Clever Firm Predicts Patients Most at Risk, Then Tries to Intervene Before They Get Sicker
The diabetic patient hit the danger zone.
Ideally, blood sugar, measured by an A1C test, rests at 5.9 or less. A 7 is elevated, according to the Diabetes Council. Over 10, and you're into the extreme danger zone, at risk of every diabetic crisis from kidney failure to blindness.
In three months of working with a case manager, Jen's blood sugar had dropped to 7.2, a much safer range.
This patient's A1C was 10. Let's call her Jen for the sake of this story. (Although the facts of her case are real, the patient's actual name wasn't released due to privacy laws.).
Jen happens to live in Pennsylvania's Lehigh Valley, home of the nonprofit Lehigh Valley Health Network, which has eight hospital campuses and various clinics and other services. This network has invested more than $1 billion in IT infrastructure and founded Populytics, a spin-off firm that tracks and analyzes patient data, and makes care suggestions based on that data.
When Jen left the doctor's office, the Populytics data machine started churning, analyzing her data compared to a wealth of information about future likely hospital visits if she did not comply with recommendations, as well as the potential positive impacts of outreach and early intervention.
About a month after Jen received the dangerous blood test results, a community outreach specialist with psychological training called her. She was on a list generated by Populytics of follow-up patients to contact.
"It's a very gentle conversation," says Cathryn Kelly, who manages a care coordination team at Populytics. "The case manager provides them understanding and support and coaching." The goal, in this case, was small behavioral changes that would actually stick, like dietary ones.
In three months of working with a case manager, Jen's blood sugar had dropped to 7.2, a much safer range. The odds of her cycling back to the hospital ER or veering into kidney failure, or worse, had dropped significantly.
While the health network is extremely localized to one area of one state, using data to inform precise medical decision-making appears to be the wave of the future, says Ann Mongovern, the associate director of Health Care Ethics at the Markkula Center for Applied Ethics at Santa Clara University in California.
"Many hospitals and hospital systems don't yet try to do this at all, which is striking given where we're at in terms of our general technical ability in this society," Mongovern says.
How It Happened
While many hospitals make money by filling beds, the Lehigh Valley Health Network, as a nonprofit, accepts many patients on Medicaid and other government insurances that don't cover some of the costs of a hospitalization. The area's population is both poorer and older than national averages, according to the U.S. Census data, meaning more people with higher medical needs that may not have the support to care for themselves. They end up in the ER, or worse, again and again.
In the early 2000s, LVHN CEO Dr. Brian Nester started wondering if his health network could develop a way to predict who is most likely to land themselves a pricey ICU stay -- and offer support before those people end up needing serious care.
Embracing data use in such specific ways also brings up issues of data security and patient safety.
"There was an early understanding, even if you go back to the (federal) balanced budget act of 1997, that we were just kicking the can down the road to having a functional financial model to deliver healthcare to everyone with a reasonable price," Nester says. "We've got a lot of people living longer without more of an investment in the healthcare trust."
Popultyics, founded in 2013, was the result of years of planning and agonizing over those population numbers and cost concerns.
"We looked at our own health plan," Nester says. Out of all the employees and dependants on the LVHN's own insurance network, "roughly 1.5 percent of our 25,000 people — under 400 people — drove $30 million of our $130 million on insurance costs -- about 25 percent."
"You don't have to boil the ocean to take cost out of the system," he says. "You just have to focus on that 1.5%."
Take Jen, the diabetic patient. High blood sugar can lead to kidney failure, which can mean weekly expensive dialysis for 20 years. Investing in the data and staff to reach patients, he says, is "pennies compared to $100 bills."
For most doctors, "there's no awareness for providers to know who they should be seeing vs. who they are seeing. There's no incentive, because the incentive is to see as many patients as you can," he says.
To change that, first the LVHN invested in the popular medical management system, Epic. Then, they negotiated with the top 18 insurance companies that cover patients in the region to allow access to their patient care data, which means they have reams of patient history to feed the analytics machine in order to make predictions about outcomes. Nester admits not every hospital could do that -- with 52 percent of the market share, LVHN had a very strong negotiating position.
Third party services take that data and churn out analytics that feeds models and care management plans. All identifying information is stripped from the data.
"We can do predictive modeling in patients," says Populytics President and CEO Gregory Kile. "We can identify care gaps. Those care gaps are noted as alerts when the patient presents at the office."
Kile uses himself as a hypothetical patient.
"I pull up Gregory Kile, and boom, I see a flag or an alert. I see he hasn't been in for his last blood test. There is a care gap there we need to complete."
"There's just so much more you can do with that information," he says, envisioning a future where follow-up for, say, knee replacement surgery and outcomes could be tracked, and either validated or changed.
Ethical Issues at the Forefront
Of course, embracing data use in such specific ways also brings up issues of security and patient safety. For example, says medical ethicist Mongovern, there are many touchpoints where breaches could occur. The public has a growing awareness of how data used to personalize their experiences, such as social media analytics, can also be monetized and sold in ways that benefit a company, but not the user. That's not to say data supporting medical decisions is a bad thing, she says, just one with potential for public distrust if not handled thoughtfully.
"You're going to need to do this to stay competitive," she says. "But there's obviously big challenges, not the least of which is patient trust."
So far, a majority of the patients targeted – 62 percent -- appear to embrace the effort.
Among the ways the LVHN uses the data is monthly reports they call registries, which include patients who have just come in contact with the health network, either through the hospital or a doctor that works with them. The community outreach team members at Populytics take the names from the list, pull their records, and start calling. So far, a majority of the patients targeted – 62 percent -- appear to embrace the effort.
Says Nester: "Most of these are vulnerable people who are thrilled to have someone care about them. So they engage, and when a person engages in their care, they take their insulin shots. It's not rocket science. The rocket science is in identifying who the people are — the delivery of care is easy."
[Editor's Note: This is the fifth episode in our Moonshot series, which explores cutting-edge scientific developments that stand to fundamentally transform our world.]
Kira Peikoff was the editor-in-chief of Leaps.org from 2017 to 2021. As a journalist, her work has appeared in The New York Times, Newsweek, Nautilus, Popular Mechanics, The New York Academy of Sciences, and other outlets. She is also the author of four suspense novels that explore controversial issues arising from scientific innovation: Living Proof, No Time to Die, Die Again Tomorrow, and Mother Knows Best. Peikoff holds a B.A. in Journalism from New York University and an M.S. in Bioethics from Columbia University. She lives in New Jersey with her husband and two young sons. Follow her on Twitter @KiraPeikoff.
With the pandemic at the forefront of everyone's minds, many people have wondered if food could be a source of coronavirus transmission. Luckily, that "seems unlikely," according to the CDC, but foodborne illnesses do still sicken a whopping 48 million people per year.
Whole genome sequencing is like "going from an eight-bit image—maybe like what you would see in Minecraft—to a high definition image."
In normal times, when there isn't a historic global health crisis infecting millions and affecting the lives of billions, foodborne outbreaks are real and frightening, potentially deadly, and can cause widespread fear of particular foods. Think of Romaine lettuce spreading E. coli last year— an outbreak that infected more than 500 people and killed eight—or peanut butter spreading salmonella in 2008, which infected 167 people.
The technologies available to detect and prevent the next foodborne disease outbreak have improved greatly over the past 30-plus years, particularly during the past decade, and better, more nimble technologies are being developed, according to experts in government, academia, and private industry. The key to advancing detection of harmful foodborne pathogens, they say, is increasing speed and portability of detection, and the precision of that detection.
Getting to Rapid Results
Researchers at Purdue University have recently developed a lateral flow assay that, with the help of a laser, can detect toxins and pathogenic E. coli. Lateral flow assays are cheap and easy to use; a good example is a home pregnancy test. You place a liquid or liquefied sample on a piece of paper designed to detect a single substance and soon after you get the results in the form of a colored line: yes or no.
"They're a great portable tool for us for food contaminant detection," says Carmen Gondhalekar, a fifth-year biomedical engineering graduate student at Purdue. "But one of the areas where paper-based lateral flow assays could use improvement is in multiplexing capability and their sensitivity."
J. Paul Robinson, a professor in Purdue's Colleges of Veterinary Medicine and Engineering, and Gondhalekar's advisor, agrees. "One of the fundamental problems that we have in detection is that it is hard to identify pathogens in complex samples," he says.
When it comes to foodborne disease outbreaks, you don't always know what substance you're looking for, so an assay made to detect only a single substance isn't always effective. The goal of the project at Purdue is to make assays that can detect multiple substances at once.
These assays would be more complex than a pregnancy test. As detailed in Gondhalekar's recent paper, a laser pulse helps create a spectral signal from the sample on the assay paper, and the spectral signal is then used to determine if any unique wavelengths associated with one of several toxins or pathogens are present in the sample. Though the handheld technology has yet to be built, the idea is that the results would be given on the spot. So someone in the field trying to track the source of a Salmonella infection could, for instance, put a suspected lettuce sample on the assay and see if it has the pathogen on it.
"What our technology is designed to do is to give you a rapid assessment of the sample," says Robinson. "The goal here is speed."
Seeing the Pathogen in "High-Def"
"One in six Americans will get a foodborne illness every year," according to Dr. Heather Carleton, a microbiologist at the Centers for Disease Control and Prevention's Enteric Diseases Laboratory Branch. But not every foodborne outbreak makes the news. In 2017 alone, the CDC monitored between 18 and 37 foodborne poison clusters per week and investigated 200 multi-state clusters. Hardboiled eggs, ground beef, chopped salad kits, raw oysters, frozen tuna, and pre-cut melon are just a taste of the foods that were investigated last year for different strains of listeria, salmonella, and E. coli.
At the heart of the CDC investigations is PulseNet, a national network of laboratories that uses DNA fingerprinting to detect outbreaks at local and regional levels. This is how it works: When a patient gets sick—with symptoms like vomiting and fever, for instance—they will go to a hospital or clinic for treatment. Since we're talking about foodborne illnesses, a clinician will likely take a stool sample from the patient and send it off to a laboratory to see if there is a foodborne pathogen, like salmonella, E. Coli, or another one. If it does contain a potentially harmful pathogen, then a bacterial isolate of that identified sample is sent to a regional public health lab so that whole genome sequencing can be performed.
Whole genome sequencing can differentiate "virtually all" strains of foodborne pathogens, no matter the species, according to the FDA.
Whole genome sequencing is a method for reading the entire genome of a bacterial isolate (or from any organism, for that matter). Instead of working with a couple dozen data points, now you're working with millions of base pairs. Carleton likes to describe it as "going from an eight-bit image—maybe like what you would see in Minecraft—to a high definition image," she says. "It's really an evolution of how we detect foodborne illnesses and identify outbreaks."
If the bacterial isolate matches another in the CDC's database, this means there could be a potential outbreak and an investigation may be started, with the goal of tracking the pathogen to its source.
Whole genome sequencing has been a relatively recent shift in foodborne disease detection. For more than 20 years, the standard technique for analyzing pathogens in foodborne disease outbreaks was pulsed-field gel electrophoresis. This method creates a DNA fingerprint for each sample in the form of a pattern of about 15-30 "bands," with each band representing a piece of DNA. Researchers like Carleton can use this fingerprint to see if two samples are from the same bacteria. The problem is that 15-30 bands are not enough to differentiate all isolates. Some isolates whose bands look very similar may actually come from different sources and some whose bands look different may be from the same source. But if you can see the entire DNA fingerprint, then you don't have that issue. That's where whole genome sequencing comes in.
Although the PulseNet team had piloted whole genome sequencing as early as 2013, it wasn't until July of last year that the transition to using whole genome sequencing for all pathogens was complete. Though whole genome sequencing requires far more computing power to generate, analyze, and compare those millions of data points, the payoff is huge.
Stopping Outbreaks Sooner
The U.S. Food and Drug Administration (FDA) acquired their first whole genome sequencers in 2008, according to Dr. Eric Brown, the Director of the Division of Microbiology in the FDA's Office of Regulatory Science. Since then, through their GenomeTrakr program, a network of more than 60 domestic and international labs, the FDA has sequenced and publicly shared more than 400,000 isolates. "The impact of what whole genome sequencing could do to resolve a foodborne outbreak event was no less impactful than when NASA turned on the Hubble Telescope for the first time," says Brown.
Whole genome sequencing has helped identify strains of Salmonella that prior methods were unable to differentiate. In fact, whole genome sequencing can differentiate "virtually all" strains of foodborne pathogens, no matter the species, according to the FDA. This means it takes fewer clinical cases—fewer sick people—to detect and end an outbreak.
And perhaps the largest benefit of whole genome sequencing is that these detailed sequences—the millions of base pairs—can imply geographic location. The genomic information of bacterial strains can be different depending on the area of the country, helping these public health agencies eventually track the source of outbreaks—a restaurant, a farm, a food-processing center.
Coming Soon: "Lab in a Backpack"
Now that whole genome sequencing has become the go-to technology of choice for analyzing foodborne pathogens, the next step is making the process nimbler and more portable. Putting "the lab in a backpack," as Brown says.
The CDC's Carleton agrees. "Right now, the sequencer we use is a fairly big box that weighs about 60 pounds," she says. "We can't take it into the field."
A company called Oxford Nanopore Technologies is developing handheld sequencers. Their devices are meant to "enable the sequencing of anything by anyone anywhere," according to Dan Turner, the VP of Applications at Oxford Nanopore.
"The sooner that we can see linkages…the sooner the FDA gets in action to mitigate the problem and put in some kind of preventative control."
"Right now, sequencing is very much something that is done by people in white coats in laboratories that are set up for that purpose," says Turner. Oxford Nanopore would like to create a new, democratized paradigm.
The FDA is currently testing these types of portable sequencers. "We're very excited about it. We've done some pilots, to be able to do that sequencing in the field. To actually do it at a pond, at a river, at a canal. To do it on site right there," says Brown. "This, of course, is huge because it means we can have real-time sequencing capability to stay in step with an actual laboratory investigation in the field."
"The timeliness of this information is critical," says Marc Allard, a senior biomedical research officer and Brown's colleague at the FDA. "The sooner that we can see linkages…the sooner the FDA gets in action to mitigate the problem and put in some kind of preventative control."
At the moment, the world is rightly focused on COVID-19. But as the danger of one virus subsides, it's only a matter of time before another pathogen strikes. Hopefully, with new and advancing technology like whole genome sequencing, we can stop the next deadly outbreak before it really gets going.