University of Utah researchers have utilized machine learning to reveal that simultaneous exposure to multiple air pollutants during early pregnancy significantly increases the risk of early preterm birth, challenging current single-pollutant air quality standards. Published in the Journal of Exposure Science & Environmental Epidemiology in 2026, the study analyzed data from 44,874 first-time mothers in Utah, a region frequently experiencing elevated pollution levels. Researchers focused on three primary contaminants: fine particulate matter (PM2.5), ozone (O3), and nitrogen dioxide (NO2), alongside temperature data from 2013 to 2016. To decode the complex interactions between these variables, the research team developed an epidemiological machine learning framework utilizing a self-organizing map, a neural network architecture designed to identify patterns within high-dimensional datasets. This approach enabled scientists to categorize ambient air into twelve distinct pollution clusters and track exposure patterns week by week throughout gestation. Lead author Brenna Kelly, a University of Utah Population Health Sciences Ph.D. graduate and incoming Responsible AI postdoctoral fellow, emphasized that traditional analytical methods cannot feasibly test every possible pollutant combination, making machine learning essential for mapping real-world environmental mixtures. The analysis identified late first trimester exposure to combined ozone and PM2.5 as the most critical window for adverse outcomes. Pregnant individuals exposed to this specific mixture during week 11 faced a 53 percent increased odds of early preterm birth, defined as delivery before 34 weeks of gestation. Repeated exposure across weeks 9 through 14 elevated the risk nearly threefold. Michelle Debbink, a co-author and associate professor of obstetrics and gynecology at the University, noted that early placental and vascular development makes the fetus particularly vulnerable to inflammation and hypoxic stress caused by these chemical combinations. The findings carry direct implications for environmental policy and public health monitoring. Current Air Quality Index ratings are calculated by identifying the single most harmful pollutant in a given area, often categorizing mixed, moderate-level exposures as safe. The study demonstrates that cumulative and synergistic effects can trigger serious complications, including preeclampsia and preterm delivery, even when individual pollutant concentrations remain below regulatory thresholds. Simon Brewer, a co-author and University of Utah professor, highlighted the institutional support provided by the DELPHI initiative and the university’s Responsible AI framework, which facilitated the integration of artificial intelligence into complex environmental health modeling. Researchers recommend that regulatory bodies transition toward mixture-based assessment models to better reflect actual atmospheric conditions. By applying this machine learning pipeline to other environmental hazards, public health officials can develop more nuanced exposure guidelines. The study establishes a scalable template for leveraging artificial intelligence in epidemiological research, bridging the gap between computational pattern recognition and tangible clinical outcomes in vulnerable populations.