ÿ�ճԹ�

Whether it be ventilators or vaccines, allocating limited supplies during the COVID-19 pandemic has been a persistent problem, especially in underserved, minority communities that have been affected disproportionately by the coronavirus. Now, a study by Stanford scholars has developed a promising new way to reach vulnerable populations and deliver resources more equitably.

Working in collaboration with public health officials in Santa Clara County and community leaders in East San Jose, Stanford scholars borrowed a simple concept from machine learning to prototype a new way to distribute COVID-19 diagnostic tests that were critical to better understanding the transmission of the coronavirus.

Their findings, published Aug. 27 in , showed that machine learning, when combined with local insights from community health workers working on the ground, broadened testing capacity, decreased demographic disparities in testing and caught clusters of infections early on.

“This intervention demonstrates a critical need for academics to work in partnership with community health workers and public health agencies to reach and disseminate information to vulnerable communities,” said Daniel E. Ho, senior author of the study and the William Benjamin Scott and Luna M. Scott Professor of Law at Stanford Law School.

Ho, a senior fellow at the Stanford Institute for Economic Policy Research (ÿ�ճԹ�), is the founder and faculty director of  where he and other Stanford researchers work with government agencies on a pro bono basis to adapt and leverage advances in data science and machine learning for public policy.

When COVID-19 hit, Ho and his team — including study co-author , who has helped lead the lab’s response to the pandemic, as well as with the  — began working with public health officials to investigate how they could assist with their pandemic response.

One of those officials was Dr. Analilia Garcia, the racial and health equity senior manager for the Santa Clara County public health department.

Garcia and her colleagues were concerned by how COVID-19 was disproportionately impacting the county’s Latinx residents. While this demographic makes up just over one quarter (25.8 percent) of the county’s population, they have accounted for .

As COVID-19 spread throughout the area, ensuring that this population had access to resources, including testing, became a priority.

“We knew we needed to go to where the people were at,” said Garcia. She and her team set up testing centers in local churches, shopping centers and other areas. But they soon saw it was not accessible enough.

For example, while 60 percent of East San Jose residents are Latinx, only 30 to 50 percent of visitors at the area’s two nearest testing centers, the Emmanuel Baptist Church and the Santa Clara County Fairgrounds, were Latinx.

Garcia learned that people faced challenges that prevented them from getting tested: some were unable to secure childcare or arrange transport, while others worried about finding the time off work to go and get tested.

“A lot of our most vulnerable communities are our essential workers,” Garcia said. “So we needed to make sure that our access consisted not only of the accessible locations but accessible times and also removing barriers.”

Garcia recalled seeing programs in San Francisco that provided at-home testing to some of their residents. She wondered if something similar could work in Santa Clara County.

But that raised another question: Where exactly should testers go? That’s where Ho and the RegLab team stepped in to help.

One possibility was to send testers to areas where the positivity rate is high. The positivity rate is the proportion of all tests that have a positive diagnosis and is typically seen as a measure of testing resources. But raw positivity rates can be misleading because some areas might have low rates as a statistical artifact from too few people getting tested. Thus, the indicator might suggest testing isn’t needed in the very places it’s needed most.

To avoid this, the researchers used a simple insight from machine learning called “uncertainty sampling”: Go where there is the most uncertainty about COVID-19 transmission.

“The areas where you are uncertain are probably also exactly the areas where you are most worried about the testing infrastructure not reaching everyone,” Ho said.

To allocate by uncertainty, the team used a simple technique called upper confidence bound sampling to determine where the data is most unclear about the highest positivity rates. This approach enabled the team to navigate “the explore-exploit tradeoff”: How to allocate resources that can be used to explore risk (in their case, unknown outbreak areas) and exploit known risk (known outbreak areas) that could guide community health workers to where they were needed most.

Through this model, Ho’s team created detailed maps and assigned community health workers to specific neighborhoods in real-time based on the daily intake of cases.

“Stanford’s innovations, infrastructure and technology and resources really facilitated our ability to pinpoint where we needed to go to implement the intervention,” said Garcia. “This idea of door-to-door testing and offering a free Covid test was not random. Stanford played a very important role in helping us use the data to inform which doors we knocked on.”

Taking a data-driven, community-centered approach

To deliver door-to-door tests, the researchers collaborated with the local co-operative META, an acronym for Mujeres Empresarias Tomando Acción (“Entrepreneurial Women Taking Action”) to recruit promotores de salud (“community health workers”), Spanish-speaking community healthcare workers who are able to serve as trusted advocates between Latinx residents and the healthcare system.

Involving people that the community could trust was critical. As Garcia and the researchers had learned, some immigrants avoided testing altogether because they were distrustful of the healthcare system.

“These are moms, wives and leaders who were suffering the burden of Covid themselves,” said Garcia. “META rose to the occasion when our County called them and said ‘we are here to serve, we are going to stand with you.

Promotores were trained by the Santa Clara County Public Health Department to go door-to-door to provide at-home COVID-19 testing kits to households in East San Jose.

To complement the data-driven approach, the team also gave promotores flexibility to determine where to conduct testing based on their own experiences as community members themselves, like their knowledge of a neighbor’s social gathering that had violated Covid safety protocols.

“The promotores were pivotal to answering questions, engaging with residents and gaining trust at a difficult time,” said Ho.

Expanded outreach

In total, promotores expanded testing resources in East San Jose neighborhoods between 60-90 percent relative to the baseline over two months.

The researchers found that the method of going door-to-door boosted testing among Latinx populations significantly.

For example, while 49.0 percent and 30.7 percent of individuals at the Church and Fairgrounds were Latinx, respectively, 87.6 percent of those tested via the door-to-door method were Latinx, representing an 80 percent to 184 percent relative increase in Latinx individuals reached, the researchers said in the paper.

Going door-to-door also reached 13 percent of individuals aged 65 and over, compared to 8 percent and 6 percent at the Church and Fairgrounds.

The overall positivity rate for door-to-door tests was 6.8 percent and uncertainty sampling uncovered the most positive cases, with a positivity rate of 10.8 percent.

Power of collaboration

For both Ho and Garcia, the results demonstrate what can happen when academia, local government and the community come together to tackle a problem.

“It was a really beautiful trifecta because you had three different entities with different expertise and experiences coming together in response to a crisis. It was an opportunity to leverage all of that collective knowledge to respond to a life and death crisis,” said Garcia.

Ho said he hopes that approaches combining science, policy and the expertise of trusted community members can be helpful in addressing other inequities, including vaccination efforts, which the team is now turning its attention to.

“The pandemic has laid bare profound social and health disparities,” he added. “What is exciting is that this intervention points the way toward using science and community partnerships to close these kinds of demographic gaps and ensure a more equitable future.”

Ho is also affiliated with the Political Science Department in the School of Humanities and Sciences and the Institute for Human-Centered Artificial Intelligence (HAI).

Other authors on the paper include first authors Ben Chugg, Lisa Lu and Derek Ouyang, all affiliated with ÿ�ճԹ�. Benjamin Anderson and Raymond Ha, also from ÿ�ճԹ�’s RegLab, contributed. Alexis D’Agostino, Anandi Sujeer, Sarah L. Rudman and Analilia Garcia from the county of Santa Clara Public Health Department are co-authors on the paper as well.

A version of this story was first published by Stanford News Service.