Studies focusing on deteriorating neighborhoods, electronic health records, and gun trafficking highlight this month's research roundup

Yale School of Public Health researchers have developed a new approach for improving genetic risk prediction by tapping into the vast, often underused information contained in electronic health records (EHRs). The method—called Electronic Health Record Embedding Enhanced Polygenic Risk Scores (EEPRS)—integrates modern embedding techniques with traditional genome-wide association study (GWAS) data to produce more accurate and clinically meaningful predictions of disease risk.

Embedding techniques are used to turn information, such as electronic health records, into numbers that computers can easily analyze. In EEPRS, those methods include well-known applications like Word2Vec, as well as newer approaches that use large language models, such as GPT, to capture patterns in patients’ health data.

Current polygenic risk scores (PRS) rely on simplified, predefined disease categories, usually treating conditions as binary traits—case or control. But this approach overlooks the rich, multidimensional patterns that EHRs capture across thousands of diagnoses, symptoms, and clinical encounters. The new EEPRS framework addresses this gap by applying natural language processing tools, such as Word2Vec and GPT, to generate numerical representations of clinical phenotypes. These embeddings are then incorporated directly into creating risk scores, using only GWAS summary statistics.

In evaluations across 41 traits in the UK Biobank, EEPRS consistently outperformed single-trait PRS methods, with the largest gains appearing in cardiovascular-related phenotypes. The team also introduced EEPRS-optimal, which uses cross-validation to select the most effective embedding strategy for each trait, and MTAG-EEPRS, a multi-trait extension that further boosts prediction accuracy.

The manuscript was published in The American Journal of Human Genetics. Dr. Hongyu Zhao, PhD, Ira V. Hiscock Professor of Biostatistics, and Professor of Genetics and Statistics and Data Science, is corresponding author.

Lead author Leqi Xu, a doctoral candidate in biostatistics, said the work highlights the enormous potential of combining modern embedding approaches with large-scale biobank data. “By capturing the nuanced relationships embedded in electronic health records, EEPRS allows us to build more powerful and more interpretable genetic risk models that reflect the true complexity of human health,” Xu said.

If adopted widely, the EEPRS framework could help accelerate precision medicine by uncovering subtler genetic signals and improving early-risk identification across a broad range of diseases.

Journal Reference: Xu, Leqi et al (2025). Improving polygenic risk prediction performance by integrating electronic health records through phenotype embedding. The American Journal of Human Genetics. DOI: 10.1016/j.ajhg.2025.11.006

A study of older Americans led by researchers at the Yale School of Public Health suggests that long-term exposure to physically deteriorating neighborhoods may raise the risk of diabetes and inflammation. The report is one of the first to track neighborhood disorder over time and link it to biomarkers for chronic disease.

The study, published in the Journals of Gerontology: Series B, analyzed six years of neighborhood observations from the National Health and Aging Trends Study, a nationally representative cohort of Medicare beneficiaries. Interviewers assessed visible signs of disorder—such as trash, graffiti, and vacant buildings—around participants’ homes. Using latent class analysis, researchers identified four distinct patterns of exposure: stable low disorder, stable high disorder, increasing disorder, and decreasing disorder over time.

Compared with older adults living in consistently low-disorder environments, those exposed to stable high disorder had significantly higher levels of hemoglobin A1c (HbA1c)—a key marker of chronic glucose elevation—and high-sensitivity C-reactive protein (hsCRP), an indicator of systemic inflammation.

“Our findings show that the physical state of a neighborhood is not just a cosmetic issue—it can leave a measurable biological imprint on older adults,” said Postdoctoral Associate Dr. Jiao Yu, lead author of the study. “Improving neighborhood environments may be an impactful way to promote healthier aging.”

The researchers used machine-learning–based inverse probability weighting to adjust for socioeconomic, demographic, and early-life factors. Dr. Xi Chen, Associate Professor of Public Health (Health Policy) is the study’s corresponding author.

Journal Reference: Yu, J., Cudjoe, T. K. M., Mathis, W., & Chen, X. (2025). Uncovering the Biological Toll of Neighborhood Disorder Trajectories: A Machine Learning–Based Weighting Analysis of Biomarkers in Older Adults. Journal of Gerontology: Series B – Psychological Sciences and Social Sciences. DOI:10.1093/geronb/gbaf242

In a commentary published in the American Journal of Epidemiology, Yale School of Public Health Assistant Professor Lee Kennedy-Shaffer, BS ’13, warns that guns moving across state borders are weakening the impact of state-level firearm laws and challenging how researchers assess their effectiveness.

The article, “Spillovers and Effect Attenuation in Firearm Policy Research in the United States,” was co-authored by Alan H. Kennedy, BS ’06, of the College of William & Mary. The authors argue that the free flow of firearms between states—often along routes such as the “iron pipeline” on Interstate 95—creates spillover effects that dilute the benefits of stricter gun regulations.

“Policies that might work well if adopted nationwide can seem ineffective when only a few states implement them,” said Dr. Kennedy-Shaffer, a faculty member in the Department of Biostatistics. “We need better data systems and methods that account for these spillover effects so that good policies aren’t abandoned because the statistics are misleading in limited-scale pilots or single-state implementations.”

The study highlights a critical limitation of many firearm policy analyses: most assume that each state functions independently when measuring the effects of new gun laws. In reality, firearms often move easily between states, meaning that stronger restrictions in one jurisdiction can be undermined by weaker laws in neighboring states—a “bypass effect” that makes policy impacts appear smaller than they truly are.

Journal Reference: Kennedy-Shaffer, L., & Kennedy, A. H. (2025). Spillovers and Effect Attenuation in Firearm Policy Research in the United States. American Journal of Epidemiology. DOI: 10.1093/aje/kwaf220