Company Spotlight: Inflammatix TriVerity - Reading the Immune System's RNA to Predict Sepsis

Maya: Welcome to AI in Medicine - Smart Summaries. Just a quick heads up before we start, our voices are AI-generated.

Sam: And today we are doing a Company Spotlight. We are looking at a public document, specifically the FDA 510(k) summary for a product called TriVerity, made by Inflammatix.

Maya: The hook for this one is really compelling. Imagine an emergency department physician has a patient with suspected sepsis. This document shows a test that takes blood, isolates the RNA, and in about 30 minutes, generates scores that differentiate between a bacterial and a viral infection. Plus, it determines the likelihood that the patient will need a ventilator, vasopressors, or renal replacement therapy within 7 days.

Sam: That is exactly the kind of clarity you want in an emergency. To be clear, we are not looking at a company pitch deck today. This is the FDA's official 510(k) summary published on January 10, 2025. It outlines exactly what this device is, the legal basis for its clearance, and the real testing data the company submitted.

Maya: Right. This episode is not sponsored, and our goal is just to unpack what the data actually says. So let us start with what the TriVerity test actually is. The summary describes it as an automated, semi-quantitative in vitro diagnostic test. It runs on a specific machine called the Myrna instrument.

Sam: How does it actually work? Because we talk a lot about AI analyzing images or patient records, but this is happening right at the molecular level.

Maya: It is measuring the relative expression levels of host response genes. Instead of trying to find the specific bacteria or virus in the blood, it looks at how the patient's own immune system is reacting. The test isolates RNA from whole blood collected in a PAXgene Blood RNA tube. Then it uses a process called reverse transcription loop-mediated isothermal amplification, or RT-LAMP.

Sam: Okay, RT-LAMP. What exactly is it looking for during that process?

Maya: The summary lists the measurands. It targets 29 informative host response RNA transcripts, which are specific genes like ARG1, CD163, and IFI27, plus 3 housekeeping transcripts used as baselines. The cartridge contains all the reagents. Once the sample goes in, all extraction and amplification are fully automated.

Sam: And where does the AI or machine learning come in?

Maya: The test quantifies the amount of each of those 32 transcripts based on fluorescence detected by the instrument. Then, it takes those 32 values and feeds them into two fixed classifiers. Those classifiers output three separate scores.

Sam: Three scores. What are they telling the clinician?

Maya: Each score falls into one of five discrete interpretation bands: Very Low, Low, Moderate, High, and Very High. The first score is for the likelihood of a bacterial infection. The second is for a viral infection. And the third is for severe illness, which is specifically defined as the need for mechanical ventilation, vasopressors, and/or renal replacement therapy within seven days.

Sam: That is a lot of information from one 30-minute test. Who is the intended patient here? Can this be used on anyone?

Maya: The indications for use are very specific. It is for adult patients with suspected acute infection or suspected sepsis presenting to the emergency department. It is also meant to be used in conjunction with clinical assessments and other laboratory findings as an aid. It does not replace the doctor's judgment.

Sam: Got it. Now, because this is a 510(k) clearance, the FDA requires the company to show that their device is substantially equivalent to a legally marketed predicate device. What was the predicate here?

Maya: The predicate device is called SeptiCyte RAPID, which was cleared under K203748. They are both classified as Class II devices under the product code PRE, which stands for RT-qPCR Assay For mRNA Transcript Immune Biomarkers.

Sam: Are they identical, or are there major differences?

Maya: There are some big differences outlined in the comparison table. SeptiCyte RAPID uses reverse transcription polymerase chain reaction, or PCR, and it only measures two mRNA transcript immune biomarkers. TriVerity uses RT-LAMP and measures 29 host response genes plus the 3 housekeeping genes.

Sam: That is a massive step up in complexity. Two genes versus 32 total transcripts.

Maya: Exactly. Also, the patient population is different. SeptiCyte RAPID is indicated for patients suspected of sepsis on their first day of Intensive Care Unit admission. TriVerity is for patients presenting to the emergency department.

Sam: Moving it up to the front door of the hospital. That makes sense. Let us talk about the testing. How did they prove this thing actually works, both in the lab and on real patients?

Maya: The summary breaks down extensive analytical testing first. For example, they did a multi-site reproducibility study using contrived samples across two external sites and one internal site, with multiple operators and 12 different instruments. The standard deviation measured in each of the 15 bands was within their acceptance criteria.

Sam: Did they test for interferences? Emergency department blood samples can be pretty messy.

Maya: They did. They spiked potential interferents into contrived samples at concentrations higher than normal clinical ranges. They tested things like Bilirubin at 400 mg/L, Hemoglobin at 10 g/L, and drugs like Vancomycin at 100 mg/L and Dopamine at 500 mg/dL. The document states no interference was found for any of the substances they tested.

Sam: What about the actual clinical trial? Lab tests are great, but real humans with suspected sepsis are complicated.

Maya: They ran a prospective, multi-center clinical study called SEPSIS-SHIELD. It included emergency departments in 21 locations across the US and one in Europe. Out of 1,441 participants who consented, they ended up with 1,222 evaluable participants for the diagnostic endpoint and 1,120 for the prognostic endpoint.

Sam: And how did they determine what the patients actually had? The 'ground truth', so to speak.

Maya: They used clinical adjudication as the reference standard. They broke the results down into two populations: a 'forced adjudication' group where patients had to be classified as infected or non-infected, and a 'consensus adjudication' group, which only included patients where the clinical adjudicators were certain of a Yes or No.

Sam: Let us look at the consensus group numbers, since that is where the ground truth is most solid. How well did TriVerity spot bacterial infections?

Maya: In the consensus group, the prevalence of bacterial infection was 63.1%. When the test returned a 'Very High' bacterial score, the negative percent agreement, meaning how well it ruled out false positives, was 95.5%. The probability of the patient actually having a bacterial infection if they fell into that band was 93.2%.

Sam: And what about the other end of the spectrum? If the test said 'Very Low'?

Maya: In the 'Very Low' band, the positive percent agreement was 97.2%. The probability of no bacterial infection was 87.9%. The area under the curve, or AUROC, for the bacterial result was 0.83.

Sam: An AUROC of 0.83 is solid for a complex condition like sepsis. What about viral infections?

Maya: Viral performed even better. In the consensus group, the AUROC was 0.91. For the 'Very High' viral band, the negative percent agreement was 98.6%, and the likelihood ratio was 40.93. The probability of having a viral infection in that band was 92.9%.

Sam: Wait, really? A likelihood ratio of 40.93? That is remarkably high. That means a positive result in that band heavily shifts the probability that the patient actually has a virus.

Maya: Yes, it is a very strong signal. And on the prognostic side, predicting severe illness within 7 days, the prevalence in the study was 10.9%. The AUROC for the severity score was 0.78. In the 'Very High' illness severity band, the negative percent agreement was 98.7%, with a likelihood ratio of 11.33.

Sam: So if a patient comes in, and the test says 'Very High' for severity, they are significantly more likely to need a ventilator or pressors in the next week. That is something a hospital system would definitely want to know on day one.

Maya: Exactly. The summary notes that 79.6% of the severity results fell into the clinically actionable outer bands. So it is not just dumping everyone into the 'Moderate' middle ground. The majority of patients get a clear directional signal.

Sam: How does this compare to what emergency departments are already using? Things like procalcitonin or white blood cell counts?

Maya: The document actually includes a direct comparison table. Using consensus adjudication, TriVerity's bacterial AUROC was 0.83. In that same group, C-Reactive Protein was 0.74, Procalcitonin was 0.71, and White Blood Cells were 0.76. So the TriVerity score outperformed those standard biomarkers.

Sam: That is a meaningful jump in accuracy. Did they look at how this performs across different patient demographics? Sometimes biomarkers do not work equally well for everyone.

Maya: They did, and this is one of the most interesting tables in the summary. They broke down the accuracy of TriVerity versus Procalcitonin by race. For Black or African American patients in the consensus group, the TriVerity bacterial score had an AUROC of 0.83, while Procalcitonin was only 0.66.

Sam: Wow, 0.66 is not great at all for a clinical decision tool.

Maya: Right. The summary explicitly states that the AUROC ranges across different races were narrow for TriVerity, a maximum variance of 0.10, whereas they were markedly wider for Procalcitonin, a variance of 0.18. The document notes this potentially indicates an impact of race on the accuracy of Procalcitonin, and that TriVerity could substantially improve diagnosis in minorities where procalcitonin underperforms.

Sam: That is a crucial equity point for hospital buyers to consider. What about patients with compromised immune systems? If this test measures the host response, does it fail if the host's immune system is weak?

Maya: It is a smart question. They analyzed that specifically. A total of 219 participants, which is 17.9% of the forced population, were immunosuppressed. This included patients with malignancies, HIV, solid organ transplants, and those on steroids. The likelihood ratios and AUROCs did not differ markedly based on the presence or absence of immunosuppression.

Sam: So the machine learning model was robust enough to handle the altered immune responses. That is impressive. But we have to look at the limitations. What does this 510(k) summary not show us?

Maya: A few things. First, this is exclusively cleared for adult patients. The document says nothing about pediatric populations, so we do not know how it performs in children. Second, the severity prediction is strictly capped at a 7-day window. It does not speak to 30-day mortality or long-term outcomes.

Sam: And the test still requires the clinician to act on it. The FDA letter is very standard but clear. It reminds the company that clearance does not mean the FDA has determined the device complies with all other federal statutes. And it explicitly says the test is indicated for use in conjunction with clinical assessments.

Maya: Exactly. Another practical limitation is specimen stability. The document says the blood drawn in the PAXgene tube must be run on the machine within 12 hours if kept at room temperature. They tested stability at 15 °C and 25 °C up to 14 hours and it passed, but 12 hours is their recommended window. That requires decent operational flow in the ED.

Sam: Let us tie this together for the health system leaders and medtech companies listening. What is the big takeaway here?

Maya: For hospital leaders, this shows that we are moving beyond single, unreliable biomarkers like procalcitonin. We now have FDA-cleared tools that can read 32 RNA transcripts at the bedside in 30 minutes to provide a highly accurate, race-neutral risk profile for sepsis.

Sam: And for companies building in this space, it proves the FDA is comfortable clearing fixed machine learning classifiers that translate complex transcriptomic data into simple, actionable interpretation bands. The fact that Inflammatix mapped this to 15 discrete bands across three scores shows a very thoughtful approach to user interface in high-stress environments.

Maya: That is a great wrap-up. Today we covered the FDA 510(k) summary for Inflammatix's TriVerity test. We looked at how it uses RT-LAMP to analyze host response genes, its impressive clinical accuracy in predicting both infection type and severe illness, and its performance across diverse populations.

Sam: As always, there is a link to the full FDA document in the show notes so you can read the data yourself. And remember, this podcast is for informational purposes only and is not medical advice.