About This Project

Chronic stress is a silent driver of health disparities across U.S. communities. We hypothesize that the CDC’s Social Vulnerability Index (SVI) can serve as a population-level proxy for chronic stress. Using machine learning and public health data, we will identify high-stress regions and predict downstream health outcomes. Our goal is to build a stress-aware AI model that helps target interventions and advance long-term health equity.

Ask the Scientists

What is the context of this research?

Chronic stress contributes to many leading causes of death in the U.S., yet it’s rarely included in predictive health models. Direct stress measures are scarce in public datasets, but the CDC’s Social Vulnerability Index (SVI) may serve as a scalable proxy—capturing stressors like poverty, housing instability, and healthcare inaccessibility. While widely used in emergency planning, SVI hasn’t been validated as a proxy for stress in health prediction. Advances in AI now allow us to test this relationship using population-level, geospatial data. Our hypothesis is that SVI can help create stress-aware AI models that more accurately predict disease burden in vulnerable areas. This approach is promising because it fills a critical data gap with an established index. Prior barriers were conceptual (no consensus on stress measurement) and technical (lack of tools for large-scale modeling). This project addresses both, with implications for more equitable health interventions.

What is the significance of this project?

This project introduces stress as a missing variable in health equity modeling by validating the CDC’s Social Vulnerability Index (SVI) as a proxy for chronic stress. SVI includes structural risk factors—like poverty, disability, and housing insecurity—that align closely with long-term stress exposure. Yet SVI has not been tested systematically as an input to AI health models. We will combine SVI with county-level health outcomes to build “stress-aware” models that detect underrecognized risk patterns. This approach offers a scalable way to guide more precise interventions for overdose, chronic illness, and maternal health disparities. By integrating public datasets that are often siloed, we aim to provide backers with an open-source framework that can be used by public health leaders, AI developers, and policymakers to make more equitable, data-driven decisions.

What are the goals of the project?

This project aims to validate the CDC’s Social Vulnerability Index (SVI) as a proxy for chronic stress in predictive health models. We will build machine learning models predicting health outcomes such as overdose, maternal complications, and chronic disease incidence using county-level data. To test the predictive value of SVI as a stress proxy, we will compare model performance with and without SVI variables included. Positive controls will include known stress-related outcomes (e.g., ER visits for anxiety or hypertension), while negative controls will include outcomes unrelated to stress (e.g., appendicitis rates). Model predictions will be validated using out-of-sample testing, AUC scores, and comparison to published benchmarks. The goal is to determine whether incorporating SVI significantly improves predictive power for stress-linked conditions—laying the groundwork for stress-aware, equity-focused public health tools.