A RIGOROUS, NATIONWIDE ANALYSIS
To build the most comprehensive picture to date of technology's impact on population health, we merged two of the nation's most robust datasets. Our analysis covers 3,143 U.S. counties, 6,166 hospitals, and a population of over 343 million people. The sheer scale of this study provides a powerful foundation for identifying significant, nationwide patterns.
American Hospital Association (AHA) Annual Survey (2023 data): This is the definitive source for understanding hospital operations across the U.S. We used this self-reported data to identify which specific AI and robotics technologies each hospital has adopted, reflecting real-world organizational commitment and investment.
County Health Rankings & Roadmaps (CHR): Provided by the University of Wisconsin Population Health Institute, this dataset offers standardized, county-level metrics for health outcomes (like premature death rates), health behaviors (like smoking and obesity rates), and key socioeconomic factors.
OUR ANALYTICAL APPROACH
Our study employed a multi-faceted approach to analyze this complex data:
1. Geospatial Analysis: We first geocoded all 6,166 hospital locations. We then conducted a network analysis to create 30-minute driving-time zones around each hospital with AI or robotics. By overlaying this with high-resolution population data, we could precisely calculate what percentage of the U.S. population has access to these technologies, quantifying the "digital divide."
2. Moderation Analysis: To test the "Great Equalizer" hypothesis, we used statistical regression models. This allowed us to examine whether the presence of hospital AI (the moderator) changed the strength of the relationship between unhealthy community behaviors (the independent variable) and poor health outcomes (the dependent variable).
3. Quasi-Experimental Methods: To estimate the direct impact of specific technologies while accounting for the fact that hospitals don't adopt AI randomly, we used advanced methods like Inverse Probability of Treatment Weighting (IPTW). This technique helps create more comparable groups of "adopting" and "non-adopting" counties, allowing for a more robust estimate of the technology's association with health outcomes.
A NOTE ON CAUSALITY
As a cross-sectional study, our work identifies powerful associations, not definitive cause-and-effect relationships. The robust patterns we found, however, are lent plausibility by a growing body of clinical research and provide a critical baseline for future longitudinal studies and immediate policy consideration.