How Google Street View data can help improve public health

Big data and artificial intelligence are changing the way we think about health: from detecting diseases and recognizing patterns to predicting outcomes and speeding up response times.

In a new study analyzing two million Google Street View images of New York City streets, a team of researchers from New York University evaluated the usefulness of this digital data in informing public health decision-making. Their findings, published in the Proceedings of the National Academy of Sciencesshow how relying on Street View imagery alone can lead to inaccuracies and misplaced interventions, but how combining it with other knowledge expands its possibilities.

“There is a lot of excitement about leveraging new data sources to get a holistic picture of health, including using machine learning and data science methods to gain new insights,” said Rumi Chunara, associate professor of biostatistics at the NYU School of Global Public Health, associate professor of computer science at the NYU Tandon School of Engineering, and lead author of the study.

“Our research highlights the potential of digital data sources like Street View imagery to improve public health research, while highlighting the limitations of data and the complex dynamics between the environment, individual behavior, and health outcomes,” said Miao Zhang, a PhD candidate at the NYU Tandon School of Engineering and lead author of the study.

A street-level view of health

In recent years, researchers have begun using Street View imagery to link an area’s environment and infrastructure to outcomes like mental health, infectious disease or obesity — a task that’s difficult to measure by hand.

“We know that the built environment of a city can affect our health, whether it’s the availability of sidewalks and green spaces for walking, or grocery stores that sell healthy foods,” Chunara said. “Some research shows that the availability of sidewalks correlates with lower obesity rates, but is that the whole story?”

“Our motivation for this study was to dig deeper into these associations to see if there are possible factors underlying this,” Zhang said.

Chunara, Zhang and their colleagues analyzed more than two million Google Street View images of every street in New York City, using artificial intelligence to assess the availability of sidewalks and crosswalks in the images. They then compared this information with localized data on obesity, diabetes and physical activity from the Centers for Disease Control and Prevention to see whether the built environment predicted health outcomes.

The researchers found that neighborhoods with more crosswalks had lower rates of obesity and diabetes. However, no significant link was found between sidewalks and health outcomes, contrary to previous research.

“That may be because many sidewalks in New York City are in places that people don’t use, such as along a highway, on a bridge, or in a tunnel. So sidewalk density doesn’t reflect the walkability of a neighborhood as well as crosswalks do,” Zhang said.

They also found issues with the accuracy of the AI-generated labels for the Street View images, warning that they may not match the “ground truth” and are not a reliable metric on their own. When comparing existing data on sidewalk availability in New York City with the labeled Street View images, they found that many were incorrectly labeled as having or not having sidewalks, possibly due to cars or shadows obscuring them in photos.

If you build it, will they come?

While crosswalks were associated with lower rates of obesity and diabetes, the researchers applied a public health lens to determine what might explain this association. Their analyses of the CDC data revealed that physical activity — not just crosswalks, as measured in Street View images — was responsible for the decreases in obesity and diabetes.

One study found that increasing physical activity can lead to a four-fold greater reduction in obesity and a seventeen-fold greater reduction in diabetes than adding more zebra crossings.

“We found that physical activity drives the benefits of zebra crossings, so it is important to consider such mechanisms, especially when they operate at different levels, such as the built environment versus individuals,” Zhang said.

Based on their findings, the researchers conclude that public health decision-making should not only rely on new data sources, but also consider domain knowledge. When analyzing street view images, it is critical to integrate knowledge from computer science, for example how image processing techniques can improve accuracy or how bias in algorithms can be corrected, and knowledge from public health, which drives associations between the built environment and health outcomes. Overlaying this expertise on big data can inform how programs are designed and implemented to improve public health.

In this case, adding more sidewalks and crosswalks would be less effective in improving health than the same increase in physical activity, for example through local community exercise classes.

“While growing amounts of digital data can be useful in informing decision-making, our results show that simply using associations from new data sources may not lead to the most useful interventions or the best allocation of resources,” Chunara added. “A more nuanced approach using big data combined with expertise is needed to make the most of these new data.”

Salman Rahman and Vishwali Mhasawade of NYU Tandon were also authors of the study.