By JULIE EGGINGTON, EKATERINA CLEARY & LEEZA OSIPENKO

When CMS issued its Request for Information under the Comprehensive Regulations to Uncover Suspicious Healthcare (CRUSH) initiative in February, it zeroed in on a long-festering problem: fraud, waste, and abuse in laboratory testing, especially in genetic and molecular diagnostics.

The laboratory industry will respond. And when it does, its arguments will sound polished, familiar, and deeply reassuring. They will also be either disingenuous or unproven.

If policymakers want this effort to succeed, they should be prepared to handle three claims that have long shielded problematic practices from meaningful oversight.

Claim 1: Fraud, waste and abuse is limited to a few bad actors

Expect labs to argue that fraud, waste, and abuse is rare, isolated, and already addressed through enforcement actions. The narrative will feature a handful of egregious cases, presented as outliers in an otherwise trustworthy ecosystem.

But the problem is not a few rotten apples. It is the orchard’s design.

Take “code stacking” for example, in which laboratories bill multiple individual genetic test codes rather than a single panel code, often inflating reimbursement. In one analysis, laboratories used between 1 and 12 billing codes for hereditary cancer panels with the same indications for testing, with estimated average charges ranging from $679 to $8,589 for ostensibly comparable tests. The repetition of these behaviors across companies suggests systemic incentives, not isolated misconduct.

Ample Medicare billing data, whistleblower cases, and Department of Justice settlements point to patterns, not anomalies: high-volume genetic panels ordered with little clinical justification, molecular pathology tests billed under grab bag and overly permissive billing codes, and aggressive marketing and patient harvesting practices targeting vulnerable populations.

A key driver is opacity. Many laboratory-developed tests (LDTs) are marketed under similar or identical names despite meaningful differences in design, accuracy, and intended use. To a clinician or payer, they appear interchangeable. In reality, they are not.

This naming ambiguity allows lower-quality tests to ride the coattails of better-validated ones, while still commanding reimbursement. Fraud, in this context, is not always a dramatic act. It is often embedded in routine billing.

Claim 2: Precision medicine advanced by genetics is worth the cost due to improved patient outcomes

The second argument will appeal to aspiration. Labs will emphasize that genetic testing is the backbone of precision medicine and therefore a worthwhile investment for CMS, despite the ballooning costs.

Laboratory lobbyists and industry groups will use the ‘age of precision medicine’ argument to justify the fact that genetic testing is now costing Medicare almost as much as all other outpatient laboratory testing combined. In the most recent analysis from 2024 Medicare data, genetic tests accounted for 43% ($3.6 billion) of total Medicare outpatient laboratory spending, despite representing only 5% of all Medicare outpatient laboratory tests performed.

There is some truth here to the value of genetic tests. Selected genetic tests have demonstrated clear clinical utility, improving diagnosis, or guiding treatment that has resulted in the extension of life.

But the leap from “some” tests leading to improved patient outcomes to “most” tests leading to improved patient outcomes is where the argument breaks down.

The evidence base for many marketed genetic and molecular tests remains thin, heterogeneous, or entirely absent. Clinical utility, in the rare instances where users or a test developer has demonstrated it, is often extended as a halo over other tests from different labs offering similarly marketed tests. While the me-too unproven tests share functional branding and billing codes with the proven test, they often differ in methodology or performance in nontransparent ways. In short, due to inherent differences in methodologies, the me-too tests may not improve patient outcomes at the same effectiveness as the proven test, and perhaps not at all.

This is not precision medicine. It is approximation at scale.

For policymakers, the distinction matters. Without evidence that a test informs clinical decisions in a way that improves health, its value to Medicare and Medicaid remains unproven.

If labs wish to claim that their tests save or improve lives, they must demonstrate it for each distinct test, whether it be market leader tests or the me-too tests.

Claim 3: Oversight on labs slows innovation at the expense of lives saved

Finally, expect labs to issue a warning: stricter oversight of labs’ patient recruitment and billing practices will slow innovation, delay access to cutting-edge diagnostics, and ultimately harm patients.

This argument rests on a critical assumption that has not been established, namely that the current volume and acceleration of precision medicine related laboratory testing is delivering life-saving benefit at scale.

In many cases, that evidence does not exist, or exists to the contrary.

Innovation in healthcare is not defined by the number of tests doctors order for patients, or the increasing complexity and expense of molecular pathology testing, but by their impact on patient outcomes. Requiring that the right patients receive the right tests (which have proven utility), at the right time, does not inherently suppress innovation; it helps distinguish between meaningful advances and the noise

Well documented efforts by some genetics labs to recruit as many patients as possible results in the dilution of the true value of genetic testing. Even if, for example, a cancer genetic testing laboratory has a test with proven clinical utility for patients with certain cancers, testing patients without the relevant cancers would dilute its measurable clinical utility. And due to the existence of false positives in diagnostics, testing patients recruited inappropriately could ultimately harm more patients than the test helps.

Furthermore, the absence of billing standards can distort innovation. When reimbursement is available without robust evidence requirements, the market rewards proliferation over validation. Without oversight, high-quality tests must compete with lower-cost, lower-evidence alternatives that are marketed as equivalent or better than the proven tests.

Stricter oversight of laboratories patient recruitment and requests for reimbursement would provide CMS opportunity to ensure that the goals of precision medicine are achieved. It would not halt innovation. It would redirect it toward demonstrable value.

A path forward

If CMS wants to reduce fraud, waste, and abuse in laboratory testing, it should focus less on chasing individual bad actors – which is proving to be an infinite game of whack-a-mole – and more on correcting structural weaknesses.

That begins with transparency. Tests that differ in design or performance should not share indistinguishable names in billing systems. Clear differentiation would make it harder for lower-quality tests to masquerade as established ones.

It also requires evidence discipline. Coverage and reimbursement should be tied to demonstrated clinical utility at the test level, not inferred from category-level evidence.

The CRUSH RFI presents an opportunity to reset expectations. The laboratory industry will make its case. CMS should be ready to look past the familiar narratives and ask a simpler question: not whether a test could advance medicine, but whether it does.

Julie Eggington is CEO of the Center for Genomic Interpretation and founder of Grandview Consulting. Ekaterina Cleary is a Data Scientist and Leeza Osipenko is CEO at Consilium Scientific, where she leads the development of This post is part of their work at Evimeter, a quantitative framework for evaluating the strength of clinical evidence supporting breakthrough medical devices and diagnostics, funded by Arnold Ventures.

So it’s a little absurd to be worrying about fraud and corruption in health care. But apparently HHS is. At least Oz and RFK Jr are going on about Somalis defrauding Medicaid and Armenians running fake hospices in California. (Let’s not even consider the optics of a Turkish citizen with close ties to the Erdogan regime criticizing Armenians–I mean the genocide was over a century ago!)

But of course, fraud and corruption in health care has been going on forever. Back in 2011 a Florida man was convicted of Medicare fraud to the tune of tens of millions and got a 50 year sentence. Don’t be surprised that Trump commuted his sentence. And that’s just one of thousands and thousands of cases, mostly by providers inventing fake patients to defraud Medicare or Medicaid.

But the ones who get convicted and go to jail are the amateurs.

If you’re a big company in health care, you fight with lawyers and you settle. For example, every big pharma company has settled for things like off-label promotion of their drugs. GSK paid $3bn, Pfizer over $2bn, J&J over $2bn. In fact back in the 2000s THCB had a regular correspondent called The Industry Veteran who basically suggested that whistleblowing in qui tam suits inside big pharma was the way to wealth and fame.  And of course HCA in its days when Rick Scott – now (somehow not a) convicted felon as well as Florida senator – settled for $1.7bn. This was all back in the 1990s and early 2000s, but it’s all still going on.

The venue though may have moved. Risk adjustment in Medicare Advantage has become one of the biggest venues for fraud. The key here is that the DOJ and HHS found that while Medicare Advantage plans were upcoding their patients, and therefore getting paid more for them, they weren’t actually delivering more services.

Stock analyst Michael Ha, appearing on Big Short guy Steve Eisman’s podcast showed that after United HealthGroup in 2017 acquired HealthPartners – the Los Angeles medical group – somehow its risk adjustment factor (RAF), which is what determines payment and upcoding, went from 1.0 to 1.5 within a few years. Did those patients really get 50% sicker?  Well at roughly the same time a WSJ investigation found that many patients who were upcoded by multiple plans (to the tune of $50bn) didn’t actually get any more treatment. CMS’ attempt to audit overpayments and then impose across the board fines were stymied by a court decision in 2025 that meant that they could only impose fines about the very small number of patient cases they had actually audited.

It’s amazing what a great return spending on lawyers and lobbyists gets you. That’s why it’s one rule for the giants and another for the little guy.

Stanley Warren has been on a tear on Linkedin looking just at CVS/Aetna. A couple of months back they settled an upcoding case for $117m. But Stanley’s calculation is that over the course of the upcoding they’d probably generated some $24 billion in additional payments. 

Sometimes the fines do exceed the profits. EMR vendor Practice Fusion made a quick $1m enticing doctors to push Oxycontin for Purdue Pharma but they ended up getting fined $154m (although to be fair that wasn’t just for the oxy). I suspect Aetna thinks it got a better deal by paying $117m when it received $24 billion! 

If (and it’s a big if) we ever get back to the rule of law in this country, we need a deep investigation into all these criminal corruption activities in health care and real consequences including prison sentences for the executives involved. But if we are being honest that’s very unlikely. We’ve gone through decades of allowing this kind of bad behavior and writing it off as the cost of doing business. 

For that matter we’ve done nothing much about much bigger scandals. OK, a few people went to jail over Watergate, but not Nixon. Iran Contra was basically a launchpad for Ollie North’s punditry career. Apart from a brief stint in jail for Cheney’s chief of staff, no one was held to account for the lies getting us into the Iraq war, the CIA torture sites or the corruption in the run up to the 2008 crash. And while the Koreans and Brazilians put their Presidents in jail for attempting coup d’etats, we let Trump get re-elected.

America does have a glorious history of trying to prevent this corruption. Then Senator Harry Truman ran a very successful committee that saved billions by highlighting waste in military spending in WW2. In the 1970s and 1980s Democratic senator Wiliam Proxmire gave out the Golden Fleece awards for the $435 hammer and $600 toilet seat. But in general we seem to have lost interest. Trust in government is so low, and partisanship so tribal that it’s unlikely we can come together on a real anti corruption program.

But what can we structurally do in health care to make this kind of thing less likely? 

I’d argue that the best way to do this is to restrict the amount of payments that get made for individual transactions. Obviously if you have a largely fee-for-transaction system, there will be temptation to either inflate the fee, inflate the transaction or fake the transaction. If instead we attribute a payment to a contractor for an individual’s care, such as an annual payment to a primary care physician for a fixed set of services, that contractor will be incented to make sure any subcontractors they use are not being fraudulent. We actually had an example of this when recently several ACOs identified ultra high cost wound care that was attributed to them.

There will still be examples where organizations benefitting from fraud don’t bother stamping it out. One instance was when Centene, a for-profit Medicaid plan, got paid for the same recipient in two states. There have been other cases when brokers switched people between plans without their knowledge. We need to remove the profit incentive from these arrangements.

But more widely at some point we are going to have to acknowledge two things. 

First we need to move the health care system to a government-financed flat fee system where most organizations should not be making profits. If the vast majority of people working in health care make a salary and most organizations are on a fixed budget, no one in that situation will be happy to see a fraudster get money that should belong to them.

Second, we need a massive culture shift in health care. The underlying ethics should be to take care of patients and not be concerned about the money that they are making or costing either the individual or the organization.Luckily most people in health care would prefer this and most clinical professionals have sworn an oath to uphold it. What we have to do is put a structure in place where doing the right thing is the easiest thing, and where committing fraud is hard.

I’m not holding my breath for any of this to happen tomorrow, but those of us in health care should be speaking out against the fraud and corruption we see, and against the overall system processes that encourage it.

Matthew Holt is the publisher of THCB

Protecting Healthcare and Our Democracy: A conversation with Mandy Cohen, Former Director, Centers for Disease Control and Prevention PLUS former NC Governor Roy Cooper & Dr. Donald Berwick, IHI founder/former president

REGISTER & DONATE

A matching fund of $70,000 is in effect for this event!

This supports the Movement Voter Project which makes grassroot investments to organize to support democracy at the local level.

I’m a co-host–Matthew Holt

Matthew: Another THCB Spotlight, I am here with Bevey Minor who a year ago I interviewed as Consensus Cloud Solutions and now your sign says eFax. So, what the hell happened?

Bevey: Interesting question, Matthew. The company is Consensus Cloud Solutions. And the company’s always been Consensus Cloud Solutions since we spun off and went public ourselves. You’ll notice at our booth we’ve got the eFax brand — it’s eFax by Consensus Cloud Solutions. The reason we are showing up as eFax is because this year at HIMSS we really wanted to set the record straight: digital cloud faxing is not the problem with interoperability. Paper faxes are, but digital cloud faxing is not the problem.

The problem is all this unstructured data — all the unstructured data that happens with faxes, with scanned images, with TIFF images. All that unstructured data can’t be queried. It can’t be part of TEFCA. You can’t query what you can’t find.

Cloud faxing is send and receive all day long, and we do that very well and have been doing it for 27 years. About three years ago, we introduced an intelligent extraction solution. That solution doesn’t even have to start with the fax, but it allows the “find” piece to actually become the critical thing that we need to do. CMS defines interoperability as send, receive, find, and integrate. Fax technology handles send and receive all day long, but can’t find. So once we introduced a “find and intelligent extraction” solution, we can fire up TEFCA.

I’ve talked to a lot of regulators, including Dr. Thomas Keane and Amy Gleason with the CMS Align networks. You can’t ignore this pile of unstructured data and just assume the industry is going to go magically everything’s on FHIR. We’re all using FHIR because all of this stuff has really important patient information in it.

What we want to solve in the industry is: don’t say we have to axe the digital cloud fax. Let’s axe the paper fax machine. Digital cloud faxing isn’t going away — in fact, it’s growing, especially as we get rural health off of paper fax machines. The next level of maturity is digital cloud faxing. From there, once it’s digital, now you can do all sorts of things with it.

When we introduced electronic health records during meaningful use — I was at Allscripts at the time — our dream was that we would take this paper record and transform it into an electronic health record, so we could just get rid of the paper. Once we did that and there were discrete data elements in that EHR, we could do population health, clinical decision support, efficacy, all sorts of things — because there are discrete data elements now inside that electronic health record. That’s what a digital fax will do with the capability to do intelligence on top of it.

So we want to make the industry understand that the fax is not the problem. Extracting it and getting rid of all that unstructured data is the solution.

Matthew: Okay. So since we last talked, I’ve had some journeys around the health care system. I was sent a referral from my lovely primary care doc at One Medical. Somehow the referral from One Medical got into the Blue Shield system and created a prior authorization. So I have an echocardiogram — I’ve written this up on my blog — and you know what’s coming next.

I call the number on the referral that One Medical wants me to call for the Marin Health imaging center, and they go, “We don’t have any referral.” I say, “No, no, no, they sent it to you, I see the PDF.” I call back two weeks later and they go, “I don’t know, we don’t have the referral.” So finally I call and say, “Can I email you the referral?” I get someone’s email and email them the PDF and that person goes, “Oh, this is a cardiology imaging referral, not a regular imaging referral.” It turns out it goes to cardiology, which is of course in the same building.

But here’s where it gets interesting. They say, “We can’t take an email” — they  told it was because HIPAA wouldn’t allow it.. So I downloaded some fax product — I think it may have been eFax — a trial that later I had to remember to cancel. So I fax everything. Now they’ve got a referral which was typed on a PDF, and may or may not have a piece of paper somewhere with the prior auth as well. It’s arrived, and now they can actually do something with it.

What is wrong with that picture, and what can eFax do in that picture to fix it and get the data out that’s required to help?

Bevey: A lot of things are wrong. Number one, you’re going to get delayed treatment. You’re going to get frustrated as a patient, and your health outcomes — you may even abandon care.

Matthew: That’s actually what almost happened. I was about to abandon it. I did keep going — I actually do have a problem. 

Bevey: But your health outcomes could have been horrific. But you bring up a really interesting use case that we are doing today with imaging centers. Hospitals that own their imaging centers, or IDNs that own imaging centers — 

Matthew: this imaging center, by the way, has both cardiology imaging and regular imaging in the same building, but those are different fax numbers.

Bevey:  So as an example — and this is an actual real use case we’re doing — we intercept the fax, we pull out all the structured data: all the patient demographic information, we pull out what the order is for. Let’s say it’s a mammogram with contrast. We map it into the radiology system so they can see the order without anyone data-entering anything. If it had come as a fax order, somebody still has to data-enter it, or it’s sitting in a pile of all these orders they’ve received.

Once it’s extracted and mapped into the imaging center’s RIS system, they can see — because we read that this isn’t just a mammogram, this is a mammogram with contrast. Now they can see: do we have a machine capable of a mammogram with contrast? Is that machine available on Monday, Tuesday, Wednesday? And once all that’s done, no human has actually touched this. It’s getting looked at and mapped into their scheduling system, so you can send a text message directly to the patient within 24 hours: “We’ve got your order and we have opportunities for you to come in Monday, Tuesday, or Wednesday.”

All of that is because we pulled out the structured data from that fax for that image. And for every day that a patient expected to get the next level of care but is delayed — a mammogram or some sort of imaging — 25% of trust in the system is eroded. That imaging center needs to get that referral because it’s their lifeblood. The patient is going to go somewhere else, or they could just say to the doctor, “They haven’t called me — find another place.”

Matthew: So if they had plugged in eFax somewhere in the middle of that workflow, the referral would have been intercepted, read, and automatically routed.

Bevey: The product that does that is called Clarity — that’s our AI data extraction piece. And once you’ve created all these discrete data elements, if the imaging center needed it in FHIR so it could be mapped, we can send that as a FHIR message. We have a semantic interface backbone. Once you’ve got all this extracted data, now you can do prompt-based queries on it — for example, “Show me all patients who need imaging with contrast because those machines are sitting unused and we need to get those patients in fast.” And then you can use prompt questions like on any generative AI solution

Matthew: I’ll tell you something amusing that leads to another question. Eventually after the fax goes through, somebody calls me back — and of course I don’t answer, it goes to voicemail. I see the transcription of the voicemail and they’ve arranged for me the very next available appointment, which happens to be in Santa Rosa — an hour from my house. So I go to MyChart thinking maybe I can change it. MyChart shows me the appointment but doesn’t give me the option to change it — only to cancel it. Eventually I got it changed to a location 10 minutes from my house.

But this leads to a question. You have faxes going back and forth between these departments, and this one is still very manually treated — manual appointment scheduling.But there is a whole lot of data in the EMR (Epic, Oracle, et al)  There’s a lot of information sitting in there, with people trying to get it in and out by different methods, not all electronic. How does what you’re doing play into that?

Bevey: What’s interesting is that there’s a really big effort around patient access to their data. MyChart is an example. This administration — Trump’s promise and campaign was that every American would be able to access their medical information. And now we have a bunch of pledges happening around patient access to their data, with QR codes and things like that. But what do you do with all the unstructured data? A patient may have lab results, clinical information — the USCDI data set that gives the patient their entire history and summary — sitting in unstructured data that can’t be queried.

So the QHINs are only operating on a small subset of data that is FHIR-enabled structured data. It’s going to be biased, it’s going to have missing pieces. Patients are going to think, “Here’s all the information I need” — until they drill down and say, “Two years ago my hemoglobin A1C was pre-diabetic, I want to find that.” And it’s an unstructured PDF somewhere.

When you look at an electronic health record, there’s a lot of PDFs attached to the patient’s chart. 

Matthew: I was just looking at mine. Both One Medical and Epic pieces have different structures. Some of them they processed internally and show me my lab numbers with a nice little indicator, but I’m not sure they’re all in one place. 

Bevey: We have a lot of work to do to get a common record for patients. I have two MyChart accounts. It’s easy to link them, but I also see another system that uses FollowMyHealth or something like that, and it’s not integrating as it should.

I know the promise is great — I remember the days, and you probably do too, Matthew, where every patient was going to carry around a USB drive or a magical smart card.

Matthew: You can do that in France. Probably not available in America.

Bevey: But I do think there’s promise, and the promise is to really recognize that there’s a whole bunch of data. I look around at a lot of these companies here at HIMSS and they boast they do workflow enhancements, prior authorizations, population health — but when I ask them what they’re doing with all the unstructured data, they say, “We have an OCR solution.” Whenever I hear that, it’s not really extracting data at an intelligent level. OCR knows that something is a B, a V, and a Y — my first name — but doesn’t even know it’s a name. If you use intelligent extraction, it does contextual work: “Go find me the name of this patient in this form,” which could be in the upper left-hand corner or the bottom. It really understands how to ask the right kind of questions. OCR is a gap.

So we are playing a role where we can create very sophisticated intelligence off of all this unstructured data, so many of these workflow solutions can have better data going in.

I’ve talked to Dr. Thomas Keane, head of ONC, and I said: “Do you really want to make TEFCA successful? You’ve got to talk about how we manage getting all the unstructured data into a FHIR-enabled TEFCA framework.” And rural health — health equity has been a passion of mine —  Rural health can’t rip and replace. They can’t support all the FHIR-enabled solutions they’re supposed to support. The CDC in their data modernization initiative is saying all electronic case reporting has to be done via FHIR — well, the community clinic definitely needs to populate these repositories, otherwise those databases are going to be completely biased.

And if you look at the prior authorization rule 57 going live January 1st for Medicare Advantage patients — what is that substance abuse clinic that needs to ask for an authorization so a patient can stay in treatment for another 90 days? They can’t send a FHIR-enabled prior authorization to get that approved. We intercept their fax. We say to them — and we do a lot in rural health — “Just keep doing what you’re doing. We’ll intercept it in the middle and send that as a FHIR prior authorization.” The industry needs to understand that not every provider has tech equity.

Matthew: Alright, give me your forecast. There are all these different holes — unstructured data, ambient AI, people asking CharGPT questions. Do you think the consumer, patient, smaller provider (or any provider) experience gets a lot better in two years, or are we still fighting this? Is Amy Gleason’s team going to succeed, or are we still going to be in a war in five years’ time?

Bevey: I still think healthcare is transactional. If everybody was in a closed system, everybody talked Epic, it would be great. But we don’t have that kind of system. And this is an administration of deregulation — open for whatever you think you can do from an innovation perspective. When you have tight regulation, you see solutions emerge that actually help providers and technology companies satisfy that regulation — which is happening with prior authorization, which is why we’re really helpful for those that don’t have FHIR resources.

But with deregulation and pledges, I think in two years if you gave me the same interview, we would see some baby steps. And also because there’s a growing consciousness among patients — “I can actually do a lot with AI. I can push hard on my docs. I can say this is the best treatment.” And not always use Western medicine to do that. Patients are going to hold a ton of power, and with that power they’re going to say, “I have a responsibility to take care of my health.” I think that’s going to move faster than we even thought, especially with ChatGPT Health, Gemini — patients are using these tools. They’re reading their reports and asking questions.

Matthew: I’ve uploaded a lot of my reports from my imaging and I now use it all the time.  But you know, there are people who have been dragging their feet, and a couple of years into this, it’s starting to feel like fantasy being repeated three or four years in a row. So I’m hopeful, but I’ll acknowledge the joke of the moment. I do think there’s a lot of consensus that we’ve got to get some stuff done.

You and I have been doing this long enough to remember when you could go anywhere with your data, and now everyone else is fighting over information blocking.

Bevey:  The patients don’t care what the pipes look like at the back end. The docs don’t even care. The docs are saying, “I don’t care what you do as a technology vendor. I don’t need to see all the stuff flowing through. I just need information so I can make a more informed decision, treat my patients the best way, and make money doing it.”

We’re going to have some headwinds where a lot of hospital systems are going to have to deal with uncompensated care. HIMSS is a little light this year, and I think hospitals are looking at budgets and saying, “We can’t give you the same IT budget to go shopping. We’re going to have to shut down service lines — behavioral health, women’s health — just to manage the uncompensated care.” That’s a new world we never thought we’d see. But I think we’ve got to weather through that too.

Matthew: I’ve been speaking with Bevey Miner from Consensus Cloud Solutions, Thanks for your time

By KIM BELLARD

Most of us can identify dogs from cats just by the sounds they make. We could probably even separate a dog’s bark from a wolf’s howl. If you are a nature lover, you might be able to identify different species of birds by their calls.  If you are a cetologist, you might be able to separate the vocalizations whales make versus those dolphins make. Across the animal world, we’ve learned the different sounds that different species make, which has been useful in our survival.

But did you ever wonder if you can identify, say, e coli from other bacteria?

It turns out that you can, thanks to research at Delft University of Technology (TU Delft) in the Netherlands. Four years ago, they showed that bacteria made noise, which was, in itself, a startling finding (admit it: would you have ever guessed that?). They used a thin layer of graphene to create a graphene “drum” small enough to fit a single bacterium. Team member Cees Dekker observed: “What we saw was striking! When a single bacterium adheres to the surface of a graphene drum, it generates random oscillations with amplitudes as low as a few nanometers that we could detect. We could hear the sound of a single bacterium!”

The team used this finding to accomplish an important purpose: to find out if bacteria were resistant to specific antibiotics. If an antibiotic was applied and the sound continued; it hadn’t worked. If the sounds stopped, the bacteria had been killed.

The team wasted no time in creating a start-up – SoundCell – to commercialize the finding. It promised to identify the “right” antibiotic in one hour, rather than subjecting patients to rounds of different antibiotics in search of one the bacteria wasn’t resistant to.

The team isn’t resting on their laurels. Some of them got to wondering, huh, I wonder if different bacteria make different sounds. And, their latest research shows, not only do they but, through machine learning, those different species can be distinguished. Team lead Farbod Alijani says. “With this new study, we take a significant leap forward: we show that each bacterial species has its own nanomotion signature.”

Mind. Blown.

The researchers focused on three bacteria that are common in hospital settings: E. coli, S. aureus (which causes staph infections) and K. pneumoniae (which causes pneumonia). They tested two different machine learning models; one correctly classified the bacteria 87% of the time, and the other 88% of the time.

“By combining SoundCell’s existing antimicrobial testing prototype with this machine learning model, we can identify the bacterial infection and determine which drug is effective at the same time, based purely on the sound of a single bacterium,” says SoundCell CTO, Aleksandre Japaridze. Leo Smeets, physician microbiologist at RHMDC adds: “This approach eliminates the need for culturing, which normally takes days. And because the diagnostic steps are no longer performed sequentially, we can save even more time.”

“It’s a completely different way of interpreting the different species,” Dr. Japaridze says. “Not chemically or biologically, with markers and genes, but just purely on…mechanical behavior.”

Their paper concludes:

To sum up, our results show that combining the high sensitivity of graphene nanomotion sensors with ML enables fast, label-free AST and identification of bacteria. Since the trained models analyze nanomotion signals from individual cells, results can be obtained within 1-2 hours, eliminating the need for time-consuming culturing steps. With further development, this approach could establish nanomotion spectroscopy as a powerful platform for real-time diagnostics and for studying cellular biophysics and antimicrobial resistance.

They’ve been testing sensors in the lab, so one of the next steps is to show they can be used in actual hospital settings. They’re testing a prototype at two Dutch hospitals (RHMDC and Erasmus Medical Center). Professor Alijani believes: “This close partnership between scientists at TU Delft, a start-up and a hospital is quite unique. We have the entire knowledge chain working together.”

The potential impact is huge, with over 1 million deaths due to drug-resistant bacteria annually. “We have already shown that we can reduce antimicrobial susceptibility testing to one hour,” says Dr. Japaridze. “If we can combine that speed with species classification using the new machine learning model, we could create a globally unique device that dramatically accelerates diagnosis and treatment. And that would be highly valuable in the worldwide fight against antimicrobial resistance.”

———

I love the kind of curiosity that makes one wonder, hmm, do bacteria make noise? That’s not a question most people would ask themselves. I love the scientific expertise that figured out a way to actually detect that noise, at the level of a single bacterium. I love the realization that perhaps different bacteria make different noises, and the expertise to use machine learning to distinguish them. And, of course, I’m excited that all this might lead to practical applications that could save lives and avoid needless rounds of antibiotics.

Next thing you know, we might find out that bacteria not only make noise but use them to communicate. It wasn’t that long ago that we were arrogant enough to think that only humans communicate vocally, only to find that that many animal species use sound to communicate. Heck, we’ve even found that that plants “scream,” sending out messages we’re oblivious to.

It makes you wonder: what else are we missing?

I have this wild thought that our bodies are a cacophony, with all our cells and all of cells of our microbiota chiming in. When we’re healthy, perhaps they combine to create a finely tuned symphony, but when something is off it’s like an instrument in the symphony is badly tuned, off the beat, or missing. Perhaps if we listened the right way, we could use those sounds to more quickly and more accurately diagnose and treat the problem.

That’d be some 22^nd century medicine.

So kudos to the scientists at TU Delft, good luck to the entrepreneurs at SoundCell, and to all you researchers in the world: keep asking these weird questions!

Kim is a former emarketing exec at a major Blues plan, editor of the late & lamented Tincture.io, and now regular THCB contributor