NEW YORK – After the publication of a new study in the New England Journal of Medicine AI last week, artificial intelligence-based diagnostics company Prenosis is preparing for the clinical implementation of its US Food and Drug Administration-approved sepsis software.
The NEJM AI study is the publication of the data Prenosis submitted to the FDA for its de novo marketing authorization of the Sepsis ImmunoScore, which it received in April. The software uses 22 parameters, including an assortment of biomarkers and patient clinical data, to generate a risk score indicating a patient's likelihood of developing sepsis. The AI- and machine-learning-based tool is integrated into hospital electronic medical records to provide insights about patients who have been admitted to the hospital.
Chicago-based Prenosis has built a proprietary biobank and dataset of more than 113,000 blood samples from more than 28,000 patients that it used to develop the Sepsis ImmunoScore. The software's parameters include demographic data, vital sign measurements, comprehensive metabolic panel measurements, complete blood count panel measurements, lactate levels, procalcitonin, and C-reactive protein.
The primary endpoint of the NEJM AI study was sepsis presence within 24 hours of test initiation, while the secondary endpoints included length of hospital stay, intensive care unit admission within 24 hours, mechanical ventilation use within 24 hours, vasopressor use within 24 hours, and in-hospital mortality.
The Sepsis ImmunoScore had an area under the curve of 0.81 in the external validation cohort for diagnosing sepsis. In addition, the risk score categories of low risk, medium risk, high risk, and very high risk "demonstrated good predictive ability" for the secondary endpoints, the researchers wrote.
The researchers also analyzed the software's performance in discriminating between patients who had sepsis at initial evaluation from those who did not have sepsis within 24 hours, finding an area under the curve of 0.84 in the external validation set. When determining the prognostic performance of the software to discriminate between patients who did not have sepsis at initial evaluation but who later developed sepsis within 24 hours and patients who never had sepsis, the software had an area under the curve of 0.76.
Prenosis CEO Bobby Reddy said that the new publication will help the company and its software build trust and credibility in the healthcare world. "There is a lot of skepticism about AI in general in healthcare," he noted. "At all stages, we take all opportunities to try to prove to the world and to ourselves that we are doing AI in the right way that protects patients and improves care."
While many hospitals have sepsis alert systems that measure vital signs to indicate when a patient may be showing signs of sepsis, Reddy said that those tools can have particularly high rates of false positive results. The firm's Sepsis ImmunoScore can be implemented in hospital systems to filter out those false positives and visually display how the risk score was calculated for the patient so clinicians can review it, while also providing predictive information.
"An AI tool for sepsis allows us not just to diagnose sepsis, it allows us to introduce a whole element of prediction," Reddy said. "We can move from being very reactive, which is how a lot of sepsis care is today, to being a lot more proactive, but in order to be proactive you need that predictive information."
Currently, the software has been implemented in multiple health systems across the US, but it hasn't yet been used clinically on patients due to the time it takes to get hospital approvals, train clinicians, and get the software integrated into a hospital's IT system, Reddy said. However, he noted that clinical use of the software is "very close."
The firm is also working on reimbursement for its sepsis software. Although inpatient diagnostic tests are usually grouped together by the length of a patient's stay for reimbursement and not paid individually, Prenosis plans to apply to the US Centers for Medicare and Medicaid Services for a new technology add-on payment (NTAP). In order to receive this payment, the medical service or technology must be new, costly enough that the regular rate paid for inpatient hospital stays is determined to be inadequate, and must demonstrate a substantial clinical improvement over existing services or technologies.
Reddy said that it can be a lengthy process to submit and get approval for NTAP, but there are a lot of financial incentives for hospitals to improve their sepsis care, and there's "motivation for them to want to pay for Sepsis ImmunoScore."
The firm has also received a vote of confidence from Roche, which signed a distribution agreement with Prenosis for Sepsis ImmunoScore in April. Under the deal, the software will be available on Roche's Navify Algorithm Suite. Roche's "huge sales force" and connections with health systems will allow Sepsis ImmunoScore to reach a larger number of hospitals and patients, Reddy noted.
Prenosis is currently focused on the US market, but Reddy said that the firm is investigating the clinical studies it will need to run for European regulatory approval.
Sepsis can be particularly difficult to diagnose and treat due to its variations in the source of infection.
"It's just very difficult to treat any condition when there's so much heterogeneity and so much variability in how we understand the condition," Reddy said.
However, sepsis has been a growing area of interest for many in the diagnostics industry, with startups and established companies alike developing tools for diagnosis, prognosis, and risk stratification. Immunexpress has an FDA-cleared, qPCR-based rapid test that uses gene expression to determine whether a patient's inflammatory response is characteristic of sepsis, while Italian firm Diasorin received CE marking last year for a chemiluminescent immunoassay that stratifies patients by disease severity and the risk of unfavorable outcomes.
Meantime, startup Cytovale has raised $184 million in two recent funding rounds to commercialize its FDA-cleared IntelliSep prognostic test to stratify patients based on their risk of sepsis, and Swiss diagnostics firm Abionic received FDA clearance in October for its ELISA-based test that uses the pancreatic stone protein biomarker to detect sepsis early.
Sepsis is the first frontier for Prenosis, Reddy noted, but the company has larger plans to bring precision medicine and new therapeutics to the acute care setting with its products. The firm is developing software that functions like the Sepsis ImmunoScore but is associated with specific treatments, such as indicating whether a certain treatment may provide a decreased mortality risk.
"We need to enable clinicians to see very deeply into the biology of each patient and then use that information about the biology of each patient to personalize treatment," he added.
Prenosis has a prototype software that it has trained on data from more than 9,000 patients and validated with about 400 patients, but Reddy said the firm still must conduct an interventional randomized control trial before it will be ready for clinical implementation. The company has had a meeting with the FDA and has two more planned for next year to finalize its trial design. It hopes to begin the trial by the end of 2025 and seek FDA approval within the next three years if the trial is successful, Reddy said.
The firm hopes to bring existing drugs that are shown to be safe and effective for other conditions into the acute care setting, such as drugs to help control immune response in patients with lupus or rheumatoid arthritis, Reddy added.
Prenosis plans to use that software to expand more broadly into the therapeutic space for acute care settings. The firm intends to partner with large pharmaceutical companies to run clinical trials for immunotherapies that may be used for common conditions in the acute care setting, such as stroke or heart failure. Using its software to determine patient eligibility and dosages for the therapies, Prenosis will generate the data and, depending on the outcome of those trials, potentially become the drug sponsor for FDA approval and sell the therapy with the pharmaceutical partner.
"We finally have the capability to use tools like AI to take a lot of different elements of data together to holistically evaluate patients" for clinical trial eligibility, Reddy noted.
According to him, Prenosis' broader ambitions beyond sepsis put the firm in a better position than some of its competitors who are exclusively zeroed in on sepsis, due to the difficulties in diagnosing and treating a variable disease like sepsis.
"A lot of these companies that are very focused only on doing sepsis diagnostics or sepsis alert systems, they're kind of chasing a ghost in many ways," he said. If "you're trying to diagnose sepsis or you're trying to predict sepsis and that's your whole company's purpose, I just don't think it's sustainable in the long run."