FDA approves unique new wearable for people with diabetes

This week the FDA approved the first wearable device that monitors both glucose (sugar) and ketones (a byproduct of fat breakdown) in the blood. The device provides absolute levels of both biomarkers and can also tell the user whether those levels are rising or falling.

COMMENTARY: This device has the potential to be a lifesaver, especially for people with Type 1 diabetes. When there isn’t enough glucose in the bloodstream (hypoglycemia), the body burns fat for energy instead, releasing ketones into the blood. If this process goes on for too long, it leads to a condition called diabetic ketoacidosis, which, if not recognized or treated, can be fatal.

We have long been intrigued by wearables — Mara has tried lots of them over the last year. Stay tuned for a feature issue dedicated to wearables this fall.

Biomarkers for impending disability in elders

Research on increasing the human lifespan is getting a lot of press these days. But, as we note from time to time, it’s not our lifespans that need to be increased so much as our healthspans: the amount of time we have in a healthy, functioning body. And if our healthspans are about to reach their limit, it would be good to know that, so we can plan for it.

Research published this spring demonstrated a way to determine whether an elderly person is at risk of developing disability. In the paper, disability was defined as needing help with shopping and daily decision making, as well as help getting up, standing up, or standing on one leg.

It found that higher levels of beta-2-microglobulin (B2M) and cystatin C were consistently associated with a greater likelihood of developing disability. Each increase in these biomarkers was associated with significantly higher disability risk (hazard ratio 1.35 for B2M and 1.42 for cystatin C), even after accounting for age, sex, kidney function, lifestyle factors, and other potential confounders. The research was initially conducted in Japan and then validated in Italy, and the results held up across ethnic groups.

COMMENTARY: These results make sense, given that both of these biomarkers increase overall inflammation in the body: B2M by triggering the release of pro-inflammatory molecules and immune cells, and cystatin C by inhibiting molecules and cells that decrease inflammation. However, they don’t tell us how close to disability a given person is. We also don’t know whether the trajectory is reversible.

Like most (maybe all) new discoveries - it is only the first step. If this diagnostic can predict disability, can it be used to understand the natural history or the condition? We hope that more work is done to understand that.

The best tests may be the simplest ones

Most of the articles we run in this newsletter cover new ways to diagnose disease. But sometimes a brand-new technique or biomarker isn’t what’s needed. Sometimes, to improve disease diagnosis, we just need to effectively use the tests and / or biomarkers we already have.

We came across two cases in point this week.

Improving CKD diagnosis with inexpensive, consistent, targeted screening

Chronic kidney disease (CKD) is the ninth leading cause of death worldwide, and is projected to rise to number five by 2040. However, up to half of cases may remain undetected even in high-income countries — despite the availability of simple urine and blood tests. The situation is even worse in middle- and lower-income countries, as well as among women and people who aren’t white. This spring, the Lancet published a series of three research articles devoted to CKD, the first of which focuses on diagnosis.

COMMENTARY: In its mild to moderate stages, CKD is asymptomatic, but can be detected based on changes in the amount of protein in the urine. That doesn’t necessarily mean we need to screen absolutely everyone for the disease, the authors note, because we also know what the risk factors for CKD are. In addition to having a family history of kidney problems, they include a rogue’s gallery of modern-day chronic disease: diabetes, high blood pressure, obesity, and heart disease. Routinely testing folks with those issues — using inexpensive tests that we already have — could make a big difference, the authors note.

Looking for liver disease with an underutilized analyte

People with early liver disease are in a similar situation to those with early CKD — while they have no symptoms, a biomarker is silently rising in their bloodstreams. In their case, the warning sign is an increase in bile acids, which the liver produces to help digest fats, and which normally don’t show up much in blood. Unfortunately, a bile acid measurement isn’t included on most routine blood test panels for “healthy” folks.

Research published earlier this month offered a simple way to determine whether bile acid levels in blood are too high — a condition known as cholemia. When red blood cells (RBCs) are exposed to these acids, their cell membranes get more rigid, which makes the cells less likely to burst.

The test measures how tough RBCs have become by exposing them to an NaCl solution, otherwise known as salt water. When the solution has less NaCl in it than the cells do, water rushes into the cells. (That’s good old osmosis at work.) Healthy cells burst when this happens. Cells that have been exposed to bile acids have tougher membranes, so they don’t. (See graphical abstract of the paper below.) And the higher the bile acid levels, the more rigid the cells become, so you can measure the level of cholemia by adjusting the amount of NaCl in the solution.

COMMENTARY: The liver disease paper is a proof of concept done in mice, so further research is needed to make sure human cells behave the same way. But given that roughly 25% of US adults have metabolic dysfunction-associated steatotic liver disease (MASLD), it would be good to have a simple, inexpensive test to find people whose livers are suffering in silence.

Cyclospora

This summer’s cyclospora outbreak is probably the largest foodborne outbreak in history. As of this week, the CDC has logged 17,180 confirmed cases, and is investigating an additional 11,844. (Last year by this time we had 1,180 cases.) And that’s just what the CDC knows about, which, they note, is almost certainly an undercount.

Ebola Bundibugyo

The Ebola outbreak in the Democratic Republic of Congo is now the fastest-growing in history, with 5,514 cases and 2,642 deaths. Its case-fatality rate is between 40% and 60%.

Measles

Two unvaccinated people have died of measles in Lancaster County, Penn., the first deaths in that outbreak. And earlier this week, a medical center in the same state reported that a visitor to its neonatal intensive care unit may have had the disease. Since measles is transmitted through the air and can linger there for up to two hours, that means medically fragile infants too young to be vaccinated may have been exposed.

The current larger outbreaks in the US are shown below: Arizona / Utah, Ohio, Pennsylvania, Washington.

We will be off next week in anticipation of Labor Day. We will return on Thursday, September 10th with a new issue and new podcast. Enjoy the “unofficial” end of summer.

Another way to solve the pulse-ox problem

Blood oxygen level is one of the core vital signs. For decades, clinicians have measured it using a pulse oximeter — that little clip-on doohickey that goes on your finger. But during the COVID pandemic, it became clear that pulse-oxes can deliver inaccurate results in people with darker skin colors.

Back in January, we reported on the development of a device intended to improve the pulse ox. Instead of simply looking at light that passes through the skin, the sensor uses polarization to filter out the light from the layers of the skin where pigment is present. Another option we recently learned of does the opposite. Instead of filtering out the patient’s skin color, it checks skin color first, and then adjusts both the light it uses to obtain a signal and the results it gets accordingly.

Previous research showed that the tool worked extremely well on people of all skin tones whose blood oxygen was at normal levels (its oxygen-saturation measurement was accurate to within 1.4% of a standard reference oximeter). This new research demonstrated that it does just fine on folks whose oxygen saturation is low, too (it was accurate to within 2.9% of a standard reference oximeter, which meets the FDA requirement). And, as promised, skin tone didn’t bias the device’s measurements, even at very low oxygen levels.

COMMENTARY: It’s time to ditch BMI as proxy for obesity

Folks, a person’s BMI doesn’t necessarily tell you whether they’re obese. The amount of abdominal fat they have is what matters, and BMI isn’t the best proxy for that. Their waist size is.

This topic really shouldn’t go under “New and Noteworthy,” because by now the idea isn’t new. The European Association for the Study of Obesity updated its guidelines to this effect over two years ago (we covered it back then). But over the past few months, several papers have come out that hammer the concept home.

The two most recent ones appeared this month. One showed that waist size alone was almost as good at predicting obesity-related health issues as a slightly more complicated metric from the Lancet that also involves waist-to-hip and waist-to-height ratios. The other found that waist circumference alone was a better predictor of cardiovascular disease than BMI. And back in early June, a study warned that a quarter of Americans whose BMI is normal are, in fact, “clinically obese.” (The paper used the Lancet measurements plus evidence of health issues associated with extra fat to make that determination, and it made a big splash in the lay press.)

So clinicians, get out your fabric tape measures. This test is as cheap and easy as it comes. You don’t even have to use a calculator. What’s not to love?

When it comes to psych illness, the f in fMRI stands for “fallible”?

Psychiatrists have long been in desperate need of objective biomarkers to supplement traditional diagnosis based on patterns of symptoms. For a while, it seemed like functional MRI (fMRI) of the brain might fit the bill.

MRI’s superpower is its ability to image soft tissues, which it does by exciting and relaxing protons (aka hydrogen ions) using huge magnets and radio frequency pulses. Different soft-tissue types react differently to this prodding, allowing computer analysis to create images.

The “f” in fMRI makes use of the fact that when neurons are active, they need more fuel, which means more blood flow. Blood is about 80% water, and water (H₂O) has lots of hydrogen ions. So fMRI shows which parts of the brain are working harder than others. It was, therefore, reasonable to theorize that psychiatric disorders would exhibit detectably different patterns of neuron activity across areas of the brain.

Unfortunately, they don’t, as a recent paper In Nature Neuroscience convincingly demonstrates. The researchers examined decades of evidence, and while they found lots of fMRI differences, they didn’t see consistent patterns for specific disorders.

COMMENTARY: MRI and CT combined, used with an arsenal of task-specific contrast agents, have been very successful tools for diagnosing physical disorders that are characterized by structural changes. (Some examples: the myelin destruction of multiple sclerosis and the brain atrophy of advanced Alzheimer’s.) Why does fMRI not succeed at finding psych disorders?

The answer lies in its resolution and its speed. Compared to CT, MRI has very low resolution: While a standard CT scan can detect structures as small as about 1 cubic millimeter, the best MRI is about 10 times worse. It’s also slow — it takes several minutes to get that image. Trying to detect millisecond-duration changes in neurons that are packed 50,000 is just a bridge too far. This alone might be enough to doom any potential in psych disorders. (If you’re wondering why we don’t just use CT, that’s because CT can’t “see” soft tissues well.)

What of the future? Is there any hope that developing fMRI techniques will solve the resolution problem? Maybe, but is brain behavior just too complex, dynamic, plastic, and diverse for a “what happens where, when” approach to ever work at any resolution? Lots of very innovative thinking is going into finding out.

As we get older, screening by age makes less and less sense

Who should be screened? When and how often? Most clinical guidelines use age as the primary factor driving the answers to these questions, but a recent study demonstrates just how arbitrary and wrong this can be.

As a person ages, on average they compile more and more health problems (aka comorbidities). At the same time, the older a group of people gets, the bigger the differences there are between one person’s set of comorbidities and the next. For example, you might have diabetes, high blood pressure, and chronic back pain, while I have cardiovascular disease and arthritis in my hips and knees, and my friend has cataracts and sleep apnea. (Technically this is an application of Taylor’s Power Law.)

Studying the health records of 240,000 lower-income adults in four different health systems, researchers found that one-size-fits-all screening programs based on age work less and less well the older we get. Instead, they found that it would be more effective to screen people based on the group of comorbidities they have.

COMMENTARY: This recommendation would require an enormous paradigm shift, but we think it’s one whose time has come. Clinical guidelines for screening and care need to become much more tailored to individual health. (Of note, we receive more comments and questions about screening guidelines than any other topic in the Newsletter.) Given that we now have both electronic health records for nearly all receiving care and the powerful tool that AI provides, this should not be rocket science. Of course, the difficulties are more practical. Access to care is “jagged,” to use an AI term of art, and screening compliance, amongst both patients and physicians, leaves a lot to be desired.

Nevertheless, individual screening recommendations need to be made with comorbidity data in hand. This is a Bayesian concept – average prevalence of any given disease in a population should only be the starting point (aka the prior). Comorbidity and personal history (aka the likelihood factor) can then be used to provide individual screening recommendations.

🎙️ New Episode Alert: Mike Klein, CEO of Genomenon, tells us how his company's human and AI curators mine the global scientific literature for evidence describing genetic variants.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

AI enables personalized disease risk prediction to manage health

AI’s superpower is its ability to find patterns across diverse data sets. If we use that superpower to its maximum — by analyzing our genetics, past behavior, the environment we live in, and the history of our interactions with the healthcare system — we ought to be able to find predictive patterns. The holy grail would be the ability to use personalized health predictions to take tailored, pre-emptive preventative actions.

Real progress is being made toward this goal. Two recently published models (one last November and one this month) seek to integrate genetics with electronic health record data to calculate a personal health-risk profile. The profile could then be used to design a plan to maximize health over time: what to be screened for, when and how often; which preventative drugs to take and procedures to undergo; which behavioral changes to make.

The more recently published model starts by finding disease signatures (things that typically go wrong as we age). It then assembles a tailored mix of them for each person (their portfolio of comorbidities) and predicts what each person is at risk of developing both in the next year and up to 10 years out. The model was much more accurate than traditional single-disease risk tools (e.g., PREVENT for heart disease).

COMMENTARY: The tragedy of 21st-century healthcare is that in spite of all its advances and technology, at the patient level it leaves a lot to be desired. The system remains predominantly reactive (because doctors are paid for procedures, not prevention). In addition, it’s episodic, siloed by physician specialty, and has a severe long-term memory problem.

For patients up to age 30 or so, this is a reasonable approach. But beyond that, these failings become increasingly important. This study shows the way forward for primary care: personalized risk assessment based on longitudinal, dynamic, multi-omic, multi-dimensional data, generating clinical predictions that inform interventions to maximize healthspan.

GFAP makes MS trajectory more predictable

On its “What is MS?” page, the National Multiple Sclerosis Society defines MS as “a chronic, unpredictable disease of the central nervous system.” Most commonly, MS progresses in a “relapsing-remitting” fashion: Periods of remission punctuated by episodes in which the disease reappears and things get worse. But for some people, progression simply plods on, without remission. Unfortunately, just as the National MS Society says, it’s hard to predict which version a patient will have — and someone who starts in one camp may eventually switch to the other, also unpredictably.

MS is generally thought to be an autoimmune disease in which the body’s attacks on its own nervous system cause inflammation in the affected tissues. We already have a biomarker for that: neurofilament light chain (NfL). But while high NfL levels tell you that the patient’s neurons are inflamed and that a relapse is imminent, they don’t tell you whether the patient’s disease is about to progress or not. Until recently, we didn’t know of a biomarker that could make that call.

According to a paper that appeared in JAMA Neurology earlier this month, we might have one now. It’s glial fibrillary acidic protein (GFAP). We already knew that patients with high GFAP levels were likely to have progressive disease eventually. With this study, we now know that MS in folks with high GFAP levels is 40% more likely to progress within the next year. In addition, when GFAP levels decreased in patients receiving treatment for MS, their likelihood of disease progression decreased, as well.

Better sensors mean better tests

The idea behind an electrochemical sensor is simple: It’s made of a material whose measurable electrical properties change when it binds to a chemical that drives or reflects a specific disease. In practice, it’s very hard to design a sensor that only binds to one thing. Generally, the more sensitive the sensor, the more stuff sticks to it, most of which you don’t want.

Measuring dopamine in tears

Last week, we highlighted one electrochemical-sensor-based device that manages this trick: It measures ketones in breath (they rise when the body burns fat for energy). This week, we came across another: a sensor that can measure dopamine levels in tears. Those levels directly affect mood, and low levels are also the key driver of Parkinson’s disease. To a lesser degree, low levels of dopamine also drive schizophrenia, Alzheimer’s disease, and depression.

COMMENTARY: Over the past several years, many less-invasive tests based on non-traditional analytes have been found to be just as effective at diagnosis as techniques that rely on traditional samples. Two good examples: Using blood instead of tissue to diagnose or monitor cancer (aka liquid biopsy), and using plasma instead of spinal fluid to identify brain disorders. At the same time, novel materials research is enabling the creation of selective and sensitive electrochemical sensors that can do for pennies what previously cost hundreds of dollars. This is a great thing, as lower-cost, less-invasive tests increase access, which can then drive adoption and earlier treatment.

Wearable sensors that are invisible — or meant to be seen

If you or I had to have some aspect of our health monitored using a wearable device on, say, our upper arms, we probably wouldn’t think twice about it. We’ve all seen people wearing glucose monitors there, right? How about on the chest or belly? Sure, fine — it’ll be under a shirt, what’s the difference?

How about on your face?

I don’t know about you, but I might want to think about that one.

That’s exactly why a group of researchers have developed a skin sensor that is effectively invisible. They’re incredibly thin, stretchy electrodes — so thin, they not only can’t be seen, they also can’t be felt (see image). And yet, in tests, they could be used to measure electrical activity in the eyes, facial muscles, and brain (EEG). In fact, in some cases they did a better job than conventional gel electrodes.

Of course, there’s one group of patients who might want to be able to see sensors on their faces — if they looked like, say, dinosaurs. Or maybe butterflies.

For those kiddos (and other body-art enthusiasts), there are paint-on electrodes — think temporary tattoos that can measure electrical impulses.

Painting the electrode onto the skin serves a functional purpose as well as an aesthetic one. The conductive, polymer-based ink is applied to the skin while it’s wet, so it adheres to every crevice and flows underneath hairs. That gives it the best possible contact with the skin, without any air gaps. Once it dries, the material is stretchy, so the wearer can move without dislodging it. And it’s porous, so sweat can flow right through it. How sure are the researchers that it works? They’re sure — one of them even wore one during their exercise routine.

Next-generation AI still produces racial and gender stereotypes

Back in 2024, research showed that when GPT-4 (the second foundation model behind ChatGPT) was asked to describe fictional patients with various medical conditions, it frequently reproduced racial and gender stereotypes. Things haven’t gotten any better over the past couple of years.

When researchers tested newer models o3-mini and DeepSeek-R1, which are supposed to be better at reasoning, the results were the same or worse. Where GPT-4 demonstrated racial and gender stereotyping for 67% of the medical conditions it was faced with, the newer models showed racial stereotyping for 78 – 89% of medical conditions and gender stereotyping for 56 – 67% of conditions.

Yes, p-tau217 can diagnose Alzheimer’s. But do we really need it?

Physicians have been using cognitive tests to diagnose Alzheimer’s disease for over 100 years. Do we really need anything more? The simple answer is yes, very much so.

First some history, because it is not the first time this question has been asked. Back in 2013, Medicare (CMS) declined to cover then newly FDA-approved amyloid beta PET scans, because officials were skeptical that patients’ health outcomes or survival would benefit from the tests. A roughly 11,000-person study designed to address this question conclusively found that PET scanning did make a big difference. Quicker, more definitive diagnosis drove less follow-up testing, fewer emergency visits, and fewer hospitalizations (see this 2018 summary).

A smaller study presented at this July’s Alzheimer’s Association International Conference confirmed that a blood test (combined percentage of p-tau217 and Aβ42/Aβ40 ratio) changed diagnoses for about 30% of patients in both primary and specialty care. Most strikingly, specialists found that 18% of those that they had judged to have Alzheimer’s after extensive cognitive testing, a brain CAT scan, and a 60-minute visit did not have the disease. In fact, primary-care physicians with blood-test results and a 15-minute visit were about as good as the specialists at making the diagnosis.

COMMENTARY: It’s notable that absent this test, specialists were generally prone to over-diagnosing Alzheimer’s (a hammer and nail bias?). So using the test to rule it out in those displaying early cognitive symptoms is very valuable, particularly since we know that about 45% of those suffering cognitive impairment have other disorders, many of which have effective treatments.

On the other hand, p-tau217 positivity alone has relatively low positive predictive value: about 60% of the cognitively unimpaired who test positive will not develop dementia in the next 10 years. Thus, if primary-care physicians yield to worried-well requests for a blood test (absent concerning cognitive symptoms), over-diagnosis still has the potential to become a significant concern.

AI + EHR = Early ADHD diagnosis

The earlier a child with ADHD is diagnosed, the better they do in school over the long term. While the average age at diagnosis is somewhere between seven and eight years, for many kids (especially girls and / or kids with other neurological issues) it can take much longer. It would be great to have a reliable way to find these kids early on.

A paper in Nature Mental Health offers a potential solution: An AI foundation model that uses data from electronic health records to diagnose ADHD. A foundation model, also sometimes called a general-purpose AI system, is a model that’s been trained on a very wide range of data, and that can produce a lot of different kinds of outputs. People can then build applications that use the foundation model for more specific purposes.

After being trained on the records of over 140,000 patients, the model was able to predict the likelihood of an ADHD diagnosis with about 92% accuracy by the time a given patient reached age five.

Kids who have autism as well as ADHD are even harder to diagnose than those with ADHD alone. To learn more about how one company is using maternal autoantibodies to enable early autism diagnosis, listen to this week’s On Target podcast, featuring Bob Wassman of MARABio.

Sit down and let me take your vitals

Looking for a health-tracking device but don’t want to commit to wearing something all the time? We’ve got just the thing — and we can almost guarantee it’s something you never knew you needed.

What you need is a smart toilet seat.

Just imagine! You can measure your blood pressure, heart rate, blood oxygen saturation, and respiratory rate, and monitor trends, all while doing crossword puzzles or scrolling through Instagram. What could be better?

No really, this exists. And according to the company, its results are as accurate as those obtained by instruments used in hospitals. You get the results through an app (of course), and the seat can keep track of up to five registered users.

And don’t worry — you won’t accidentally end up with blood-pressure measurements from friends who’ve come to your house for dinner and needed to use the restroom. The seat won’t measure anything unless the user is registered with it. (We know you were concerned.) And it doesn’t have any microphones or cameras — it uses a single-lead electrocardiogram, a photoplethysmogram, and a ballistocardiogram to take your vitals.

Is your lost weight fat or muscle? Breathe out and see

The biggest concern for most GLP-1 users is whether significant weight loss reflects fat loss (good) or muscle loss (bad). Fortunately, there is a way to determine which one is happening, because they are two different processes that generate different biomarkers. Muscle breaks down into amino acids, which end up in urine. When fat is consumed for energy, one of the end results is acetone, which is exhaled in breath. (See this 2015 review for more detailed biochemistry.)

A lot of inexpensive consumer devices that can do this are already available (we found 26 in an Amazon search, ranging in price from $22 – $180). The big challenge is accuracy, and it is this that a new smartphone-linked handheld device sets out to address.

The device’s creators chose a sensor that is as good as mass spectroscopy at picking out acetone from among all the other volatile organic compounds (VOCs) that can fool cheap sensors, such as alcohol, intestinal / microbiome gases, volatile flavors like coffee, etc. Next it addressed breath volume. Deep breaths, which can have a volume of 3 – 5 liters, can dilute the acetone concentration by up to 90%. So the device provides real-time feedback to be sure to collect just the normal tidal volume (about ½ liter). Putting this all together requires careful clinical testing and calibration, which most of the consumer devices cannot even get close to accomplishing.

COMMENTARY: This particular device is only a prototype, but the research in this paper can be used as the benchmark against which to compare other devices.

🎙️ New Episode Alert: Bob Wassman, Chief Medical Officer and Executive Vice President of Clinical Affairs at MARABio, explains how testing for maternal auto-antibodies can drive early autism diagnosis and treatment.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

Do new Alzheimer’s drug results mean pTau217 tests are worth it?

Blood tests for pTau217 are now the diagnostic of choice for detecting early Alzheimer’s disease (see our April review and this July 2026 report). However, until recently, the clinical rationale for early Alzheimer’s diagnosis was limited. Yes, it permitted patients and their caregivers to plan ahead (and researchers to focus clinical trials). But testing really pays off when it enables effective therapies to be matched to the patients who can benefit. And even though we now have two FDA-approved drugs for Alzheimer’s, it wasn’t at all clear that they were truly effective.

The two drugs are definitely good at one thing: eliminating beta-amyloid plaques. However, they only demonstrated minor (largely imperceptible) cognitive benefits. Research showed that Lecanemab (Leqembi®) slowed decline by 26 - 27%, and Donanemab (Kisunla®) by 29 - 35%. That translates to at best just a six-month delay in progression, while 21 - 37% of patients experienced excess fluid in the brain (ARIA) requiring treatment suspension.

But all that data came from clinical trials. In 2023, the LEADER study began tracking Lecanemab’s effectiveness in a more real-world environment (i.e., generally older, with more comorbidities). At July’s Alzheimer’s Association International Conference, the drug’s manufacturer (Biogen/Eisai) reported interim results that are much more promising. After 17 months, 83% of 432 patients benefitted from treatment (75.9% were “stable,” while 6.6% improved), while ARIA occurred in only 12.3%.

COMMENTARY: With such strong results, why do we keep “work” in quotes? We are always cautious with interim manufacturer-reported data. In this case, no placebo group was reported, and diagnostic “stability” is neither a precise nor very demanding term. In this context, it means that a patient didn’t change major categories in the Clinical Dementia Rating – Sum of Boxes diagnostic (CDR-SB). The boundaries of those categories are quite wide, especially once symptoms beyond “mild” are apparent.

More testing before prostate biopsy just makes sense

After the FDA approved prostate specific antigen (PSA) testing for screening in 1994, deaths from prostate cancer declined by 50%. Definitely a great achievement. The problem is that PSA has a very low positive predictive value — only about 13% of PSA-positive (>3ng/ml) men develop clinically significant prostate cancer, so lots of men get unnecessarily referred to biopsy.

MRI helps in the diagnostic journey because it reveals potentially cancerous changes in the size, shape, and structure of the prostate. But just like PSA, on its own it has limited ability to distinguish malignant from benign changes. A recent report shows that traditional digital rectal examination and MRI are about equally effective when it comes to identifying aggressive cases with poor outcomes.

A lot of effort is being and has been directed at improving the prostate diagnostic pathway. A recent report evaluates a multi-component diagnostic pathway that includes PSA, plasma protein biomarkers, genetic risk profile, and MRI, plus clinical presentation. It missed far fewer cancer cases than PSA alone (false negatives reduced from 26% to 10%), although false positives were about the same.

COMMENTARY: Clearly pre-biopsy diagnostics make sense. MRI is part of the solution, so the question is which of the available supplementary tests should go with it. While we have no head-to-head comparisons to guide our choices — Which heritable mutations should we be looking for? Which plasma proteins? — any one of these may be capable of improving outcomes. We hope and expect to hear more about these multi-omic tests.

One at-home test to rule them all?

During the pandemic, we all got used to the idea of at-home testing for COVID. Since the end of the pandemic, we have seen FDA approvals for at-home tests or at-home sampling for infections that range from the flu to HPV, chlamydia, gonorrhea, trichomoniasis, and syphilis.

Are we ready, then, for a single at-home testing unit that can diagnose all sorts of things, from infectious diseases to . . . cancer?

That’s the eventual goal for the lead author of a recent paper describing such a device. Fundamentally, the tool is a fluorescent-light detector. That’s useful because there are several laboratory techniques that use the presence, absence, or strength of fluorescent light to distinguish between positive and negative test results (RT-LAMP, RPA, CRISPR / Cas, and TRAP). According to the paper, this test unit is compatible with all of them.

So how small is this thing? The original version, which could only measure one sample at a time, is roughly the size and shape as an AirPods case (see photo — the white object is an AirPods case, and the device is the black rectangular item in the back). The newer version of the tool, which can measure a test sample and a control sample simultaneously, is a little bigger (it’s the black rectangular item in the front).

COMMENTARY: Cancer testing at home seems like maybe not the best idea — if there was ever a test result that needed to be accompanied by a clinician’s immediate advice, it would be that one. But the idea of a multitasking tool that could diagnose a variety of less consequential ailments and / or monitor chronic disease patients is intriguing.

That said, we also remember the home PCR units that appeared during the COVID era. While those promised more reliable results than the rapid antigen tests we all got used to, the price point for both the testing device itself and the individual test cartridges was much too high for most people at home, and even for many physicians’ offices. If this thing is aiming to be the killer app for at-home diagnostics, both the tool and its consumables need to be affordable for everyone.

Cyclospora

The outbreak of the month is an unusually unpleasant one. We won’t go into the symptoms, as we’re sure you’ve heard more than enough about that already. So what do you need to know about cyclospora from a diagnostic perspective?

The big deal about cyclospora, diagnostically speaking, is that it’s hard to test for it, both in people and on produce.

Testing in people

The parasite doesn’t take up stain well, so it’s hard to see under a microscope. It isn’t included on standard GI-disease test panels, so clinicians have to order PCR tests — which, as we all know, can take a while.

Testing produce

It takes very little cyclospora to make people sick. It also can’t be cultured in a lab. The combination of these two factors means that labs can’t just swab a head of lettuce to see if it’s contaminated — if you don’t swab the right spot, you could get a false negative.

Instead, labs have to wash produce carefully to try to dislodge the parasite (which itself isn’t easy, because the little buggers are tenacious), separate out the debris from the liquid, and then run PCR on the debris.

Challenging epidemiology

The other difficulty with cyclospora is that after you eat contaminated food, it can take up to 10 days before symptoms appear. Good luck remembering what you ate that long ago.

As we write, the FDA is investigating four different cyclospora outbreaks, including the enormous one potentially linked to iceberg lettuce from Taylor Farms de Mexico. Our advice? If you have the option, buy produce grown near you. According to Your Local Epidemiologist, “there has never been a documented cyclosporiasis outbreak linked to a local farm.”

Ebola Bundibugyo

On July 2, the WHO granted its first emergency use listing (EUL) for a molecular test that detects the Bundibugyo species of Ebola — the one causing the current large outbreak in the Democratic Republic of Congo. On July 22, the FDA granted the first emergency use authorization (EUA) of a test for that purpose.

The two tests are made by different companies — the WHO-approved test comes from a Chinese company, while the FDA-approved test comes from a firm in the US.

As of July 28, the DRC has reported 2,423 confirmed cases and 967 deaths, for a case fatality ratio of roughly 40%.

Measles

The total number of measles cases in the US this year has topped 2025’s record-breaking total, and it’s only July. Due mostly to the ongoing outbreak in Utah and the now-ended one in South Carolina, we’re now at 2,318 official cases, the most in 35 years.

The current larger outbreaks in the US are shown below: Arizona / Utah, Pennsylvania, Virginia, Washington.

A better diagnostic target for Alzheimer’s?

The leading tests and treatments for Alzheimer’s disease all target the same things: toxic beta amyloid (Aβ) plaques and tau neurofibrillary tangles in the brain. The tests, whether liquid biopsies (e.g., those that test for p-tau217) or the gold-standard PET imaging studies, look for evidence that those plaques and tangles are either present or on the way. The drugs (Lecanemab/Leqembi^® and Donanemab/Kisunla^®) have both proven highly effective at removing Aβ plaques.

But the drugs’ impact on cognitive decline isn’t all that substantial. Clinical trials found evidence of about a 25% reduction in the rate of decline, but the decline itself continued. Perhaps these drugs will prove more effective when administered earlier in disease progression and / or for longer than the duration of the drugs’ clinical trials, but for now, the jury is out. Meanwhile, many die with, but not because of Aβ plaques.

What else could be going on? A recent Alzheimer’s & Dementia paper suggests that a set of smaller, stable aggregates are what actually drive cognitive decline and dementia, by disrupting essential connections between neurons. These clumps are made of a type of amyloid that never forms larger fibrils or plaques – hence their name, nonfibrillar aggregates of Aβ (NFA-Aβ).

COMMENTARY: The Alzheimer’s diagnostic challenge is a tough one. Plaques and tangles are clearly involved, but there are serious questions about whether they are either necessary or sufficient to cause dementia. Synapse dysfunction has long been a potential alternative to the dominant amyloid hypothesis. It has also been offered as a possible additional factor in the disease, with several studies claiming better correlation with dementia than plaques alone (see this 2020 review from Ageing Research for a full description).

The NFA-Aβ study was done in mice, so it is pre-clinical. But if its hypothesis stands the test of further research, neither drug development nor diagnostic design would prove challenging. That said, there is a long history of unsuccessful Aβ-targeting drugs, some of which differentially targeted smaller oligomers of Aβ but still failed (e.g., Aducanumab / Aduhelm^®, Crenezumab, and Gantenerumab). Is NFA-Aβ the right target after all? It’s too early to tell.

A new way to diagnose both acute and chronic Lyme?

I think we can all agree that ticks suck. Literally and figuratively.

Among the many nasty pathogens they carry, one of the most frustrating is Borrelia burgdorferi, which causes Lyme disease. An initial infection is bad enough, but generally treatable with antibiotics. What’s even more problematic is the small number of cases that progress to a syndrome of ongoing body aches, fatigue, and difficulty concentrating. (This chronic trifecta isn’t unique to Lyme, which makes it even more problematic, as cases that aren’t and maybe never were Lyme are often attributed to the disease. For the best unpacking of that scenario and how it should and shouldn’t be addressed, see this op-ed by Jake Scott.)

Research published in Clinical Microbiology offers a new way not only to diagnose acute Lyme earlier than current tests, but to tell whether a case of post-infection syndrome was indeed caused by Borrelia infection. It suggests that antibodies to some of the fatty chemicals that make up the bacterium’s membrane (phospholipids) may be the biomarkers we need.

The study found that “three antiphospholipid antibodies were elevated during acute infection, with two (antiphosphatidic acid and antiphosphatidylserine) significantly elevated” even before current tests would have caught them. Some of the patients diagnosed with what the paper calls post-treatment Lyme disease (PTLD) continued to have elevated levels of antiphosphatidylserine, which wasn’t found in “patients with look-alike autoimmune disorders.”

COMMENTARY: Borrelia isn’t good at making its own phospholipids, so instead it scavenges them from its hosts and cobbles together a membrane. This and the fact that elevated levels of antibodies against phospholipids have been seen in other autoimmune diseases (multiple sclerosis, for example), makes it reasonable that antibodies against these fundamental cell building blocks might cause PTLD.

Unfortunately, the results in this paper aren’t exactly a slam-dunk. While antiphosphatidylserine levels in patients with PTLD did tend to be higher than those who didn’t, there was significant overlap between the two (see graphic). Still, we’ll take any steps in the right direction that we can get.

Proteomics for rare-disease diagnosis

Researchers are examining whether the proteome — the proteins encoded by DNA — might provide diagnoses for rare diseases. Research published in Genome Medicine showed that a mass-spectrometry-based method of proteome analysis was able to identify more than 8,000 proteins in certain blood cells, corresponding to more than half of the genes in those cells that are known to cause disease. It also allowed researchers to spot disease-causing genes that hadn’t been identified before. When used to diagnose mitochondrial disease, it did a better job than clinical respiratory chain enzymology, which is the standard type of test that, along with genomics, is used to diagnose those diseases.

To get people to keep screening, make it easy to get screened

Back in 2025, we reported on a study which showed that mailing a stool-based test kit (FIT) to folks who were due for colon-cancer screening increased screening rates dramatically — from 30% to 80%. But what happened after that? Do folks who get a negative result on a FIT test ever get screened again?

A study published this June looked at that question. It followed 492,812 adults from four large health systems for up to a decade after a negative home FIT test. The answer? About half of patients got screened in more than 75% of subsequent years, about 30% got screened at least once and 16% never did it again.

The biggest predictor — by far — of retesting rates was which of the four health systems provided the patient’s care. Even after researchers adjusted for patient differences, the odds of consistent testing varied more than 20-fold among the four systems studied.

The recipe for success won’t surprise anyone. The organization with the highest adherence used organized population outreach strategies, including continuing to mail FIT kits to patients who were overdue and making kits available for anyone to pick up at primary-care offices, laboratories, pharmacies, and flu clinics.

COMMENTARY: You can manufacture the best test in the world, but if you don’t make it easy for patients to take it, that might not matter. And for those who quibble that FIT isn’t the best test — yes, colonoscopy remains the gold standard. But at the end of the day, we remain committed to the idea that the best test is the one that the patient actually takes.

🎙️ New Episode Alert: Suchi Saria, founder and CEO of Bayesian Health, tells us about her company's sepsis-detection system - and how to successfully bring such a system into the clinic.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

Breast cancer is the second most common cancer in US women, after skin cancer. It’s also the second most common cause of cancer death in women, after lung cancer. Though death rates due to the disease have been declining steadily for decades, cases have been creeping up by about 1% annually for the past several years. The fight continues, as scientists continue to press forward with new innovations and discoveries in risk assessment, early diagnosis, personalized treatment, and monitoring for recurrence.

Risk assessment

Stats models fall short for risk prediction

According to the American Cancer Society, US women as a group have a 1 in 7 chance of getting breast cancer over the course of their lifetimes. Women with family history of the disease are at higher risk. To determine exactly how much higher such a woman’s risk might be, clinicians can use any one of several statistical models.

Unfortunately, according to a recent Cochrane review, none of them are terrific. Two of the four they assessed either overestimate risk (Tyrer‐Cuzick / IBIS) or underestimate it (BRCAPRO). The other two (Gail and BOADICEA) don’t have those problems, but neither of them do a great job of telling whether a woman is likely to get breast cancer or not - an ability called discrimination. BOADICEA performed the best, but even it had only moderate discriminatory accuracy.

To learn more about how one company is using AI to predict breast-cancer risk, listen to this week’s On Target podcast, featuring Connie Lehman of Clairity.

Diagnosis

Shortening the post-mammo wait with AI triage

Mammograms are the cornerstone of breast-cancer diagnosis. But they can’t tell you everything. When mammography results are abnormal, women often have to wait agonizing weeks for the follow-up testing that tells them definitively whether or not they have breast cancer.

AI triage could change that — by shortening the wait for high-risk women and decreasing the anxiety of those whose risk is lower. In a recent study in press at Nature Digital Medicine, researchers were able to do that using an open-source, previously validated AI model. The model assesses a woman’s risk of having breast cancer within one year, based on their mammogram results.

In this real-world study, women whose risk was in the top 10% were offered expedited follow-up care. Those who accepted got their mammograms read by a radiologist the same day. When necessary, they got biopsies as soon as possible, often on the same day. As the paper noted, “Time to screening results, diagnostic evaluation, and biopsy were significantly shorter in expedited patients versus control patients with screen-detected cancers, with reductions of 99.1%, 99.1%, and 87.2%, respectively.”

The quick care was warranted, too: Cancer was detected in 60/1000 women in the high-risk group, compared to 2.3/1000 in those who were not high-risk.

Per ACP guidelines, you may not need a yearly mammogram

Getting a mammogram is a yearly ritual for millions of women. However, according to updated guidelines from the American College of Physicians (ACP), women who have an average risk of breast cancer don’t need to get checked quite as often as that.

For women between ages 50 and 74 who are at average risk, the ACP now recommends mammography only every two years (biennially). Average-risk women who are 40 – 49 should talk with their doctor about whether they should undergo biennial screening, and women older than that no longer need screening at all.

So what does “average risk” mean? Here’s the list:

• No personal history of breast cancer

• No personal history of high-risk breast lesions

• No genetic mutations that are known to increase risk

• No family history that puts you at higher risk

• No history of high-dose radiation to the chest at a young age.

COMMENTARY: Will these new guidelines change clinical practice? We have doubts about implementation, at least initially. The annual mammogram is ingrained not only in the medical community but in popular culture, as well. Since many other preventative screens have not been broadly adopted (think low-dose CT for lung cancer), we are reluctant to see mammogram guidelines that take away from this strong “role model” system. We’ll see what happens.

Avoiding biopsy with ultrasound + DOT

Biopsy is the gold standard for definitive breast-cancer biopsy diagnosis. But nobody wants to undergo a biopsy if they don’t have to. For women who find a lump in their breast, ultrasound can help with diagnosis — if it shows that the lump is fluid-filled, that lump is unlikely to be cancer. However, “unlikely to be” isn’t the same as “isn’t,” so these women may end up getting biopsies anyway — only to find out that the lump wasn’t cancer, after all.

Work published in Breast Cancer Research showed that pairing another diagnostic technique with ultrasound can give women better answers. The additional piece of the puzzle is diffuse optical tomography (DOT). DOT uses near-infrared light to measure the oxygen levels around the mass as well as the concentration of hemoglobin in the area. High levels indicate that the mass is very active and a lot of blood is flowing to it, which means it’s more likely to be cancer.

This small study (238 women) showed that ultrasound-guided DOT would let 25% more women avoid biopsy than ultrasound alone — it was able to show that those women’s lumps were much less likely to be cancer. While it wasn’t perfect, the false-negative rate was low, at slightly under 2%.

Monitoring and guiding treatment

New biomarkers for triple-negative breast cancer liquid biopsy

When someone is treated for cancer, the first thing they want to know is whether the treatment is working. For women with triple-negative breast cancer (TNBC), that information is even more critical, as that form of breast cancer is very aggressive. (TNBC cells have neither estrogen receptors, progesterone receptors, nor the HER2 protein, hence the name.) These women need to know whether their cancer is still spreading, and they need to know asap.

Tumor DNA in a patient’s bloodstream is a clear indication of metastasis. But finding that circulating tumor DNA (ctDNA) can be very challenging, because there’s so little of it. Plus, we didn’t have very many biomarkers for triple-negative breast-cancer cells.

We now have four more markers than we used to, though, thanks to work published in Cancer Research Communications. Using all those markers together not only improved the ctDNA detection rate significantly (from 40% to 95% in a mouse model), it also highlighted receptors that future treatments could one day target.

🎙️ New Episode Alert: On Target sits down with Dr. Connie Lehman, Founder and CEO of Clairity, Inc.

In this episode, Dr. Lehman shares how AI-driven insights can improve early detection, personalize screening, and ultimately change the future of breast-cancer care.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

For many serious conditions, early diagnosis is the key to successful treatment. This week we highlight a group of tests that improve the odds of getting that critical answer more quickly.

To know if the pancreatic cancer is truly gone, use ddPCR

As we’ve often said in this newsletter, pancreatic cancer remains difficult to successfully diagnose and, therefore, treat. Part of the problem - and one we haven’t touched on much - is how often it recurs after treatment. The most common type of pancreatic cancer is pancreatic ductal adenocarcinoma (PDAC), and after it’s removed it comes back about 57% of the time. Just as with initial diagnosis, early diagnosis of recurrence is key - for prognosis as well as treatment.

A recent study evaluated two blood tests that monitor for pancreatic-cancer recurrence. Both look for circulating tumor DNA (ctDNA) that includes the KRAS mutation, which is found in about 90% of these cancers. (KRAS is the mutation that’s targeted by the new pancreatic-cancer drug that everyone’s rightly excited about.) One relies on next-generation sequencing (NGS), which is the type of test most commonly used in this situation. The other uses digital droplet PCR (ddPCR), a technique in which the sample is divided into thousands of tiny droplets before analysis.

The ddPCR technique is known as a more sensitive test than NGS, and this small study (106 patients) bore that out. After surgery, ddPCR detected tumor KRAS DNA in 56% of patients, compared with 9% for NGS. After chemo, the results were similar: ddPCR was positive in 60% of patients, versus 5% of those tested with NGS.

Thirteen seconds to know whether a TBI patient will wake up

After a severe brain injury, one dire question must be answered as quickly as possible: Will this person ever wake up? According to a recent study, clinicians may be able to get the answer in 13 seconds flat, using a low-tech eye test.

The test, known as the late light-off response, turns a well-known eye test upside down. Instead of evaluating how a person’s pupils respond to a sudden, bright light (that’s the pupillary light reflex, or PLR), it measures how long it takes for the pupils to respond to sudden lack of light.

Brain-injury patients whose pupils took more than three seconds to respond appropriately to sudden darkness were significantly less likely to wake up over the next seven days, even if their other levels of responsiveness were normal. The test takes about 13 seconds to administer, using the same standard tools used for PLR.

You already own a device that can dx mild cognitive decline

It’s your cell phone. Okay, technically it’s an app, but it’s one that’s designed for use at home, on mobile devices.

The app itself has already been shown to be capable of diagnosing mild cognitive decline. What’s new is a paper demonstrating that the app can make that diagnosis in a shorter period of time (30 weeks) than standard, in-person cognitive tests. That’s because people can test themselves frequently at home, while the standard tests are only administered once or twice a year, at office visits. In addition — and for the same reasons — the app can also track subtle cognitive changes over time.

Diagnosing CSF infection in real time

Too much fluid in or around the brain is a dangerous thing. Many brain injuries and diseases require clinicians to place drains that continuously remove excess cerebrospinal fluid (CSF), but those drains can also be a pathway for infection. ICU nurses send fluid samples for testing as often as is reasonable, but that ends up being only once a day or two - those nurses have a lot to do, and getting the samples is time-consuming.

A paper in Science Translational Medicine provides proof of concept for a device that can test CSF continuously. The tool is about the size of a smartphone and attaches directly to the tubes that drain fluid from the patient. Its sensors measure the fluid flow rate, as well as three biomarkers of infection: glucose, lactate, and pH. Since the device’s measurements correlated well with laboratory results, it could one day be used in ICUs to give clinicians a heads-up about impending infection in real time.

Continuous ultrasound for high-risk pregnancies

Ultrasound is the cornerstone of fetal monitoring, but it only gives clinicians a snapshot of how the fetus is doing at the moment the exam is performed. For high-risk pregnancies, it would be very useful to have more consistent monitoring, to catch abnormal changes before they have a chance to impact fetal development.

A paper in Nature Biotechnology describes a tool that can solve this problem. It’s a flexible, adhesive patch about the size of your palm that uses ultrasound to continuously monitor blood flow in both the umbilical cord and one of the fetus’s major arteries. It can also provide measurements that allow clinicians to track fetal growth. The device works even when the mother and / or fetus are moving — a critical feature, given how active some of those little ones can be.

For now, the authors envision the device being used in hospitalized patients. They are working on a wireless device, which could one day be used in patients at home.

Does this pup have what it takes?

Training a guide dog is wicked expensive. Depending on the type of training they need, it can cost up to $50,000 — and 60% of dogs that start guide training wash out. That translates to a loss of more than $12,000 per dog. So it would be awfully nice to know if a particular pup has the Right Stuff before training even starts.

Genomic testing might be able to make that call. One team of researchers looked at genomic sequences for 1,100 Labrador retrievers that were part of the International Working Dog Registry (IWDR). The researchers correlated the genetic information with the dogs’ scores for 17 traits on the IWDR Behavior Checklist - the ones that are likely to cause a puppy to fail its guide-dog training.

They found that genomic data was a better predictor of a dog’s success for at least 11 of the 17 traits. Not only could that be helpful to decide whether to start a pup in training, it can also be used to determine which dogs should be bred. Over time, that could improve the percentage of puppies that complete the training.

🎙️ New Episode Alert:

Madhuri Hegde, SVP and CSO of Revvity, discusses how her company leverages partnerships and broad expertise in both life sciences and diagnostics to solve patient problems.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

Insights from the European Society of Human Genetics

Steroid side-effect prediction

Oral corticosteroids are one of the most commonly prescribed classes of drugs in all of medicine. But their wide-ranging anti-inflammatory and immune-dampening capabilities come with the potential for significant side effects that include stroke, osteoporosis, and cataracts, especially when the drugs are used long-term. Since clinicians don’t know in advance whether a given patient is part of the 10% of people who will develop these problems, they often start patients on doses that are lower than the ideal for their condition - or they avoid steroids altogether, opting for more-expensive treatments that may be less effective.

Genetic testing could identify those folks in advance, according to research presented at the annual conference of the European Society for Human Genetics (ESHG) earlier this week. It showed that people with certain genes were more likely to develop certain side effects: CYP3A4 for osteoporosis and CTLA4 for stroke and cataract, for example. Using a polygenic risk score improved their prediction ability even further, especially for osteoporosis in younger people.

Comprehensive prenatal genetic testing that’s not invasive

Another abstract presented at the European Society of Human Genetics annual conference described non-invasive fetal sequencing (NIFS), a comprehensive blood-based screening test for fetal anomalies. While noninvasive genetic tests for fetuses already exist, the range of diseases each of them tests for is limited. Only invasive tests - amniocentesis and chorionic villus sampling (CVS) provide a complete picture. But many opt to avoid them.

The new NIFS technique relies on deep sequencing of the fetal exome, using cell-free fetal DNA found in maternal blood samples. Their study found 95 - 99% of the genetic variants that invasive tests discovered in the same patients, and over 97% of the genetic variants that cause clinically important conditions. It was accurate as early as 10 weeks into a pregnancy.

COMMENTARY: This technique relies on lab capabilities that already exist, so it may be much cheaper than invasive tests. Definitely a technology to watch in larger studies.

Reddit chatter + wastewater for the World Cup win

Two weeks ago, we reported on a project led by Georgetown University’s Health Security Operations Center (HSOC) and MedStar Health, a Washington, DC-area hospital chain, to monitor World Cup venues for signs of infectious-disease outbreaks. Looks like the team scored. One of the project’s partners (Project Stethoscope, led by Your Local Epidemiologist’s Katelyn Jetelina) noticed a spike in Reddit chatter about gastrointestinal illness in the Seattle area, and notified the rest of the group. Wastewater testing in the area confirmed a spike in adenovirus, which causes mild, self-limiting GI illness, particularly in kids. Fortunately, it does not seem to have impacted the games.

Rare disease roundup

Diagnosing a rare disease is the ultimate needle-in-the-haystack endeavor. Here are a few new methods that scientists hope will improve the odds of making these elusive discoveries.

A single genomic test that could replace 15 individual tests

Genetic sequencing is the cornerstone of modern rare-disease diagnosis, as about 80% of these diseases have a genetic cause. Typical genetic testing relies on short-read sequencing, which looks at sections of DNA that are typically about 300 to 400 base pairs long. Long-read sequencing is a technique in which thousands of base pairs or more are sequenced all at one time. While the process is more expensive than short-read sequencing on a per-sample basis, research published in the New England Journal of Medicine showed that for diagnosing rare diseases, it’s also more effective, yielding about 3% more diagnoses (the reads for the test assessed in the paper were 20,000 base pairs long). And its higher cost may be offset by the fact that one long-read test can replace 15 short-read genomic tests.

Off-the-shelf LLM + clinician = improved diagnostic rate

And then, of course, there’s AI.

In a paper that appeared in NEJM AI last week, researchers found that an off-the-shelf large language model was able to mine genomic and other patient data to diagnose rare diseases that clinicians had been unable to identify. As NBC News reported, “the researchers provided the o3 model with clinicians’ notes about the case, a description of the patient’s symptoms, and a filtered list of certain genes that might be responsible for the patient’s symptoms.” The final diagnoses were then made by humans, and included 18 new diagnoses among 376 cases.

Listen for more

Once you’ve diagnosed a rare disease, the next question looms: Does a treatment for this disease exist? To hear former deputy FDA commissioner Janet Woodcock’s take on how the agency could change its approach to approving these treatments, tune in to this week’s episode of the On Target podcast.

Assessing the flatus each of us passes

Introducing self at a July 4th party: So, how often do you fart?

Just kidding - don’t answer that. You don’t actually know. No really, you don’t - and we know that because people have been testing it, using a new device described in a Biosensors and Bioelectrics: X article as “Smart Underwear.”

The title is kind of a misnomer, because this thing isn’t a pair of underwear you put on. It’s a little device just a smidge larger than a stack of five quarters - we checked - that snaps onto the outside of whatever underwear you want to wear it with.

The paper notes that all but one of the 19 study participants reported “no significant discomfort during routine daily activities,” and on average, participants wore the device for over 11 hours a day. You’ll also be happy to hear that the study authors decided to use silver oxide batteries “housed in custom-designed brass retainers” as a power source. (Standard lithium-ion batteries were deemed to be too much of a fire hazard!)

Why does the world need a device like this?

First of all, because you can’t know what’s abnormal unless you know what’s normal. If you’re farting (or in technical terms, “passing flatus”) 150 times a day, is that bad? If 20 is the average number for a healthy person, sure. But not if the real average is 200. (Studies to determine the answer to this question are ongoing.)

And that brings us to the second reason: People are really bad at estimating how often they fart. Until now, the only devices that could measure flatus frequency required rectal tubes. Good luck getting study participants to wear those for 11 hours a day.

The third reason is that flatus is a potential diagnostic sample. Yes! The chemical makeup of a person’s flatus depends on the makeup of their gut microbiome, which in turn changes based on what they eat. The device was able to detect changes in the hydrogen level of flatus with about 95% sensitivity. Now that’s something to celebrate, maybe with some little, um, explosions . . . or even some fireworks!

Influenza

An influenza outbreak has sickened nearly 300 people at the US Air Force’s boot camp at Lackland Air Force Base. This April, Defense Secretary Pete Hegseth made flu shots optional for incoming military recruits - and according to the Associated Press, only about 40% of recruits who arrived at the Lackland base after the change chose to get vaccinated. On Wednesday, the Defense Department announced that the flu vaccine will once again be mandatory for all recruits.

Measles

The Utah measles outbreak has now been going on for more than a year - according to Medpage Today, the first case was reported on June 20, 2025.

The current larger outbreaks in the US are shown below: Arizona / Utah, Pennsylvania (one outbreak in Lancaster county and one in Lebanon county), Virginia, Washington.

🎙️ New Episode Alert:

Janet Woodcock, distinguished FDA veteran and past deputy commissioner of the agency, gives us her take on AI, companion diagnostics, drugs for rare diseases, and how to earn trust for new technology.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

Note: We will be off next week for the Fourth of July. We hope that you have a safe, relaxing, and enjoyable holiday weekend. We will be back on July 9th.

Could mRNA profiles tell you how much time you have left?

In the recent bestseller The Measure, by Nikki Erlick, every adult on Earth receives a package containing a string, the length of which reveals exactly how long they have left to live. According to a recent Nature investigation, mRNA (aka transcription) profiles are the equivalent of those strings. Whoa!

Well, maybe. The study actually claims that those profiles can be used to detect underlying inflammatory, metabolic, immunological, and epigenetic factors that combine to drive biologic age and mortality.

Mice were used to develop the list of the most important transcriptomic profiles. The authors then assessed the same factors in rats, macaques, and humans. They found that important patterns are common across all four mammals studied, in proportion to each species’ lifespan – mice live about two years, versus humans’ 80 years.

COMMENTARY: We are naturally skeptical of the large and growing “longevity” industry, because it spans the full gamut from the scientifically likely to full-on wishful thinking. This paper falls into the first category. It is complex and nuanced, and it adds to the growing preponderance of evidence linking mortality and disease to the specific biological pathways involved and the clinically measurable biomarkers that reflect them. That includes most of the usual systemic suspects: inflammation, metabolic disorder, DNA damage / mutation, and epigenetic modifications.

We are still some distance from a practical pan-morbidity, pan-mortality, all-in-one diagnostic test. But several different methodologies promise to lead us there - with the potential to reveal which, when, and for whom behavioral and therapeutic interventions would be best leveraged. And that, in the end, is really the point of these “biological age” tests, as a recent JMIR review notes. Knowing that your body or one of your organs is “older” than it should be isn’t all that helpful, unless it is possible and you are willing to do something about it.

Could a liquid biopsy detect lung cancer “5 years early”?

According to research in a recent Cell paper, an increase in the levels of 14 plasma proteins can be used to detect the most prevalent kind of lung cancer five years earlier than it would otherwise have been found. The cancer type is non-small-cell adenocarcinoma; the research also appeared in the New York Times and was reviewed in detail by Eric Topol’s Ground Truths.

The proteins are not produced by the cancer cells themselves — they are stress signals from normal cells surrounding the diseased tissue. An inflammatory cascade begins when immune cells (macrophages) spot a problem. That causes them to release IL-1β (a very, very pro-inflammatory cytokine), which is what causes these proteins to increase in abundance. Because of their source, the proteins in this panel also reveal other forms of lung damage, including COPD, pulmonary fibrosis, and airborne irritants.

Even more excitingly, an FDA-approved drug that can block IL-1β already exists: canakinumab/Ilaris®. However, it only stops cancer progression if it is used before metastasis (i.e., before stage IV).

COMMENTARY: The paper is not itself a clinical trial. Its conclusions are based on an investigation into an incidental finding that showed up in a previous clinical trial of canakinumab’s use in cardiovascular disease. That trial, published in 2017, found fewer cases of lung cancer in patients who received the drug. This new trial convincingly demonstrates that levels of the 14 proteins in question rise as surrounding tissues mount a response to early insults, including lung cancer (at least some forms, including EGFR-driven adenocarcinoma).

A small quibble. The data in this paper comes from the UK Biobank, which means it was collected between 2006 and 2010, before lung-cancer screening with low-dose CT was standard of care for high-risk patients. That leaves open the question of whether modern screening techniques would erode the “five years early” claim.

However, there is no question that this panel could (should?) become an essential part of lung-cancer workup. For every 55 individuals who are flagged as high risk based on this test, IL-1β drug treatment will prevent one lung-cancer case. That would be a huge win.

That said, IL-1β activation is an absolutely essential part of the innate immunity that keeps us alive, so blocking it requires a careful risk/benefit analysis. Beyond the very impressive conclusions of this study, this research is also an example of AI’s ability to identify obscure/hidden patterns in massively complex data sets. That superpower is going to be a huge driver of diagnostic progress.

Inexpensive, easy to use, 3D ultrasound

Ultrasound is such a nifty imaging method. It’s silent, it doesn’t hurt, and it doesn’t expose you to potentially damaging radiation. But it has its own downsides. It’s highly operator-dependent (you really have to know where to point that probe), it doesn’t do well when air is in the way, and bone blocks it altogether. And — as is the case with pretty much all imaging to some degree or another — in order to interpret it, you really have to know what you’re looking at.

A new “augmented reality” system described in Communications Engineering could help solve a couple of these problems. Instead of creating a two-dimensional grayscale image, the system creates one in three dimensions, visible via a virtual-reality headset. The image appears there as a cloud of colored points, superimposed over the anatomy that would be expected to appear in that location.

Three-D ultrasound visualization systems already do exist, but they’re expensive and require a lot of computing power. This system manages to do more with less - it’s less laggy and more spatially accurate than current systems, and is also cheaper. Plus, the novice users in this small study found its images much easier to understand.

The researchers expect that their tool could be used to teach students how to perform ultrasound scans and to help experienced ultrasonographers place biopsy needles. It could also allow clinicians to see the heart wall move in 3D, in real time.

The FDA has given 510(k) clearance to an updated version of a commercially available pulse oximeter - one that is much more accurate across a wide range of skin tones. The company published a verification study of the device in January, showing that it was more accurate than the agency’s draft guidance required.

Cochrane switch on prostate cancer screening not as big as it looks

The UK’s Cochrane Collaboration distills practical clinical implications from a sea of often conflicting research reports on medical practices. They tend to do a good job of it, and are viewed as the world’s most objective body performing these meta-analyses. This month Cochrane is in the news because it changed its assessment of one of diagnostic’s most debated topics: prostate-cancer screening.

Here’s the timeline:

2006: Cochrane finds “insufficient evidence to either support or refute” the use of any prostate-cancer screening.

2010: Additional evidence leads them to conclude that “screening did not significantly reduce prostate cancer-specific mortality.”

2013: Still “no significant reduction in prostate cancer‐specific and overall mortality.”

2026: After reviewing evidence added over the last 13 years, Cochrane appears to reverse their stance, stating that “Screening likely reduces prostate cancer‐specific mortality and may reduce overall mortality.”

That sentence has spawned a lot of headlines over the past week (Nature, STAT News, the Guardian, etc.). But Cochrane’s stance hasn’t actually made a 180-degree change - it’s really more like five degrees. There is now, per Cochrane’s assessment, enough evidence to say that prostate-cancer screening eventually reduces mortality, but only by a very little bit.

Cochrane’s evidence comes from 789,000 men enrolled across six studies. According to the new review, 16 out of every 1,000 men are estimated to die of prostate cancer without screening, versus 15 of those screened (for the stats nerds: the 95% confidence interval ranged from one fewer deaths to two fewer). And the evidence for that assessment is, they said, “low certainty.”

COMMENTARY: This is hardly compelling new evidence for the benefits of widespread screening: One to three deaths in an age cohort of which 491 per 1,000 died for any reason. The benefits of screening only show up if trials last a long time (23 years for some of those included here), but that means keeping clinical protocols constant, even if best practice has moved on.

All but one of the six trials considered were strictly PSA-only, with a range of thresholds from 3-10 ng/ml. The Finnish ProScreen trial is closer to modern clinical practice (a three-step diagnostic process: if PSA >3ng/ml, then a kallikrein protein panel, then to MRI if positive). But it’s still in process. Cochrane included ProScreen’s early and preliminary results, which showed “little to no effect” on prostate-cancer diagnosis, whether the cancer was localized or advanced. But its results having to do with mortality are not yet known.

A urine test for autism on the far horizon?

Finding reliable physiological tests for psychiatric disorders has been an ongoing study in frustration. Autism spectrum disorder (ASD) in younger children is a case in point. Diagnoses have been rising – from 1 in 10,000 in 1980 to 323 in 10,000 in 2022 – but except for the 10% of cases with genetic origins, there is no consensus about what causes it.

The gut-brain axis has come under increasing scrutiny as a possible driver of the condition. At least 40% of children with ASD have chronic digestive problems, and the more severe these problems are, the worse the ASD symptoms.

Hence, the metabolites that gut bacteria shed were the focus of a recent small (about 100 children) study reported in Molecular Psychiatry. Those metabolites eventually end up in the urine, and 600 of them were found to be significantly different in the urine of children with ASD versus those who are unaffected. AI narrowed this number to a panel of the 19 most influential. The total amount of these 19 metabolites was highly correlated with ASD.

COMMENTARY: The gut microbiome is only one of several possible drivers of ASD. However, it’s quite likely that a substantial subset of cases are indeed driven by developmental damage caused by the toxic metabolites that some bacteria shed. If these can be identified early enough (i.e., before behavioral problems become apparent), the potential for successful therapy is promising. A urine test would permit this.

The diagnostic challenge is complex. Harmful and benign bacteria produce similar metabolites, albeit in different proportions. However, some metabolites (e.g., p-Cresol, produced by 60 different bacteria) appear especially harmful.

All that said, this study is quite small, and it needs to be confirmed in much larger samples. Assessing 600 variables in a 100-person sample carries a high risk that the findings reflect only these specific 47 ASD children (aka “overfitting the data”). But if bigger studies bear these results out, this test could potentially help a lot of kids.

Watching the World Cup’s wastewater

Whenever big groups of people from all over the world gather together, public-health folks get a little antsy. When people cross borders, whatever microbes they’re carrying cross with them. And if a lot of people are crossing borders at once, that’s a lot of potential pathogens on the move.

That’s happening now, as thousands of people come to Mexico, Canada, and the US for the World Cup. So a team based in Washington, D.C. is watching for any sign of an outbreak by monitoring wastewater results, hospitalizations, and local health reports in the cities where the games will be played.

Under normal circumstances, one would expect the CDC to lead this team. But the current administration has drastically cut staff and funding to that agency. So this effort is a joint project of Georgetown University’s Health Security Operations Center (HSOC) and MedStar Health, a hospital chain in the DC area.

While a lot of attention has been paid to the possibility that the games could bring Ebola to the US, HSOC officials are much less worried about that. Their number-one concern is measles, along with tropical diseases such as dengue and respiratory diseases like the flu. Wastewater surveillance can detect all of those and more, quite cost-effectively, and it provides evidence of outbreaks earlier than hospitalization data. (Unfortunately, under the administration’s current budget plan, funding for the US’s National Wastewater Surveillance System will go from about $125 million a year to $25 million, so we won’t have as much of this information after this coming September.)

The project is sending daily status reports to FIFA, the organization that runs the World Cup, as well as to government officials and hospitals. The rest of us can keep tabs on the situation via an online dashboard, which features analysis by our favorite science communicator, Your Local Epidemiologist.

Gynecological disease has a fraught history when it comes to diagnosis (and treatment). Social stigma, squeamishness, and lack of respect for women bear most of the responsibility for the problem, and unfortunately all of those factors are very much in play today. Add to that the fact that some of these diseases - notably ovarian cancer - are just flat-out tough to diagnose until they’re very far along, and you have a track record that’s not the most successful.

In this issue we offer a roundup of some of the latest analyses, ideas, and innovations that, we hope, will make the world a better place to be a person who has ovaries, a uterus, and a cervix.

Cervical cancer

Cervical-cancer screening increases when self-collection offered

In a JAMA Network Open study we covered in February, only 20% of women eligible for cervical-cancer screening said they would prefer self-collection of test samples. However, what we say and what we do are sometimes different. A recent NEJM Catalyst study showed that, in one health system, 2.2% more women completed screening once self-collection was an option. (That 2.2% increase was statistically significant.)

Ovarian cancer

Inexpensively “Macgyvering” an ovarian-cancer diagnosis

Ovarian cancer is tough to diagnose. One of the ways you can do it is by measuring the amount of a cancer-associated gene in a sample, a technique that typically requires a spectrophotometer. That’s an expensive tool, and it’s hard to come by in low-resource settings.

A proof-of-concept study published in Oncoscience demonstrated that you can make the same measurement using a low-cost, portable device made of an LED light, an inexpensive color sensor, a sample tube, two rubber bands, and an eraser. (Seriously - look at the image from the study.)

To learn more about how one company is transforming ovarian-cancer diagnosis, listen to this week’s On Target podcast, featuring Dawn Mattoon of Mercy Bioanalytics.

IDing those who’ll respond to PARP inhibition for ovarian cancer

One treatment for ovarian cancer works by blocking an enzyme called PARP, which the cancer cells use to repair their DNA. Right now, PARP inhibition treatment is used in women whose cancer has a biomarker called HRD. But not all of those women respond to the treatment, and some without the biomarker do well with it. So something else is going on.

Research published in Nature Communications showed that the people who actually do better on PARP inhibition treatment have a combination of traits involving five different biomarkers that are part of the immune system. (For those who want to get into the weeds, the biomarkers were CCR4, which sends certain T cells to attack the cancer, as well as IL-6 and IL-8, which increase inflammation, and CCL17 and CCL22, which help the tumor hide from the immune system. The folks who did better had higher levels of CCR4 and lower levels of everything else.)

When the researchers created an algorithm called CUP-CC that took into account levels of all the relevant biomarkers, it was much better at predicting who’d respond well to PARP inhibition than HRD was.

In addition to being more predictive, the new test is also less invasive - it’s a blood test, while the HRD test requires a tissue sample.

Endometriosis

Lighting up inflamed uterine tissue

Discovering that you have endometriosis usually requires a diagnostic odyssey of 5 - 10 years. Fortunately, efforts to decrease that time through the use of less invasive diagnostics than the traditional surgery are ongoing. (Note: These days, endometriosis is more typically diagnosed based on symptoms and history, but surgery remains the gold standard.)

One interesting technique described in The Lancet Obstetrics, Gynaecology, & Women’s Health may allow the disease to be diagnosed using imaging. Researchers investigated whether ^99mTc-maraciclatide, a radiotracer that binds to tissues that have inflammation, could spotlight the abnormal uterine tissue involved in endometriosis. Most of the time, it could. Though the study “was not powered for definitive diagnostic accuracy,” the test’s sensitivity looked to be about 82%, with 100% specificity.

Finding endometriosis as soon as it begins

Endometriosis begins with the onset of menstruation and ends at menopause. So in theory, it could be diagnosed almost as soon as menses begin. That’s the goal of the Simplified Adolescent Factors for Endometriosis (SAFE) score, a questionnaire-based test described in an eClinicalMedicine study.

The score is based on a set of six risk factors for endometriosis: frequency of pelvic pain, seeking treatment for that pain, taking painkillers for that pain, heaviness of menstrual flow, period pain up to age 19, and history of endometriosis in other family members. Each factor the patient reports experiencing adds one point to the score - and doubles the likelihood that they have endometriosis.

The authors hope that this tool could be used with adolescents as soon as they begin having menstrual periods, thereby shortening the amount of time before they get diagnosed.

Obstetrics

Spotting placenta accreta before labor begins

Placenta accreta is a life-threatening condition in which the placenta attaches too deeply to the uterine wall. It’s dangerous because when the placenta detaches or is removed after birth, potentially lethal bleeding can result.

In the 1960s, placenta accreta was quite rare, at about one in 30,000 pregnancies. Today it happens in a bit over one in 500. That’s because the condition is much more likely in women who’ve had a caesarian section, and those have become more and more common. But only 30% of cases are diagnosed before the woman gives birth.

That could change with the use of AI-augmented ultrasound, according to research presented at the Society for Maternal-Fetal Medicine (SMFM) 2026 Pregnancy Meeting™. When an AI model was trained using recordings of prenatal ultrasounds, it was able to tell whether other ultrasound recordings showed evidence of placenta accreta with 88% accuracy (100% sensitivity, 75% specificity, and no false negatives). Caveats include the fact that all of these images came from the same medical system, so the model will need to be tested against images from other institutions.

🎙️ New Episode Alert: Dawn Mattoon, CEO of Mercy Bioanalytics, explains how her company is using extracellular vesicles to successfully diagnose early stage cancer.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

Liquid biopsy identifies microscopic environments around tumors

Liquid biopsies got that nickname thanks to their ability to tell us what kind of cancer a person has based on material in their blood. Typically they do that based on some biomarker that the cancer drops into the bloodstream - for example, a strand of circulating tumor DNA (ctDNA) with a specific methylation pattern that indicates both its tissue of origin and the presence of cancer. (That’s how most multi-cancer early detection (MCED) tests work.) Positive results of these “biopsies” give you knowledge of the type of cancer cell you’re dealing with.

Research published in a recent Nature article describes a liquid biopsy that provides completely different information about a tumor. Instead of focusing on what type of cell makes up the cancer, it shows what the tissue in tiny microenvironments in and around the tumor look like. (Each microenvironment is roughly the width of a human hair.) That matters because, the researchers discovered, those tumor-surrounding microenvironments (TME, aka spatial ecotypes) influence cancer aggressiveness and vulnerability to immunotherapy treatment.

So how can you get information about a three-dimensional environment just from stuff that’s floating in the bloodstream? The researchers used traditional tumor biopsies to identify nine distinct TMEs that show up in all different kinds of cancer. Each TME has its own pattern of DNA methylation, so if you see that pattern in cell-free DNA in the blood, you know that particular TME is present. (As you might imagine, it took a bit of AI to find these patterns.)

An exploratory analysis of 100 melanoma patients showed that these TMEs were useful in predicting anti-PD-1 and anti-CTLA-4 immunotherapy treatment success.

COMMENTARY: The diagnostic power of liquid biopsy continues to expand and astound. Immunohistochemistry doesn’t consistently predict immunotherapy treatment success, so adding these tumor microenvironment results has the potential to be a huge advance if it becomes more broadly applied.

Biomarker patterns in blood have utility far beyond cancer. For a discussion of how they can be used to monitor treatment of multiple sclerosis, listen to this week’s On Target podcast, featuring Doug Biehn of Octave Bioscience.

Dx tests don’t seem to reduce mortality - why?

A provocative and insightful two-part perspective from the Diagnostic Detective points out a strange and unsettling fact about diagnostic tests. These tests (and we’re talking about “tests in patients with suspected disease,” not screening tests) usually fail to demonstrate a mortality benefit when they’re evaluated with a randomized clinical trial (RCT). It’s true. Even very accurate tests that would otherwise appear to be life-saving don’t seem to actually save lives. How is this possible?

According to the author’s analysis, the problem isn’t the tests themselves - it’s how we’re evaluating them. In an RCT, the question we ask is this: If we test all the people who might have this disease, does using this test change how we manage these folks in a way that makes them live longer? The author argues that this question is too broad. He explains that within the group of people who might have this disease, there are really three different groups:

• People who most likely don’t have the disease (shown in red below). A positive test isn’t going to change the fact that these people won’t get treated.

• People who most likely do have the disease (shown in green here). A negative test isn’t going to change the fact that these people will get treated.

• People in the gray zone (here shown in yellow) - we’re not sure if they have the disease or not, so the result of the test will determine whether they get treated or not.

The Diagnostic Detective argues that the only people truly affected by the test are the folks in that third (gray / yellow) group. But since the members of the other two groups “dominate the data,” they hide the effect of the test on the people in Group Three. His conclusion: “Diagnostic trials do not fail because tests are unimportant. They fail because most of the patients enrolled were never in a position to benefit.”

COMMENTARY: Mortality reduction is a high bar to set for diagnostic utility. Is a test truly only useful if it saves a life? We would argue that a test can be useful if it changes a life: It may also guide treatment that improves quality of life (aka healthy lifespan) without actually lengthening that life.

One good example: pathogen identification tests, which can reduce the time and cost wasted by using trial and error to find an effective treatment. Other valuable tests confirm the presence and likely prognosis of diseases for which no effective current treatments exist. Even without a treatment, being able to identify and name the disease can be helpful for a patient and their family.

Though these articles explicitly excluded screening tests from the discussion, we could not help thinking of the implications of this analysis for the large UK NHS trial of a multi-cancer early detection test. The hope had been that detecting asymptomatic cancers at earlier and more treatable stages would reduce mortality. While the test was able to diagnose disease earlier, it recently reported its failure to meet a mortality benefit endpoint.

The bar to conclude that an asymptomatic population screening test is worthwhile needs to be high for many reasons. One of the most important is the fact that even the most accurate tests generate many more false than true positives (see our prior perspective on MCEDs’ challenges and potential benefits), every one of which drives expensive and ultimately unnecessary follow-up testing.

Want to know whether you’re going to live to a ripe old age? Maybe get your bile acids and steroid levels checked. A recent study of people who lived to age 100 found that folks of that age had higher levels of certain bile acids and more stable levels of certain steroids than people who didn’t make it to that age.

Ebola Bundibugyo

Last week, we told you that if you live in the US, you almost certainly don’t have to worry about the Andes hantavirus unless you’ve been hanging around folks who were on the cruise ship where the outbreak started. This week, we’re going to give you essentially the same advice about Ebola - unless you or someone you know was recently in the Democratic Republic of Congo (DRC) or in Uganda, where the outbreak is happening, you almost certainly won’t get it.

But in eastern DRC and western Uganda, the situation is very bad. As CIDRAP reported yesterday, “No vaccines or therapeutics are available for the Bundibugyo strain of Ebola, which is causing this outbreak.” This strain of the virus is estimated to kill about 40% of the people it infects. And the combination of regional violence and the US’s slashing of aid to the DRC has made the healthcare and public-health response to the outbreak slow and patchy.

While the WHO has designated the outbreak a public health emergency of international concern, they still don’t know exactly how many people have been infected or where the virus is spreading. The agency estimates that over 600 people have had the disease and 160 of those have died, but as of Tuesday, May 19, only 30 cases had been confirmed. The WHO is rushing tests to eastern DRC and the US is working with a small biotech company to develop a treatment, but as the top WHO official in the DRC warned, it will take some time to even begin to get this outbreak under control.

Measles

The current larger outbreaks in the US are shown below: Arizona / Utah, California (outbreaks in Riverside and Sacramento counties), Colorado (one outbreak in Weld county and one centered in Broomfield county), Florida, Pennsylvania, Texas, Virginia, Washington.

CORRECTION: Last week in our perspective on ldCT screening we stated incorrectly that “one-by-one cancer screening still has superior positive and negative predictive value” versus multi-cancer early detection (MCED) tests. With the exception of colonoscopy, which has a 100% PPV by definition, all the other cancer-screening tests recommended by the US Preventive Services Task Force have a PPV in the single digits. In a 2025 Nature Communications review of all the 111,000 MCED tests ordered in the real world through October 2023, 0.4% resulted positive, of which 49.4% were correct, a far better accuracy than any one-at-time screening test.

However, as we stated before, retiring multiple separate tests in favor of a single MCED will require detailed apples-to-apples data analysis by cancer type and cancer stage, stratified by patients’ known risk status (age, smoking, genetics, etc.).

🎙️ New Episode Alert: Doug Biehn, President and CEO of Octave Bioscience, talks about the difference that a biomarker test can make in the management of multiple sclerosis.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

We will be off next week for Memorial Day and will return on June 4 with a new issue.

Uptake of ldCT screening for lung cancer still low - 10 years on

In spite of a dramatic drop in the number of cigarette smokers (from about 45% of adults in the 1960s to less than 10% today), lung cancer remains the leading cause of cancer deaths in the US, with just a 20.5% five-year survival rate. Back in 2013, the US Preventative Services Task Force (USPSTF) began recommending low-dose CT (ldCT) screening for all adults ages 55 - 80 with at least a 30-pack-year smoking history who had quit less than 15 years ago. In 2021, requirements were broadened somewhat, to those over 50 with a 20-pack-year smoking history.

JAMA Internal Medicine (letter and editorial) reports that uptake is growing, but at 24.5% is still well below screening rates for other cancers (colorectal 67.4%, cervical 75.4%, and breast 80%). Uptake was highest among the Medicare-eligible over-65s (30.6%) and those in the VA system at all ages (39.1%). Not surprisingly, lowest uptake was among those under 65 years and uninsured (6.1%). Even more disheartening - one December 2025 study in Annals of Internal Medicine showed that only about 60% of Medicare patients whose ldCT screening showed an abnormality received follow-up care.

COMMENTARY: Cancer screening options are growing to include multi-cancer early detection (MCED) tests from a one-time blood sample. The hope is that this single liquid biopsy will eventually replace the need for one-at-a-time cancer screening – more types of cancer included, at earlier, more treatable stages; far less invasive, less inconvenient, easier to perform, and much cheaper than the battery of periodic tests currently recommended by USPSTF.

We are not there yet. One-by-one cancer screening still has superior positive and negative predictive value (probabilities that a positive or a negative result is correct).

However, even one-by-one cancer screening generates lots of false positives, and as we discussed back in 2024, ldCT is no exception. While it finds potentially malignant nodules in nearly half of those tested (48% of men, 38% of women), only 11% of these were judged “clinically relevant,” and only a tenth of those were judged “clinically treatable.” That’s just 0.3% of all those screened.

With all of that in mind, it still makes sense to continue a program of recommended screening tests. New data on MCEDs continues to clarify just how accurate they are, for which cancer types and at what stages. Hopefully soon we will know enough to add an annual liquid biopsy to periodic screening. Unfortunately, we are at least a decade away from being confident enough to actually abandon one-by-one cancer screening in favor of a single annual liquid biopsy.

Underserved oncology patients aren’t getting biomarker testing

Testing for biomarkers is now part of the standard of care in oncology. And yet, the majority of oncology patients in underserved communities never learn that such testing is an option, according to a study presented at the 2026 ASCO Gastrointestinal Cancers Symposium.

The data came from focus groups with 310 gastric-cancer patients. Bottom line: Only 9% of these patients’ physicians initiated a biomarker discussion, while 22% of the patients initiated this discussion themselves. A full 69% had no discussion about biomarkers at all.

COMMENTARY: Of course this is a very small study, but unfortunately it is consistent with similar reports of wide gaps in the utilization of biomarkers for other cancer types (e.g., lung cancer). The situation is highly ironic, considering the ever-decreasing cost and invasiveness of biomarker testing.

Why, then, is this happening? The study didn’t ask the question. It was conducted by the GI Cancers Alliance as part of their Test Your Biomarkers program, which aims to educate patients about biomarker testing, so presumably it was meant to highlight that the program is still needed. But that educational campaign began in February 2024, and this study doesn’t tell us whether more patients know about biomarker testing now than they did then. So is the program working? We don’t know that, either.

And, critically, this study doesn’t tell us why clinicians in these communities aren’t offering the testing to begin with. In an interview with MedPage Today, the authors speculated only at a very high level, citing “competing priorities and complex care pathways” as barriers.

Decision-making about testing is now very much a two-party experience, with both physician and patient taking independent but coordinated responsibility. To know how to improve access to biomarker testing, we need to know not just that certain patients still don’t have that access, but why. Can someone please ask that question?

For more discussion of biomarkers in diagnostics, listen to this week’s On Target podcast, featuring Alec Ford of Karius.

Andes hantavirus

Let’s get the important stuff out of the way first: This outbreak is not the beginning of the next pandemic. If you haven’t been in contact with anyone who was on the cruise ship where it started, it’s extremely unlikely that you have gotten or are going to get sick because of it.

How do we know? As Your Local Epidemiologist (YLE) explained it,

“ 1. We’ve successfully stopped Andes outbreaks in the past through standard contact and droplet precautions.

2. This does not appear to be a highly contagious virus between people.

3. Variants don’t appear to be a concern here. . . . . This virus is behaving as scientists expect.”

So now that none of us are freaking out, where exactly are we?

• As of yesterday (May 13), the outbreak stands at 11 cases, nine of which have been confirmed.

• 25 Americans were on the cruise ship where the outbreak began. Eighteen are currently quarantined in US government facilities - 16 in Nebraska, two in Atlanta. Seven went home after the first person on the ship died, but before an outbreak had been declared. About 11 more Americans were exposed to cruise passengers during airplane flights.

• The monitoring period for hantavirus is 42 days. For the passengers in US quarantine facilities, the CDC is counting May 11 as Day 1 - that’s the day they disembarked from the ship.

• In a May 13 briefing, CDC officials said that they are “encouraging” the people in the Nebraska facility to stay where they are for the 42 days, but did not say that they were requiring them to do so. They are also encouraging the seven other passengers and 11 exposed people to remain at home.

• According to YLE, “Hantavirus can turn positive on PCR testing before symptoms arise and before it becomes contagious.” However, because the Andes strain of the virus hadn’t appeared in the US before, no PCR test for it was available in this country. Yesterday, Wired magazine reported that a University of Nebraska team had developed and validated one, and that they expected to test the passengers in Nebraska “soon.” According to Wired, the university “now has the capacity to conduct a few hundred tests on patient samples.”

🎙️ New Episode Alert: Alec Ford, CEO of Karius, shares how genomics and AI are expanding what’s possible in infectious-disease diagnostics — and why getting the right answer faster can make all the difference for immunocompromised patients battling cancer and other serious conditions.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

We heart this MCED test . . . for doggone good reasons

We’ve written a lot about multi-cancer early detection tests (MCEDs) over the past year (17 articles in 2025), but none of those tests - in our humble opinions - is as cool as the one we learned about this week.

The company that produces the breath-based test published phase II trial results in the Journal of Clinical Oncology. The researchers reported 91% sensitivity and specificity overall for seven different types of cancer (head and neck, breast, lung, gynecologic, upper and lower GI, genitourinary), and 91% sensitivity for early stage cancers (stages I and II). These are reasonably good numbers in the MCED world, but that’s not what makes the test cool.

We like it because it involves dogs.

To get tested, people breathe into a cotton face mask for 10 minutes. At least three different trained dogs then sniff the mask and give their assessment (either “this person has cancer” or “no cancer detected”). The company then combines the three dogs’ results using a statistical model that takes into account each dog’s track record as well as the patient’s medical history. They report that the dogs’ responses vary from one another less than 2% of the time.

New cervical cancer screening guidelines include self-sampling

The American College of Obstetricians and Gynecologists (ACOG) has issued new guidelines for cervical cancer screening. For the first time, the organization has formally included patient self-sampling for human papillomavirus (HPV) testing as an option.

Listed below is the summary of their recommendations for average-risk individuals:

• Age 21-29: Pap test every 3 years.

• Age 30-65: Ideally, clinician-administered high-risk HPV (hrHPV) test alone, every 5 years. Other options can be considered if certain criteria are met:

  • Patient-collected HPV test every 3 years. This is only an option if a follow-up system is in place.

  • hrHPV + Pap tests every 5 years, or a Pap alone every 3 years. These are options only if there’s no way to do just an hrHPV test alone, or if the patient still wants one of these options even after discussion with the clinician.

• AGE 65+: If the patient has been adequately screened thus far, stop screening. If not - or if the patient is high-risk - keep screening.

  • Adequate screening = Three negative Pap tests or two negative co-tests (hrHPV + Pap) within 10 years, with the most recent test occurring within 3 years for Pap alone or 5 years if co-testing is used.

Two promising tactics for detecting pancreatic cancer

As we’ve written before, a test for early stage pancreatic cancer is desperately needed. The disease is bad enough when it’s caught early (44% survival rate for localized disease), but only 20% of cases are diagnosed at that point. Roughly half the time, pancreatic cancer is only found once it’s spread to other organs - when the chances of survival are a dismal 3%. (It has been a big week for pancreatic cancer - with the announcement of Phase1/2 trial results showing increased survival for the 90% of patients with RAS mutations.)

Two new studies give cause for hope.

Sample: blood. Tech: nanoparticle recovery and analysis

Pretty much all tumors shed nanoparticles into the blood. Some of these particles, like extracellular vesicles, are actively released. Others end up in the bloodstream when cells die. The nanoparticles carry biomarkers - typically DNA and proteins - that are specific to the kind of cancer they come from.

The problem with nanoparticles is the “nano” part. Most methods of separating particles out of blood are based on the particles’ physical characteristics, especially variations in their size and / or weight. Nanoparticles are so small, it’s hard to separate them out that way from the rest of the plasma in which they ride.

To solve this problem, researchers used tiny electrical fields to drag the nanoparticles onto a microchip. There, the biomarkers they carry can be analyzed for signs of pancreatic cancer.

According to a study they published in the journal Small (yes, that’s the name of the publication), the test had a 97% likelihood of correctly distinguishing people with cancer from those with non-cancerous pancreatic disease. That’s a lot better than the current standard test, which is an ultrasound-guided biopsy that has a 79% chance of catching the cancer.

Sample: CT. Tech: AI

The diagnostic problem is that a cancerous pancreas looks so much like a normal pancreas on CT that it can be literally impossible for a human eye to tell the difference between them. And the changes that a radiologist is supposed to look for - the presence of a mass, a dilation of the pancreatic duct, or evidence that the tumor has invaded blood vessels - don’t show up until the disease has already been present for a long time.

Research published in Gut describes an automated, AI-based solution to both of these problems. The model was able to diagnose 73% of those previously undetectable cancers at a median of about 16 months before diagnosis — nearly double the detection rate of specialists reviewing the same scans without AI assistance. It does that by looking at radiomic features of the CT scan. Those features are something that people can’t see - they’re quantitative information from the diagnostic images themselves. In this study, these radiomic features describe the texture of the tissue itself.

That, the authors note, makes sense, because during its early stages cancer is busy changing the microstructure of the pancreatic tissue. Radiologists can’t see that happening - but the AI can.

Lastly, it has been a big week for pancreatic cancer - with the announcement of Phase1/2 trial results showing increased survival for the 90% of patients with RAS mutations.

COMMENTARY: No, AI did not beat ER docs at diagnosis

Look, folks, we cover AI a lot in this newsletter. We’re excited about the way it can advance diagnostics by doing things that people simply cannot do, notably by finding subtle patterns in ginormous piles of data. But - breathless bloviating by tech bros aside - AI is not better than humans at everything, at least not yet.

Which is why the headlines about a recent study drove us bonkers. Here’s an example, from NPR: “In real-world test, an AI model did better than ER doctors at diagnosing patients.”

Um . . . no, it did not. As ER doc Kristin Panthagani explained in You Can Know Things, its findings were “based on results from only two physicians — who aren’t ER doctors — performing a task that isn’t really what ER doctors do.” (The emphasis is hers.)

Here’s what the study actually did: It gave real-world hospital records to two large language models (LLMs) and two internal-medicine specialists and asked which of them were more likely to come up with the correct final diagnosis for emergency-room patients. (If you want all the hairy details but don’t want to comb through the whole study, read Panthagani’s post.)

There are two big problems with this approach.

Number one: The LLMs weren’t compared to ER doctors. They were compared to internal-medicine specialists, who do not have the same expertise.

Emergency-medicine clinicians are trained to come up with a list of possible diagnoses based only on the following information:

• Step 1: What they learn when a patient walks in the door (triage)

• Step 2: What they learn from examining the patient and running initial diagnostics in the ER.

Internal-med specialists, by contrast, are used to working with a lot more data. They’re trained to ferret out a final diagnosis based not only on all the information that’s already been provided by the ER team, but also on follow-up examinations and testing. So asking them to diagnose based only on the few data points available in the ER makes no sense. And it makes all the headlines incorrect from the get-go. (Also of note: When the internal-med docs made diagnoses based on more complete medical records, they did just as well as the LLMs.)

Number two: ER docs’ primary job isn’t coming up with a final diagnosis - it’s making sure patients don’t die. So if you’re testing an LLM’s ability to function as an ER doc by asking it to come up with a final diagnosis, you’re asking the wrong question. The correct question would have been, “What could these patients have had that might have killed them?”

Panthagani brings up other problems, notably that the LLMs used information that only existed in the hospital records because real-life ER docs knew what to look for. But the two issues described here are enough for us. (Okay, the fact that the findings are based on the success rate of only two physicians is a problem, too.)

Honestly, the whole thing is kind of baffling. Scientific studies are supposed to very precisely compare apples to apples. Could AI do a better job - or at least as good a job - as ER doctors? Maybe. But this study doesn’t answer that question, and folks who write about medicine for a living should have seen that.

Stroke diagnosis goes airborne in Australia

Strokes are acute injuries to localized areas of the central nervous system (CNS) that happen because those areas don’t have enough blood in their blood vessels. One in four people will suffer a stroke in their lifetime, resulting in over 160,000 deaths per year in addition to widespread disability and suffering.

The problem comes in one of two flavors. Either blood gets out of the blood vessels (hemorrhagic stroke), or blood stops flowing into the vessels because they’re blocked (ischemic stroke). Quick diagnosis is urgent because “time is brain.” Without the oxygen that blood supplies, brain cells begin to die at the rate of about 2 million per minute.

Roughly 87% of strokes are ischemic. Clot-busting drugs can be very effective in these situations, but only if they’re administered within 4.5 hours of symptom onset. However, if the patient is undergoing a hemorrhagic stroke, these drugs can be fatal. Hence, the standard diagnostic pathway is to treat only after imaging (usually non-contrast CT) tells you which version of stroke you’re dealing with.

And therein lies the problem. CT machines are big and expensive. Because they use X-rays, they need to be placed in their own shielded rooms, so they tend only to live at major hospitals. Transporting patients to these places can quickly eat up critical time.

The Australian Royal Flying Doctor Service (RFDS) is testing a solution to this problem. It’s a helmet that uses AI to analyze radio waves that pass through a patient’s head. Since X-rays aren’t involved, all the bulk and shielding that CT needs isn’t required (paywall).

A 323-case study now in preprint evaluated the device’s accuracy in identifying stroke-mimicking symptoms from true strokes, and of those, which are ischemic versus hemorrhagic. Headline (no pun intended) accuracy was reported as 92% sensitivity / 85% specificity in detecting hemorrhagic strokes, and 95% sensitivity / 80% specificity for ischemic strokes.

COMMENTARY: It is impossible to tell from the current trial data just how effective this device will turn out to be in the field. The study was small, with the 92% / 95% sensitivities based on successfully identifying 12 of 13 hemorrhagic cases and 19 of 20 ischemic cases.

However, non-contrast CT has its weaknesses, too. It is almost perfect at identifying hemorrhagic strokes, but misses many smaller / earlier ischemic strokes that would benefit from rapid treatment. Various reports estimate sensitivity from 75% to as low as 16%.

The appeal of this particular AI / radio-frequency-based brain-scanning approach is that it enables true portability. The device weighs just 32 lb., while other mobile CT platforms range from 154 - 1,800 lb. If it does turn out to work well in the real world, it could be a game-changer.

For the fictionalized TV version of this medical corps, you have to watch The Royal Flying Doctor Service on Amazon Prime.

Micro probe, multi-monitor

Patients in intensive care get their blood drawn frequently, as clinicians monitor metabolic status. These patients are fragile, and their condition can change very quickly. So being able to monitor them without waiting for a lab to analyze a blood sample would be helpful.

New technology described in Nature Communications would allow clinicians to monitor multiple biomarkers simultaneously, in real time. It’s a 1.1mm optical fiber probe that uses spectroscopy to measure tissues’ response to infrared light. In this proof-of-concept study, the probe was able to simultaneously monitor ethanol, glucose, and lactate.

The myriad of ways in which mailing out tests changes screening

Around this time last year, we reported on a study that investigated whether it would help to send patients a fecal immunochemical test (FIT) along with their reminder to get screened for colon cancer. It helped a lot. Screening rates went from 37% to 80%. Eventually, cases went down by about 30%, and deaths were cut in half.

A number of studies we’ve seen before and after that study look at what else might change when clinicians send screening tests to patients.

Colorectal cancer (CRC)

A November 2024 study in JAMA Network Open found that when patients were only told to get screened for CRC during an office visit, about 10% did so. When patients who got the office-visit recommendation also received a FIT kit in the mail (with instructions and postage for returning it), 30% got screened.

A JAMA Internal Medicine study published this week looked more closely at mail-in CRC screening tests. It showed that 28% of patients at community health centers who received a FIT-DNA test in the mail along with the manufacturer’s “wraparound” support got screened, compared with 23% of those who got a FIT test and a phone call. (The FIT-DNA test is an every-three-year test, while FIT must be taken annually.)

Atrial fibrillation (Afib)

This August 2025 JAMA study asked whether mailing a screening test to older, moderate- to high-risk patients and asking them to use it would increase diagnoses of Afib. The test was a continuous electrocardiogram (ECG) monitor in the form of a patch that the patient was asked to wear for 14 days. About 84% of patients who received the patch wore them. In the group that received the patch, about 7% were diagnosed with Afib, compared to about 5% of those who did not.

Human papillomavirus (HPV)

In June 2025, we reported on a JAMA Internal Medicine study showing that 41% of folks who received a phone reminder and a mail-in HPV screening kit got screened, compared to 17% of those who only got the phone call.

Another JAMA Network Open article from October 2025 looked at the cost-effectiveness of mailing HPV tests to patients in need of screening. It found that of patients overdue for screening, 36% of those who received educational materials and a self-sampling kit in the mail got screened, while only 19% of those who just got the educational materials got screened. The mailing that included the kit was also the most cost-effective.

COMMENTARY: To us, getting a screening test in the mail seems like the logical, comfortable next step in a society that 1) is used to having everything from meals to toilet paper delivered to them, and 2) was trained to self-test for disease during the pandemic. While it’s nice to see researchers finding all these different positive ways in which mailed tests change the metrics on screening, we wish that more of them would ask the question that the study done last May took: Does mailing patients a screening test decrease the rates of disease and death (and, we’d add, patients’ quality of life)? That, in the end, is what screening aims to do.

Measles

The big measles epidemic in South Carolina is officially over, after more than 42 days without a new case. One bright spot: measles vaccination in the state has increased more than 30% this year. However, Outbreak Outlook (seasonal paywall) reports that in Utah, Texas, and Florida, evidence is mounting that

“case rates are significant undercounts; there has been reporting for months that a proportion of patients are not seeking care for measles or are declining testing (see here for just one example).”

The Arizona / Utah outbreak, for example, may be more than 6.5 times larger than has been reported. In addition, summer travel is likely to lead to more cases across the nation, as the CDC warned in a recent alert.

The current larger outbreaks in the US are shown below: Arizona / Utah, California (outbreaks in Riverside and Sacramento counties), Colorado (one outbreak in Weld county and one centered in Broomfield county), Florida, Texas, Washington.

🎙️ New Episode Alert — On Target sits down with Dr. Irene Ghobrial, Founder of Predicta Biosciences.

Irene Ghobrial, co-founder of Predicta Biosciences, explains how bringing whole-genome sequencing into the clinic is taking diagnosis and treatment of multiple myeloma and blood cancers to a new level of precision.

Note: Mara is a Director of Predicta Biosciences as her firm is an investor in the company.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

Does rAST really have only limited benefit in sepsis?

A recent report and editorial in JAMA (and a commentary in Medpage Today) found that using a rapid antibiotic susceptibility test (rAST) was no better than slower traditional microbiology when it comes to managing patients with gram-negative bacteria in their bloodstreams (bacteremia or advancing to sepsis). It found no significant reduction in mortality, length of stay, or complications. (AST tests identify which antibiotics will likely be most effective to stop growth of the relevant bacteria.)

These patients were seriously ill. All were hospitalized, and 25% of them were already septic when they were admitted. The study found some marginal benefit for the patients who were suffering from antibiotic-resistant infections, which makes sense - those were the ones for whom the benefit of finding the right antibiotic quickly was highest. But there was no way of identifying these challenging cases ahead of time without an AST test.

The mortality rate for these patients is high: 23.4% of those enrolled in this trial died within 30 days of admission, whether traditional microbiology or rAST was used.

COMMENTARY: Just how much faster is “fast”? Before either method is used, bacteria have to be grown out for 10-15 hours, then the clock starts: traditional AST takes 24 - 48 hours, while rAST takes about 6 hours. Inevitably, some time goes by setting up the tests, creating the results report etc., usually adding another 2 - 3 hours to the process. That translates to “fast” being a net 18 - 42 hours quicker. One would think that having a definitive diagnosis in hand that much more quickly would be very valuable to individual patients.

However, the benefit to individual patients is only part of the reason to use rAST. As we discussed back in January, antibiotic stewardship is a big problem, and it’s getting worse because of widespread antibiotic overuse in humans, animals, and plants. The WHO estimates that globally 17.5% of drug / infection combinations are now fully resistant. In the US, that translates to 2.8 million antibiotic-resistant infections and 35,000 deaths each year. By shortening the amount of time it takes to find the right antibiotic, rAST has the potential to reduce the use of ineffective ones. Unfortunately, the stewardship benefit is spread thinly across everyone, while conservative treatment of critically ill individual patients requires clinicians to throw every relevant treatment at the patient as early as possible.

AI may find a lot more potential CVD. That might not be good.

In medicine, when you’re looking for a specific problem but find another abnormal thing that may or may not matter, that unexpected abnormal thing is called an incidental finding. An example: Someone I know recently had an abdominal ultrasound to diagnose an intestinal problem, and the ultrasonographer made the incidental finding that he only has one kidney. (True story.)

One of the incidental findings that often shows up on a CT scan of the lungs is calcium in the coronary arteries (CAC). That’s an abnormal thing that can matter, because the more calcium you have in your coronary arteries, the higher your risk of a heart attack.

Radiologists will note CAC when they see it on a CT, but according to a recent article in STAT News (paywall), they miss it 20 - 40% of the time. However, the article notes, that may be about to change.

In March, the American College of Cardiology recommended that cardiologists should take incidental findings of CAC on CT scans into account when deciding whether to start a patient on statins. And as of the first of this month, Medicare has approved an additional $15 reimbursement for FDA-approved AI imaging review of CTs to look for CAC in some patients.

That’s a lot of CAC we’re about to find, which means a lot of follow-up testing. The total cost of all of those diagnostics could end up being 10 to 100-fold greater than that initial $15.

COMMENTARY: At first glance, the rationale for looking for CAC is compelling. Cardiovascular disease is still the number-one cause of death in the US (for both women and men), and statins are a cheap and adequate preventative for most.

CAC-specific CT studies are very good at predicting future heart-disease risk, but they are expensive, and they’re only ordered later in a disease course as symptoms become apparent. Lung CT is far more frequently performed (19 million per year in the US is widely quoted).

The key questions are whether patient mortality will decline as a result, and at what end-to-end inclusive cost? What makes the answer to these questions as yet uncertain is that as lower-prevalence populations get screened, the false positive rate increases geometrically, generating higher follow-up costs for those that will not benefit.

Age-based cancer screening doesn’t always make sense

Cancer incidence increases as we age, so most current screening guidelines suggest starting around age 50. But as a recent Nature review summarizes, cancer incidence in under-50s has been rising (see chart). Even though cancer in that age group remains much, much rarer than it is in older people, that increase has put pressure on health organizations to adopt a younger and younger screening threshold.

That change comes at a significant cost. In addition to the increased financial burden involved in screening more people (the test itself, time off work, etc.), the wider the screening net is cast, the more reported positives turn out to be false alarms that have caused unnecessary anxiety. But that only becomes evident after follow-up procedures, which tend to be more onerous and / or invasive than the initial screening was.

For example, though 10% of mammograms are initially flagged as positive, just 0.5% of them turn out to be truly positive after follow-up. Since younger age groups have lower disease prevalence, the false positive rate for them is even higher.

Another recent report in JAMA evaluates screening at the other end of the age spectrum, asking, “At what age should colorectal cancer (CRC) screening cease?” It looked at 91,952 veterans who had reached age 75 - 98% had had at least one colonoscopy by that age, and 28% of those had had polyps removed.

Over the course of the next 10 years, death from CRC was extremely rare: 0.5% for those with a history of polyps, 0.4% for those without. Those numbers were far overshadowed by the roughly 48% who died from non-CRC causes.

So is it worth it to continue screening over-75s for CRC? The answer is no, even for those with a prior history of polyps. Those over 75 are 100 times more likely to die of something else.

COMMENTARY: Age is the number-one risk factor for a cancer diagnosis, but it is far from the whole story. Screening for cancer needs to be brought into the personalized-medicine era by incorporating individual, specific risk assessment, so that those at higher risk are screened promptly and more frequently.

An example of this is cervical-cancer screening (see our recent perspective). The younger the age of HPV vaccination, the less screening required: from every 5 years for the never-vaccinated to just twice in a lifetime for those vaccinated before age 15.

Two free AI training sets gave a lot of people what they paid for

AI models are only as good as the data sets used to train them. That is why a recent preprint, summarized in Nature News, was so alarming - but not surprising. It found that two highly dubious data sets had been used as the training data for AIs described in 124 published research papers. Those AIs were cited in 86 review articles, and three of the models were being used in clinical practice.

The data sets in question appear in Kaggle, the self-described “World’s AI Proving Ground.” One is a 5,110 person data set for stroke risk assessment (downloaded 275,000 times), the other a 100,000 person data set for diabetes risk (downloaded 114,000 times).

Red flags the authors found included overconfident assertions of clinical utility, undisclosed sources, absence of expected correlations (e.g., glucose levels, BMI and HbA1c), absence of incomplete data points, and more. Most of the papers based on these two data sets were published in less-prestigious journals, but far from all of them (see chart).

COMMENTARY: Accurate, relevant and up-to-date training data is the “secret sauce” that can unlock the enormous potential of AI in diagnosing disease. That data is immensely valuable, and consequently it is becoming increasingly proprietary and hard to access. Hence the temptation for clinical AI decision-support system developers to download predigested data sets to fine-tune their algorithms.

In principle there is nothing wrong with this, providing that the data set never leaves the developer’s sandbox. In practice though, models based only on bad data are leaking into real-world clinical practice.

Today, all AI tools have to be treated with informed skepticism. They are extraordinarily capable, but they are also prone to unforced errors (overconfidence, hallucinations, etc.). That’s even more of a problem if the training data was not fit for purpose to begin with. While this particular study focused on just two data sets used predominantly in less-developed countries, it more than likely represents just the tip of the iceberg.

Diagnostic steps in the trek towards controlling bladder cancer

Bladder cancer has one of the highest recurrence rates of any cancer - somewhere between 30 and 80% of these cancers will come back. Diagnosis and monitoring typically require cystoscopy - inserting a camera through the urethra, into the bladder. Talk about invasive. So we’ve been happy to see some recent diagnostic developments that could improve the management of this frustrating disease.

cfDNA in urine for diagnosis

A small case-control study suggested that cell-free DNA (cfDNA) in urine could help diagnose cancer that has invaded the bladder’s muscle wall. The study looked at cfDNA in urine samples, investigating both large and small fragments of a range of genes. It found that small fragments of the MYC gene were the most useful, with a specificity of 97% and positive predictive value (the likelihood that a positive test is correct) of 88% for diagnosing that type of bladder cancer. (Sensitivity wasn’t the best, unfortunately, at 70%.)

ctDNA to tell if bladder removal is needed

Once patients have been diagnosed with bladder cancer that’s invaded the muscle wall, patients and clinicians face big decisions. Which of these patients should have their bladders removed, and which can be effectively treated without that life-changing surgery?

Interim results from the RETAIN2 trial suggested that a blood test might help answer that question. The small study followed 71 patients with muscle-invasive bladder cancer who had been treated with a combination of chemotherapy and immunotherapy, but not surgery.

There was some significant good news: The study found that patients who had no circulating tumor DNA (ctDNA) in their blood after treatment were unlikely to have their tumors metastasize over the next two years (97% had no metastases a year after treatment; 83% had none after two years). However, ctDNA wasn’t able to predict whether or not the cancer would come back.

A urine test to see if treatment is needed after tumor removal

Most of the time, bladder cancer is diagnosed before it gets into the muscle wall. Many patients with that lower-grade form of the disease can have their tumors removed using a scope inserted through their urethras. The question is what happens after that.

After surgery, many of these patients undergo immunotherapy treatment called BCG that has significant side effects. But not all of them need it. Scientists have tried to tell who’s who by checking cfDNA in the urine for mutations - mutated DNA could indicate that microscopic bits of tumor remain in the bladder lining. The problem is that everyone has non-cancer-related mutations in the cells that line their bladders, and the older you are, the more mutations you probably have. That’s called the field effect, and it’s true for many other organs, as well, including the breast, lung, and colon.

Using a statistical technique, a group of researchers has been able to filter out the irrelevant mutations. By looking at what was left, they were able not just to identify which patients were likely to have their cancer recur, but also to tell whether patients were more likely to respond to surgery, to immunotherapy, or to neither one.

COMMENTARY: The first two of these three steps in bladder-cancer diagnostics are promising, but relatively small. If the results from the third study hold up, they have the potential to affect diagnostics for many other cancers, which could be exciting. We’ll keep an eye out.

Enzyme IDs patients who need dual immunotherapy

Immunotherapy is one of the most promising methods of cancer treatment. It sounds ideal - sending the body’s own immune system to attack cancerous cells - but the reality is that only 15 - 20% of patients respond well to it, long-term.

So why don’t other patients respond as well to immunotherapy? Are we testing for the wrong things before we treat?

Maybe not. This is a story about a test that tells us even more than we thought it did. It’s also an object lesson into how heckin’ complicated cancer and the immune system really are, and how much we need to know about both of them in order for personalized medicine to work effectively.

A couple of weeks ago, Cell Reports Medicine published a study looking at PHGDH, an enzyme that helps cells produce the amino acid serine. Cancer cells use a lot of serine, and scientists already knew that PHGDH is found at abnormally high levels in roughly half of all colorectal cancers and 40% of breast cancers. We can test for that - and we thought we knew how to treat it. Just block the enzyme’s ability to make serine. (Drugs for that purpose are now in development.)

But the study showed that that’s not all the enzyme does. It found that PHGDH also ramps up production of a protein called PD-L1, which allows cancer cells to evade the immune system. And crucially, PHGDH does that even if its ability to produce serine is eliminated. When the scientists tested versions of the enzyme that were genetically altered so they could no longer produce serine, those dead enzymes still drove PD-L1 levels up.

Fortunately, immunotherapeutic drugs that directly fight PD-L1 already exist. And when the researchers tested a combination of both anti-PHGDH and anti-PD-L1 drugs in mice with colorectal cancer, the pairing worked better than treatment with either of the two drugs alone (50% survival vs. 20%). The test for PHGDH was telling us what to do all along - we just needed to look deeper.

COMMENTARY: So what does this mean for testing and personalized medicine? It’s a reminder that when you’re asking one complex system to attack another complex system, the answers may very well be more complicated than you think. As we continue to push the field of immunotherapy forward, an in-depth understanding of the relationship between test and treatment will be critical.

An all-in-one anti-glaucoma contact lens

Glaucoma is the second leading cause of blindness in the US (globally, it’s number one). In this disease, the pressure inside the eyeball increases until it compresses the optic nerve, eventually leading to nerve loss.

Last month we highlighted a contact lens design that could monitor interocular pressure (IOP) using a wireless system. A recent paper goes a step further. The contact lens it describes can both wirelessly monitor IOP and can automatically release drugs if and when pressure rises above a normal threshold (about 16 mmHg), a process called closed-loop dosing.

COMMENTARY: Unless it comes on very suddenly, the early stages of glaucoma are completely asymptomatic. The current clinical approach is to measure IOP whenever the need for corrective lenses is assessed - usually once a year at most.

But IOP can spike intermittently and then go back to normal. Unless you measure a patient at exactly the right time, you’ll miss that. Meanwhile, those intermittent spikes can still do damage. Recognizing them is the challenge that continuous contact-lens monitoring offers to solve. Adding drug delivery takes things another step forward. But we don’t yet know whether closed-loop dosing provides a substantial clinical advantage over daily eye drops.

Of course, there are other challenges, as well. Both this and last month’s contact-lenses have only been tested in animals, and when it comes to humans, acceptability of a more-bulky-than-normal contact lens will be an issue.

Optical genome mapping guides acute leukemia treatment

When someone is diagnosed with acute leukemia, clinicians run several genetic tests to figure out the prognosis and best course of treatment. But even as a group, those tests don’t tell you everything you need to know.

Recent research compared a technique called optical genome mapping (OGM) to the standard suite of genetic tests used in acute leukemia patients. In 18% of the cases in the study, OGM was able to find genetic variants that the other tests had missed.

COMMENTARY: In OGM, fragments of DNA are labeled with fluorescent tags and loaded into ultra-small nanochannels on silicon chips. The nanochannels arrange the DNA into straight lines - hundreds of thousands of them on a single chip. Scientists then take an image of the fluorescent tags on the chip and digitize it, creating a kind of bar code (see graphical explanation shown here). The bar codes from one individual can then be compared with bar codes based on a composite of normal DNA, showing where structures in the test subject’s DNA have changed.

Compared with other types of genetic testing, OGM is really good at spotting both very small and very large genetic variations. But just like the other types of tests, it can’t do everything - and it takes longer than some other tests. As the paper’s authors note, for acute leukemia, it may be best used as a routine addition to the standard battery of tests.

Note: To learn more about optical genome mapping, check out our latest On Target podcast, featuring Bionano CMO Alka Chaubey.

Alka Chaubey, CMO of Bionano Genomics, explains how optical genome mapping addresses the unmet needs of cytogenomics - and what it takes to get pathbreaking new technology into the clinic.

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com

To diagnose Alzheimer’s, pTau217 alone isn’t enough

Once upon a time, the “amyloid hypothesis” ruled the world of Alzheimer’s disease. The hypothesis proposes that cognitive impairment begins with the formation of beta amyloid (Aβ) plaques in the brain, which then causes tangles of Tau protein to form inside neurons. It is those later tangles that kill the cells.

As research into the disease has progressed, the hypothesis has become more controversial, for several reasons. Number one: Many with Aβ plaques never suffer cognitive decline. Number two: Cognitive decline can be caused by a diverse set of underlying problems, not related to AD.

The blood biomarker pTau217 has been shown to accurately reflect the presence of Aβ plaques on PET imaging, and thus has become the biomarker of the moment. However, recent research suggests that we probably shouldn’t be using that biomarker as the sole basis for an Alzheimer’s diagnosis.

Exhibit A: A new publication in The Lancet Neurology demonstrates that reliance on pTau217 alone leads to overdiagnosis of Alzheimer’s even among advanced cases, with a 43% false positive rate. However, if a follow up plasma eMTBR-tau243 test which indicates advanced Tau tangles is also positive, only 16% are false positives. This paper suggests a two step Dx process to confirm that AD is the cause of cognitive symptoms. If both are negative other causes should be evaluated (e.g. stroke, diabetes).

Exhibit B: Increased Aβ plaques detected by the pTau217 test are not unique to Alzheimer’s. Recent work published in Nature Medicine found high blood levels of pTau217 and pTau181 in people with either of two relatively rare diseases that don’t cause dementia: transthyretin (ATTR) amyloidosis and immunoglobulin light chain (AL) amyloidosis. (Alzforum summarized the work here.) In systemic amyloidoses like these, amyloid clumps aren’t in the brain at all. Instead, they appear in places like the heart, peripheral nerves, and kidneys.

Although symptomatic systemic amyloidosis diagnoses are rare (4,000 a year versus 900,000 for Alzheimer’s), asymptomatic systemic amyloidosis is relatively common. Autopsy data shows that it’s present in 25% of people over age 80, all of whom would likely have been false positives if they had taken a pTau217 test. That’s in addition to the roughly 50% of folks in this age group who have no cognitive decline despite having amyloid in their brains.

COMMENTARY: Unfortunately, adding an eMBTR-tau243 test to screening in a primary-care setting will not help much. In this study, all 572 people studied had at least “subjective cognitive decline” severe enough to warrant a referral to a specialist. In order to count as a true positive, the patient had to have clinically confirmed “mild cognitive impairment” or “dementia” in addition to Aβ plaques and Tau tangles. Since 142 (25%) of the 572 suffered only subjective impairment, that meant they automatically tested negative for Alzheimer’s based on these criteria - even though 63 of them had positive pTau217 tests. It is quite possible that those folks might have been at earlier stages of Alzheimer’s, before they’d developed Tau tangles.

The two now-approved Aβ-removing drugs need to be prescribed before Tau tangles become established if they are to be successful, and pTau217 is still the best tool available to identify early stage patients most likely to benefit. But these drugs are relatively expensive, onerous to administer, and incur significant (but manageable?) medical risks. Given the high false positive rate, many will receive no benefit.

AI spots signs of intimate-partner violence in medical records

Diagnosing intimate-partner violence (IPV) is a delicate and difficult business. Patients are often afraid or ashamed to reveal their situations. When IPV is diagnosed by a clinician, that clinician is often a radiologist who recognizes telltale patterns of past physical trauma. But if radiology isn’t part of the diagnostic process, it can be all too easy to miss any signs that might be present. Questionnaire-based screening tools do exist, but their sensitivity and specificity vary widely. And they require clinicians to choose to administer them.

As we often mention, the identification of complex, subtle patterns is AI’s true superpower. So a group of researchers examined the question of whether AI could diagnose IPV based on a patient’s medical records.

The resulting study, published last month in npj | Women’s Health, suggested that it can. The most successful tool they tried was a multi-modal machine learning model that was able to distinguish between a patient experiencing IPV and one that wasn’t about 88% of the time.

The authors note that the tool is not intended to explicitly diagnose IPV - instead, it’s intended to flag patients so that clinicians can then take the appropriate next steps.

COMMENTARY: IPV is shockingly common. According to the US Preventive Services Task Force, nearly half of adult women and men report having experienced it. If AI can help even a fraction of these people find help, it could be a significant force for good.

We await this screening test for pancreatic cancer with bated breath

Screening for pancreatic cancer is desperately needed, but difficult to practically perform (see this recent Nature Outlook). The desperate need exists because this type of cancer is typically asymptomatic until it’s too advanced for treatment (3% five-year survival at stage IV), and those symptoms are shared with a host of minor upper-GI problems. Could a breath test be the solution?

A team at Imperial College London has reported intriguing progress developing such a test. No detailed results have yet been published from a 700-person feasibility trial just completed, but they were interesting enough for the charity Pancreatic Cancer UK to commit funds for a larger follow-on 6,079 person study.

COMMENTARY: We know that some diseases cause people to exude unique sets of volatile organic compounds (VOCs), and dogs and even some people seem able to distinguish them. In many ways, breath tests for cancer are the “holy grail” of non-invasive tests. Many single-analyte breath tests have been with us for decades (e.g., nitric oxide for asthma, urea for H. pylori).

The great hope is that AI analysis combined with VOC chromatography/mass spec should be able to do the same. One research group is even looking into the possibility of pan-cancer screening using VOCs.

We still believe in the screening potential, but limited sensitivity combined with overwhelming false positives has thwarted past efforts. That’s especially true for low-prevalence cancers like pancreatic, for which the average person has only a 1 - 2% lifetime risk.

Only half of advanced-cancer patients get sequencing

For folks with cancer in advanced stages, genetic sequencing of the tumor is, at this point, part of the best-practice standard of care. But a recent JAMA report reveals that the majority of patients in this situation still don’t receive it. For those who have inadequate insurance coverage, who live in less well-served regions, or are members of minority groups, the situation is even worse.

COMMENTARY: This was a broad study, covering the care of 63,294 patients at 280 US cancer centers from 2018 - 22. Both community and academic centers were included. Still, a number of factors might mitigate its conclusions. Perhaps most importantly, the study included only broad next-generation sequencing (NGS). It excluded the specific gene companion diagnostic testing that academic centers more frequently order. And of course, the pandemic seriously restricted timely cancer care during the latter half of the period studied.

Nevertheless, advanced cancers present with the broadest range of genomic abnormalities. And for an increasing number of them, there are effective targeted treatments. That’s why NGS has become the standard of care for these more advanced (stage III/IV) cases - as well as for earlier stages where it is likely that NGS testing is even less frequent.

Fragmentomics could improve both liver and breast disease dx

We are learning that fragments of DNA circulating in blood/serum (cfDNA) can reveal more and more about the presence and progress of many diseases. Two recent reports reveal how AI interpretation of the nature and abundance of cfDNA can also identify a broad range of physiologic problems. One focuses on liver cirrhosis, and the other on early breast cancer detection/typing.

cfDNA’s use in diagnostics

So what exactly is a cfDNA fragment, and how does it get used for diagnostics? Most DNA in circulation is found in the form of small fragments of about 150 DNA bases. That’s because it comes from the breakup of dying cells and the body’s efforts to get rid of the resulting debris. Sequencing these fragments can reveal harmful mutations as well as drivers of gene expression, aka epigenomics (e.g., the methylation that is the core biomarker used in multi-cancer early detection tests).

Actually capturing these cfDNA variants is hard, because blood and/or urine samples contain so few DNA copies (1,500-3,000 per tube). Often there’s not enough DNA to reliably identify mutations. But fragmentomics has an additional trick up its sleeve. Different disease processes create different fragment “fingerprints”, such as fragment length, abundance, or contents. It is these patterns that AI interprets into a clinical diagnosis (see this review for more details).

For liver disease diagnosis, fragmentomics increases accuracy

The need for better, simpler, cheaper liver diagnostics is very compelling. While 1 in 5 US adults have metabolic dysfunction-associated steatotic liver disease (MASLD), and 3 in 5 are at higher risk for the condition (primarily due to metabolic disorder), the existing less-invasive tests are not very accurate. One example is FIB-4: It’s only 7% sensitive for early disease, rising to just 64% in advanced cases. The reported fragmentomic approach was substantially better at early detection (50%) and somewhat better in later cases (73%).

For breast cancer, fragmentomics could decrease false positives

In contrast to liver disease, for which there is as yet no effective screening approach, mammography is well established in breast-cancer screening. But it is relatively expensive and has a substantial (about 10-15%) false positive rate. In women with denser breast tissue, effectiveness is even lower.

Could fragmentomics solve these problems? Identifying oncogenic gene mutations and methylation markers in the early stages of disease probably isn’t the whole answer, because there is so little mutated DNA to be found in circulation. However, the fragmentomics approach that was recently reported (called “TuFEst”) reported 96% sensitivity and 87% specificity, implying 93% positive and 92% negative predictive value in the group studied.

COMMENTARY: All research compares healthy controls with known cases, which is clearly the right way to discover novel approaches. However, as we have frequently discussed, studying a population composed of roughly equal numbers of true positives versus healthy controls can dramatically overstate PPV in the real world.

In the early-breast-cancer fragmentomics study, 65% of those studied were positive cases. Real-world breast cancer prevalence in China (where the study took place) is closer to 0.03%, i.e., 2,000 times rarer. If this technique were used for screening, it would generate an unacceptable number of false positives. The liver study examined a 38% prevalence group, which is (unfortunately) much closer to the real world.

With the increasing use of AI, which at its heart is a pattern-matching technology, it is likely that the best diagnostic tests will draw on a wider and wider range of diagnostic parameters from samples collected less and less invasively. That’s the goal of multiomics: combining the most effective grouping of the most relevant features to generate diagnosis at earlier and earlier disease stages. But that can only happen if the medical community and regulatory systems recognize that approaches are going to continue to become both more diverse and more tailored to specific risk populations – a one-size-fits-all test definition is just not going to work.

To hear more about how one company is using AI and fragmentomics to improve liquid biopsy, tune in to this week’s episode of our podcast, On Target: Targeting What Matters in Testing.

Match the antibiotic to the UTI in less than a day

One of the most annoying parts of having a urinary tract infection is waiting to find out which antibiotic will kill the bug that’s in your bladder. First the lab has to grow the bacteria, and then it has to assess how well various antibiotics kill the microbes. It takes days.

That wait could soon be cut to about six hours. Research published in JAC-Antimicrobial Resistance evaluated a test that works very differently. It draws a mix of infected urine, growth medium, and a blue dye into 10 tiny tubes. Nine of those tubes contain antibiotics, and one is a control. If the antibiotics in a particular tube kill off the bacteria, the dye in that tube stays blue. If the bacteria keep growing, it turns pink.

The test worked well. Its results agreed with those obtained using the standard method 97% of the time (572/590 samples). It also worked in the presence of boric acid, which is commonly used to keep the amount of bacteria in urine samples constant while the samples are shipped to the lab (158/160 samples had the same results with or without boric acid).

PREVENT predicts risk - of more than heart disease

Back in November 2025, we reported on the American Heart Association’s rollout of the PREVENT(™) Online Calculator. The tool was intended to assess people’s risk of cardiovascular disease (CVD) over the next one to three decades as compared to others of the same age and sex.

A month later, we followed up with coverage of a study suggesting that PREVENT wasn’t any better than ASCVD+, the tool it was intended to replace. (Both scoring systems use data like cholesterol and blood-pressure levels along with patient history (Are you diabetic? Do you smoke?) to assess the patient’s risk level.) The patients in the study were folks who experienced an acute coronary event. Both PREVENT and ASCVD+ would have flagged only about half of them as folks at high risk of CVD.

Which makes a recent study all the more interesting. It found that women who PREVENT classified as high-risk were indeed at higher risk - of hip fractures. Women in the high-risk category had a 93% higher chance of hip fracture than women in the low-risk group. Looking more closely, of the women classified as high risk in the study, about 11% (118/1100) had hip fractures at some point later in life (they were followed for a median of about 10 years). Of the women classified as low risk, only about 5% fractured a hip (243/5,370). Higher cardiovascular risk was also linked to a greater risk of fractures in major areas such as the spine, forearm, or shoulder.

Measles

No new measles cases have been reported in the South Carolina outbreak in over two weeks. If that continues through April 26, the outbreak - one of the nation’s largest in 40 years - will officially be over.

Meanwhile, the outbreak in Utah is still growing, with 142 people diagnosed over the last three weeks. And the Texas outbreak in an ICE detention center has now spread to people who work in the facility, which means that people in the larger community may now be exposed.

The current larger outbreaks in the US are shown below: Arizona / Utah, California, Colorado, Florida, North Dakota, Ohio, South Carolina, Texas, Washington.

Susan Tousi, CEO of DELFI

This week, Susan Tousi shares how her company is using AI and fragmentomics to make liquid biopsy more accessible and cancer less deadly. (Disclosure: Mara’s venture firm is an investor in this company.)

Listen in: ontargetpodcast.com

Join the conversation: If you’re leading innovation in this space and want to share your perspective, we want to hear from you. info@ontargetpodcast.com