells. Most bacterial genomic DNA undergoes rapid evolution, making it difficult to compare sequences and infer evolutionary relationships. However, the genes for 16S rRNA evolve very slowly. This makes it ideal for identifying conserved genetic sequences shared between closely related bacteria. Moreover, because 16S rRNA contains multiple variable regions unique to specific bacterial families, it acts like a barcode that helps researchers detect the presence of particular bacterial species. By sequencing these regions, clinicians can map the diversity of bacteria within a sample, with certain identifiable patterns of bacterial distributions serving as indicators of disease risk.
Vaginal microbiomes were grouped into community state types (CSTs) based on similarities in distribution, focusing on features such as the dominant bacterial species and the relative abundance of other common vaginal species. Interestingly, the most prevalent CSTs varied significantly across cohorts. The variations in CSTs were partially attributed to the notable differences in the host factors that comprised each cohort; for example, the cohort for Atlanta was composed entirely of Black women, while fewer than one percent of the Wisconsin cohort participants were Black.
Using statistical methods, the researchers found that maternal education and self-reported race were the host factors that explained the most variation in vaginal microbiota structure during pregnancy. In contrast, antibiotic use and age accounted for less of the variation. Additionally, biological factors such as maternal BMI and hypertension did not significantly impact the vaginal microbiomes, suggesting that host factors are not only independently associated with CSTs, but may in fact be more influential than the traditional clinical factors to which physicians pay attention. McKee noted that this may be because social determinants like maternal education and race are intimately associated with other environmental factors not represented in the study, including urban living, discrimination, diet, and chronic stress, which have also been known to have profound impacts on vaginal community structure.
Machine Learning the Differences
The next critical question was whether the variation in microbial community structure also contributes to pregnancy complications. The ECHO researchers set out to rank the aspects of microbial community structures that were most predictive of the different host factors that correlate with pregnancy outcomes. To do so, they used a machine learning algorithm called random forest classification to predict host factors based on the most distinctive bacterial taxa—specific groups or categories of bacteria classified by their genetic and functional traits. As McKee hypothesized, the taxa that best predicted lower maternal education were also associated with pregnancy complications.
Identifying key taxa in the vaginal microbial can be like finding a needle in a haystack. The vaginal microbiome is incredibly diverse, and many species can coexist in close proximity. To accurately analyze and compare this complexity, researchers must collate vast amounts of data and be able to identify patterns across the microbiome. “We have ideas about what a healthy microbiome should look like, but there’s a million ways that it can look unhealthy,” McKee said. These deviations can include an overgrowth of pathogenic bacteria, a reduction in the overall diversity of microbial species, or shifts in the relative abundance of different taxa—each of which can signal potential health issues.
Interestingly, the researchers found that the taxa predictive of host factors were also drivers of broader disruptions in the makeup of the entire microbial community—referred to as community instability. This kind of widespread change is much easier to detect using newer machine learning algorithms; together, they can potentially provide an effective marker for clinicians to use to predict pregnancy complications.
Machine learning in healthcare does come with reservations, and for good reason. Despite the integration of artificial intelligence into various stages of screening and diagnosis over the past three decades, much of the underlying algorithms remain opaque to clinicians, making it challenging to assess their effectiveness across diverse patient populations. Ilana Richman, an associate professor of general internal medicine at Yale School of Medicine, highlighted the importance of transparency in machine learning tools used for clinical decision-making, explaining that when physicians can understand what the computer algorithms are focusing on, these tools become more useful in practice. “It can be difficult to know how effective an algorithm is if it’s tested in very idealized circumstances compared to the real world,” Richman said.
ECHO’s wealth of epidemiological data improves the likelihood that their machine learning algorithms can successfully predict disease outcomes in a variety of real-world populations. However, to be used as effective screening tools and positively impact clinical outcomes, screening practices themselves must be changed. “An increase in attention towards the fact that the rate of disease differs among populations means people are considering whether we should screen differently to reduce disparities,” Richman said. “The consensus right now is that more intensive screening in more vulnerable populations will be beneficial.”
Looking Ahead
The ECHO researchers have just scratched the surface of identifying how patients with different host factors are vulnerable to health disparities. Beyond pregnancy complications, ECHO is interested in identifying social and microbial signatures of adverse events throughout childhood and adolescence. This work represents one of the earliest efforts to provide proof-of-concept for a powerful hypothesis: differences in host factors lead to differences in the microbiome, which are directly tied to health outcomes. If proven, this could motivate collaborative efforts among public health officials, primary care physicians, and microbiome researchers to more effectively address and reduce health disparities in vulnerable populations.