Google Research reported five partner-led studies on October 6, 2026, testing whether its Population Dynamics Foundation Model could help fill public-health data gaps in the United States, Canada, Mexico and the Democratic Republic of the Congo.

The early results included better estimates of childhood vaccination coverage near the U.S.-Canada border and improved one-month dengue forecasts in some Mexican municipalities. But performance varied by task, place and forecast period.

One model, five public-health tasks

Part of Google Earth AI, the model turns aggregated search trends, mobility patterns, features of the built environment and environmental data into compact representations of places. Researchers added those representations to existing statistical or machine-learning models. They did not use them as stand-alone diagnoses or alerts.

The studies examined five tasks:

  • Estimating measles, mumps and rubella vaccination coverage
  • Estimating cardiovascular mortality
  • Forecasting dengue
  • Predicting postpartum-depression risk
  • Forecasting cholera

Cross-border data improved one vaccination estimate

In 146 U.S. counties near the Canadian border, adding Canadian location data increased the model’s explained variation in measles, mumps and rubella vaccination coverage from 0.159 to 0.216. That is a relative gain of 36%.

The finding suggests that data from across a national border can improve estimates in nearby communities that domestic data alone may miss. It does not show that the same approach will improve every health estimate.

For dengue in Mexico, researchers reported a statistically significant improvement in one-month forecasts, concentrated in municipalities with active transmission. They found no consistent improvement across all forecast periods and locations. A stronger signal during active transmission is not proof the model can reliably predict every local surge.

Promising results, not a public-health service

The study also reported gains in some cholera forecasting measures in the Democratic Republic of the Congo. It tested the model on cardiovascular mortality estimates and postpartum-depression risk prediction in the United States.

The paper says geographic data added a signal for postpartum-depression risk. It did not replace individual socioeconomic information or close screening gaps.

The studies test a research approach, not an operational outbreak-warning system. Their results show where location data may help, while leaving clear limits on how broadly the gains apply.

The work is a preprint, not peer-reviewed evidence of a deployed public-health service. Google says data from its Population Dynamics Foundation Model are available commercially in Preview, with no-cost access for select research uses.

The studies do not show that health authorities are using the model as an operational warning system. For now, the practical takeaway is narrower: adding geographic data may improve particular estimates and forecasts, but the benefit depends on the task and location.