Google Research Uses AI to Help Public Health Experts Track Disease and Health Risks
Google Research developed an artificial intelligence model that combines search trends, human mobility, and environmental data into digital location profiles. Tests show these profiles help public health experts track and predict health trends faster and across borders.
In short: Google Research developed an artificial intelligence model that combines search trends, human mobility, and environmental data into digital location profiles. Tests show these profiles help public health experts track and predict health trends faster and across borders.
Finding out where a disease outbreak will strike next often feels like trying to read a map written in invisible ink, but new artificial intelligence tools are helping scientists make those hidden patterns clear.
What happened, in plain words
Google Research software engineers Arbaaz Muslim and Gautam Prasad introduced the Population Dynamics Foundation Model (PDFM), part of Google Earth AI. This tool creates digital "fingerprints" for different locations by summarizing anonymous search trends, human mobility, environmental factors, and building density. In partnership with various research groups, Google tested how plugging these location fingerprints into existing health models could improve public health work. Five different case studies showed that the model helped researchers track cross-border vaccination rates, estimate current heart disease deaths without waiting for old census data, forecast dengue outbreaks in Mexico, predict postpartum depression risks among mothers, and anticipate cholera hotspots in the Democratic Republic of the Congo.
Key points
- Tracking cross-border health needs: Researchers at Mount Sinai Health System and Boston Children's Hospital used Canadian and U.S. location data to better predict measles vaccination coverage near the border, increasing the explained variation from 16% to 22%.
- Faster heart disease estimates: Partners at NYU Grossman School of Medicine found that the AI model could estimate current-year cardiovascular disease deaths across U.S. counties without the usual multi-year reporting lags of traditional census data.
- Early warnings for dengue fever: Working with the University of Oxford and Tecnológico de Monterrey, researchers paired the AI tool with a time-series model to forecast dengue cases in Mexican municipalities one month ahead, improving accuracy in up to 72% of active transmission areas.
- Supporting maternal mental health: University of Washington researchers found that the AI tool helped identify postpartum depression risks in mothers and could help health systems reach thousands more rural mothers or reduce false alarms.
- Anticipating cholera outbreaks: A lightweight version of the model tested with World Health Organization regional partners helped forecast cholera hotspots four to eight weeks in advance in the Democratic Republic of the Congo.
Terms explained
- Foundation Model — A large-scale artificial intelligence system trained on massive amounts of diverse data that can be adapted to perform many different tasks. Example: Think of it like a master Swiss Army knife that already knows how to do many jobs, so you do not have to build a new tool from scratch for every single task.
- Embeddings — Compact digital summaries or "fingerprints" that represent complex real-world information in a format computers can easily process. Example: It is like turning a detailed description of a neighborhood's roads, stores, and weather into a simple numerical code.
- Epidemiological Surveillance — The ongoing collection and analysis of health data to track and monitor the spread of diseases in populations. Example: It is similar to a weather radar system, but instead of tracking rain storms, health workers watch for signs of coming sicknesses.
Why it matters
Traditional public health tracking often suffers from multi-year delays and missing information, making it hard to send medical supplies or vaccines where they are needed most. By filling in these data gaps, health departments can use current conditions to guide prevention resources, spot outbreaks weeks in advance, and direct care to vulnerable communities more efficiently.
What we still don't know
The research relies on current data snapshots, and scientists are still working on making the embeddings temporally dynamic and better suited for under-connected regions. Furthermore, these tests are part of academic and partner evaluations rather than an established global standard for medical practice.
Based on reporting from Google Research. This is an independent explainer, written in our own words with AI assistance; Google Research has not reviewed or endorsed it. Read the original for the full details.