Spotlight: Interrogating the Immune System for Diagnosing Infection

Startup Inflammatix, founded in 2016 in Burlingame, California, takes a different approach to diagnosing infections. Rather than seeking the pathogen causing the infection, it uses a data-driven approach to evaluate the body’s immune system response to provide a diagnosis in as little as an hour. Big3Bio editor Marie Daghlian spoke with Dr. Timothy Sweeney, co-founder and CEO and Jonathan Romanowsky, co-founder and COO, a few days after new research was published in Nature Communications, demonstrating the strength of its sepsis test’s prognostic ability.

MD: How did Inflammatix start?

TS: Inflammatix was born out of a clinical need. My background is in medicine; I came to California (to Stanford) to train as a surgeon after medical school. During surgery residency, I grew very frustrated not knowing which of my post-operative patients needed antibiotics. I wanted to find a better way to diagnose them, so I began doing the research with Dr. Purvesh Khatri, our co-founder and scientific advisor. This work ultimately led to the founding of Inflammatix. It turned out that our research was much more successful than we had hoped. After a few years of work at Stanford we had developed gene modules to determine the presence, type, and severity of an acute infection. Together, these could form an excellent diagnostic for sepsis. We realized that these were products that needed to be commercialized and brought to market so we founded Inflammatix. It was on that journey that we found Jonathan.

JR: When Tim reached out and told me they had a validated assay for infections and sepsis, I became very intrigued because sepsis has been an area of significant unmet need—it is the largest cost to the US healthcare system, the number one killer in hospitals in the United States, and a great global burden as well. Then I saw the number of publications and the quality of the data science that had been done. My many years of experience commercializing novel molecular diagnostics, started at XDx (now CareDx), then at CardioDx and Veracyte, including launching the Afirma test for Veracyte. When I met Tim, I was pleased that they wanted to bring on someone who had commercial experience in bringing these types of tests to market. So we decided to join together to found Inflammatix. We initiated our fundraising efforts, and we closed our series A with Khosla Ventures as our lead investor about a year ago.

MD: What are you focused on now?

Dr. Timothy Sweeney, Co-founder and CEO, Inflammatix

TS: Right now Inflammatix is developing HostDx Sepsis and HostDx Fever, our two diagnostic tests, into rapid, point-of-need diagnostics in collaboration with diagnostic instrument manufacturers.

MD: How do these tests work?

TS: The goal was to figure out what the immune system is reacting to. The immune system gives external symptoms like fever or joint pain or inflammation in response to different kinds of insults. Those insults might be a bacterial infection, or a viral infection, or something entirely noninfectious like a post-surgical blood clot. Our goal is to find patterns in what the immune system is doing that allow us to figure out whether it is reacting to a bacterial or a viral infection, and do risk stratification. We use a data-driven bioinformatics approach that incorporates immune system data from hundreds or sometimes thousands of patients to find very robust gene signatures that are generalizable across a very broad population. The way the test actually works is twofold: in development, first we select a number of genes that are statistically robustly associated with the condition of interest and then on top of that we add an algorithm that interprets how the expression level of those genes relates to some immune system state. The algorithms tell us, for example, what’s the likelihood of a bacterial infection, what’s the likelihood of a viral infection. What we are looking at are genes whose levels of expression are changing as the immune system reacts to different things. We read out those levels and then our algorithms interpret those levels and put them into an actionable score. It’s a data science approach that is then reduced to practice in a rapid molecular diagnostic.

MD: How does it work in the clinic? A sick patient comes into the clinic and you draw blood that goes into the molecular diagnostic?

TS: Yes, it depends on the circumstance. We are developing two different tests so there are two scenarios. For the HostDx Fever case, a patient might have a cold and a fever and goes to their primary care physician with the question, “Do I need an antibiotic right now? I’m not feeling well, I know that I am sick with an infection, is it bacterial or viral?” The test involves a small blood draw that goes into the device and the device will tell the doctor the likelihood of the infection being bacterial or viral so he or she can quickly determine if antibiotics should be prescribed. The HostDx Sepsis test, on the other hand, is being designed for use in triage in the emergency room. Patients go to the emergency room for acute infections but they also show up for other reasons such as belly pain or a rapidly spreading rash or symptoms that aren’t obviously infectious—they may be entirely noninfectious or may be from bacteria or a virus. So in our assessment of the immune response we have to include these other possibilities of there being no infection at all or having a bacterial-viral co-infection. The other thing that the HostDx Sepsis test does is it evaluates the severity of the response separately from whether the patient has an infection or not. So, HostDx Sepsis is a test with more features.

One question is: Does this patient need antibiotics? Is that why they are here in the emergency room? The other question is: How sick is this patient? Even if we treat them with antibiotics, does it have to be IV antibiotics and do they need to stay in the hospital for a couple of days, or is it safe to send this patient home with oral antibiotics (assuming we have decided they have a bacterial infection)? So the test provides risk stratification in addition to detection.

So blood is drawn that goes into the device and the doctor gets an answer in about an hour.

MD: That’s the amazing part to me—that you can get an answer in an hour. And in the case of sepsis, the faster treatment is started, the better the likelihood of a good outcome.

TS: That’s one of the major advantages of using the immune response [for answers]. The traditional way to diagnose someone with an infection is to look for the pathogen. There are two problems with looking for the pathogen in assessment of patients with possible sepsis. If you look at patients in a hospital setting who are later judged to have had infections, only about 30 percent of them have bloodstream infections. The vast majority of patients who show up in a hospital have a local infection (like a skin or soft tissue infection, or pneumonia, or a belly infection) that never makes it into the bloodstream. But often there is no way to sample those sites. We liken [looking directly for pathogens] not just to looking for a needle in a haystack, but to looking for needles in haystacks that frequently have no needles. So it is a very difficult task.

The other problem is that the process takes a long time. There is a lot of amplification of the pathogen that has to be done in order to identify it.

The great advantage of looking at the immune response is that the genomic material we read (which comes primarily from white blood cells) is abundant in the blood. The immune response is circulating all the time. It’s not a matter of some rare event occurring, it’s only a matter of measuring the relative amount of these different genes and then applying our host response algorithm on top of them. That’s why we can be so rapid compared to traditional diagnostics.

MD: It must have taken a lot of data mining before you could create the algorithms in order to have enough data of all the various immune responses to different diseases, different pathogens, right?

TS: Absolutely. That’s the hardest thing. One could imagine that for this test to be useful in distinguishing between a bacterial and a viral infection, we would have to have seen immune responses in a huge swath of the population because all kinds of people come into the clinic with infections. It has to work in children and adults and it has to work in people across the world. It has to work for all different kinds of infections and it has to work whether the patient waited one day to go to the doctor or waited seven days. All of those [situations] contribute to this vast heterogeneity of immune response in these patients. The Inflammatix challenge and opportunity was to derive a data science pipeline that can adequately wrangle all of that heterogeneity into a single signature that works in all of those settings. The down side of that is it is a difficult task, but the upside is that now that we have been successful, we know that it will be difficult for our competitors to follow us. We have come up with this test that is really robust and is applicable in all kinds of settings for all kinds of different patients.

MD: Have you partnered with anybody to facilitate gathering all that data?

TS: Yes, some of the data are publicly available because groups such as the National Institutes of Health, Bill & Melinda Gates Foundation, and other institutions that have a mandate for public reporting funded the original data collection; and academic investigators, especially for the most recent sepsis mortality risk stratification study, privately held some of the data. This was actually a large collaboration of groups from around the world that contributed data to allow for this kind of analysis work to be done. Then as a company, we are now growing a proprietary sample bank that we can use to continue to further develop our current tests based on the input from clinical trials that we are gathering here. So it is a mix of public and private data.

MD: How did DARPA get involved? I noticed that they are one of your backers.

TS: DARPA has funded grants to develop the HostDx Sepsis test into a rapid point-of-care test. From the Department of Defense’s perspective, the need may be two-fold: one is force readiness in terms of helping wounded soldiers through the healthcare system, avoiding unnecessary antibiotic usage and getting well quickly; the other is for bio-threat detection of emerging viruses either from animals or bio-warfare, because we don’t have any molecular diagnostics for those new pathogens. By relying on the host response, which is relatively unchanged, we could potentially screen patients in a new outbreak and find those that need to be quarantined. I do not speak for DOD or the US government; that is my personal assessment.

MD: You have shown that the test has a very high negative predictive (NPV) value. What does that mean?

TS: When we say a patient does not have an infection, how often are we correct in that assessment? So an NPV of 99 percent means that if we say 100 patients in a row don’t have infections, we are only wrong once. It is a number dependent both on the test itself, and the prevalence of the disease in the tested population.

Jonathan Romanowsky, Co-founder and COO, Inflammatix

JR: And the reason why that is critical is that if a goal for using this test is to decrease the number of antibiotics prescribed and to enable physicians to send appropriate patients home as opposed to admitting them [to the hospital], or to release them sooner than they would have earlier, the physicians need to have high confidence in the testing to help them make that decision. So a very high NPV is critical for facilitating those discussions.

MD: What is on the horizon for Inflammatix? What is the status of the development of these tests?

TS: We have conducted a number of clinical trials and continue to do prospective validation. For the breadth of the claim that we can determine bacterial vs viral infection in the general population, we really need to have a solid base of clinical evidence to convince physicians that what we report is true. Some of those trials were enrolled while the research was pre-commercial at Stanford, and some of them we are actively enrolling now as a company. To develop the product for commercialization, we are partnering with diagnostic instrument manufacturers who have an ideal device for a given market. The HostDx Fever test, which is destined for the doctor’s office, needs to be really fast but also have a small footprint in order to fit into a physician’s office. That’s quite different from HostDx Sepsis, which also needs to be fast—60 minutes is fast enough—but will be in a high throughput, random-access setting in a hospital central lab. So in our partnerships we have the ability to match our diagnostic content to instruments that best fit the particular market we are aiming at.

MD: What is your timeline for these products to be available?

TS: A reasonable estimate is in about 18 to 24 months.

MD: What is your current financing situation? I know you raised a series A. Will that tide you over until the products are commercialized?

TS: We are fundraising now and that should take us past initial commercialization.

MD: It’s a promising time for raising money for diagnostics companies and investors are showing more interest in them with the advent of artificial intelligence (AI) and bioinformatics.

TS: I can’t comment on how it used to be but we have had good success so far because people understand that the diagnostics we are building aren’t really a push into the market, they are a pull. Physicians tell us they want tests like this. The health system is telling us it wants to save money. Health economics studies suggest that tests like ours should be both cost saving and an improvement in patient health. That’s the Holy Grail, right? Also, our ability to continue tap our data science pipeline will allow us to grow our menu of diagnostic content in a relatively capital-efficient way. It certainly seems to be making good inroads with investors and stakeholders.

There are two points that are often glossed over but are really the key to understanding the full value of these tests, especially HostDx Sepsis. One point, which we already discussed, is that most patients with bacterial infections don’t have bloodstream infections, which limits the diagnostic utility of traditional pathogen-focused methods. Perhaps more important is the fact that the vast majority of patients seen in the emergency room for acute infections are not septic. Most patients will be discharged and sent home. The difficulty is distinguishing between people who are ultimately likely to get better versus people who are likely to progress to needing hospitalization and further care. That’s a really hard decision to make. You might ask why a physician would withhold antibiotics from a person who he or she thinks has sepsis? The answer is that for the millions of patients per year who come to the emergency room with infections, we wouldn’t recommend withholding antibiotics from the subset who are very sick and on their way to the ICU, even before it is known whether the infection is bacterial or viral. But the majority of patients are not in that super-sick population—it could be someone coming in who has a mild case of pneumonia. They are given fluids and feel better. Now the question is whether they need to stay in the hospital or can recover at home and whether or not they need antibiotics. That’s the bread and butter patient that we think HostDx Sepsis will be useful in improving their care.

MD: It’s an unmet need and I can’t think of anything on the market right now that addresses that. It sounds like it could save hospitals money in unnecessary care and no one wants to send someone home who needs the care.

TS: One of our successes is our risk module. The recently published paper in Nature Communications demonstrated our excellent clinical performance in that area. More importantly, it wouldn’t require a second follow-up test. There are other biomarkers, like lactate and procalcitonin, that are fairly accurate at distinguishing the likelihood of 30-day mortality but they all need a follow-up test in 6, 24, or 48 hours to make that prediction. One of the nice things about HostDx Sepsis and the performance we have shown so far is that a prediction would be available at the time of diagnosis. You wouldn’t have to wait. That will be an advantage.

JR: And adding to that, we did some formal market research. We commissioned students at Arizona State University’s International School of Biomedical Diagnostics to conduct independent market research. They polled emergency room physicians and assessed their perspective on the HostDx Sepsis test. We presented them the performance of the test that we had published and 93 percent of them said they would likely recommend a test like this to their institutions for adoption, and that they would use it more than 14 times per week. It indicates that there is a lot of interest in our test and it would be used very frequently.

MD: Thank you.