Baylor wins 2024 STAT Madness with ‘smoke alarm’ for viral disease outbreaks

It began with injecting monkeys with sewage to see if they developed polio.

Building off that early polio work from the 1940s, researchers from the Baylor College of Medicine analyzed public wastewater and showed they could detect over 450 disease-causing viruses. Their study demonstrating the power of sewage as an early-warning system for outbreaks won the 2024 STAT Madness popular vote.

Wastewater monitoring became popular during the Covid-19 pandemic to predict outbreaks and assess infection trends, but the Baylor team’s goal was to go beyond any single virus and simultaneously test for over 3,000 different pathogens, including all known human viruses.

“Think of it as a smoke alarm,” said senior author Anthony Maresso, the Joseph Melnick Endowed Chair of Virology and Microbiology at Baylor. It gives you precious minutes to get out the fire extinguisher or call 911. “If we don’t have vigilance, then it could become a blaze.”

The 64 teams in STAT’s month-long, bracket-style celebration of biomedical research garnered 267,644 votes for studies on topics ranging from an electric pill that reduces hunger, inspired by the “thorny devil” lizard, to gas-trapping materials that bolster cancer treatment, manufactured similarly to Pop Rocks.

In the end, the Baylor team studying wastewater epidemiology won out with 64.8% of the vote in the final round. They beat a team from Stanford University School of Medicine that conducted a randomized controlled trial of using conversational AI to give patients personalized advice  to improve diabetes management.

Researchers from Weill Cornell Medicine won the STAT Madness All-Star award last month, voted on by attendees at the Breakthrough Summit East, for their work to create an on-demand birth control pill for men.

The Baylor team leveraged the fact that many viruses, including SARS-CoV-2, infect urinary or gastrointestinal cells, thereby shedding into urine and feces — and thus wastewater. So, the team worked with water treatment plants in Houston and El Paso, every week packing up (leak-proof) bottles of sewage and shipping them overnight to Baylor for testing.

After removing the “solids” and a couple of rounds of processing, the team deployed over a million probes, or single-stranded nucleic acids corresponding to parts of the viral genome, and sequenced the viruses that stuck onto them, according to Michael Tisza, the first author of the paper and an assistant professor of molecular virology and microbiology at Baylor. Notably, the probes had some wiggle room — roughly 15% — so they could detect mutated viruses and new strains.

In total, Tisza and Maresso identified 465 distinct viruses with this technique, most of which have never been detected from wastewater. “Hundreds of viruses are transmitting without your knowledge,” said Maresso. “If we can continue to sample these viruses over many years and we get access to very rigorous clinical datasets, we’re going to be able to start making causal associations that are going to be transformative in terms of the etiology of whole new diseases.”

The study was published in the journal Nature Communications in October 2023, funded by the Texas Epidemic Public Health Institute. The authors have since expanded wastewater monitoring into 10 cities in the state, covering up to 5 million people.

This approach has limitations, the authors acknowledged. These data are never going to tell you that this patient should be quarantined, or that a certain family is all infected, since wastewater is collected at the community level. “What we’re doing isn’t ever going to completely replace clinical testing or at-home rapid antigen testing,” said Tisza. “But what our data can do is inform public health providers, or even individuals, what they might want to be testing for.”

And some data will be missed with wastewater. In the study, Tisza and Maresso found their sequencing reads were highly correlated with the number of SARS-CoV-2, influenza, and mpox patients in the same communities, but it wasn’t perfect. For instance, they didn’t find any mpox in the wastewater in El Paso, where there were 10 reported cases over six months, but did in Houston, where there were 1,050 cases.

“We probably are not going to detect a virus from one single case,” said Tisza, “but we would, for most things, catch a small outbreak.”

The hope is to identify these outbreaks early and cheaply by continuously monitoring wastewater. “It’s kind of ironic that just about everybody in the world is doing everything they can to throw this material away,” said Maresso. “We’re one of the few lunatics that are actually trying to get our hands on more and more of it.”

The Stanford team took second place for its clinical trial testing a conversational AI that would chat with patients to adjust their insulin doses and help control their type 2 diabetes. “For many patients who need to be on insulin, it’s a critical, lifesaving medication, but the process of getting patients on the right cocktail is very complex,” said Sharif Vakili, who is the study’s co-first author and an assistant professor of medicine at Stanford.

The insulin doses need to be “titrated,” or finely adjusted, to achieve optimal blood sugar control, but that’s normally difficult to do when patients might see their doctors once every few months, he explained.

So, in this 32-person trial, half of the diabetes patients got an Amazon Alexa smart speaker that would check in with them every day, asking for their fasting blood sugar level and how much insulin they took, before recommending a new dose based on the provider-prescribed treatment plan. The other half got an online blood glucose and insulin log, as well as an Alexa that would remind them every day to complete that log. Vakili and his team used this “diabetes smart speaker” because they thought it would be easier and more convenient for older patients to use than punching data into a smartphone app.

Patients in the conversational AI group were more likely to be taking their insulin than the control group (83% versus 50%), took less time to reach optimal insulin dosing (15 days versus 56 days), and were more likely to have maintained blood sugar control by the end of the eight-week trial (81% versus 25%).

“It’s really not a big secret why,” said Vakili, given that the conversational AI group had their insulin dose adjusted an average 7.3 times, compared with 1.6 times for the other group. “You just delivered way more care in that period of time than the health system is able to offer.”

The paper was published in JAMA Network Open in December 2023 and has several limitations, such as the small sample size and most of the data being self-reported. However, if validated in larger trials, this strategy of “remote patient intervention” could improve patients’ engagement with their care plans and ultimately improve outcomes, while also addressing the primary care shortage by extending clinicians’ capacity to care for more patients.

Vakili distinguishes this approach from remote patient monitoring, like smart watches measuring EKG data or continuous glucose monitors. While those devices collect data, they don’t give patients immediate, clinically meaningful recommendations. “There’s no closed loop; you have to still bring it to the doctor for interpretation,” he said.

The idea with remote patient intervention is that an AI will act on those troves of data to deliver care, and Vakili and co-first author Ashwin Nayak, who is also an assistant professor of medicine at Stanford, have started a company called UpDoc to develop and commercialize their approach.

“There’s a lot of digital health tools and technologies out there, but this one actually changes how you practice,” said Vakili. “If you take a moment and think about it, you have now moved from the doctor prescribing a medication to a doctor prescribing a protocol.”