A start-up company created by machine learning researchers at Johns Hopkins University is emerging and has received a new sepsis model and $15 million in funding. In a world full of AI models, Bayes Health hopes to differentiate itself through data that shows the effectiveness of the tool-which is rare in the field of machine learning.
Founder and CEO Suchi Saria has been working in the field of machine learning for nearly 20 years. She published some of the first studies showing that machine learning can be used to identify sepsis at an early stage, and built a model for use at Johns Hopkins University.
Now that other models have been developed, Saria hopes that the company’s research-based approach will help it stand out from other models.
“Today, health AI is hyped. It is really difficult to do it well. If we can really solve this problem, then the chance of reducing preventable deaths will be great,” she said.
Part of the problem is that although more and more digital health solutions are released to the market, there is usually little data support, which makes it difficult to win the trust of suppliers. Saria hopes to change this situation with the results of a recent study Data published in the past On the company’s sepsis model.
Saria launched Bayesian Health in 2018 and has since raised $15 million in a funding round led by Andreessen Horowitz. The company is seeking to commercialize its machine learning algorithms, starting with its tools to detect sepsis.
The startup recently shared the results of a prospective study evaluating the use of the tool by doctors.Although it is only a preprint-it has not yet passed peer review-but it is a different approach Most models only use data for evaluation This was collected before implementation.
The model was tested in five Johns Hopkins hospitals from 2018 to 2020. Of the approximately 9,800 patients who were later diagnosed with sepsis, the model labeled 82% of them.
Of these, 3,775 patients had no antibiotic orders before the alert, but received orders within 24 hours. Importantly, approximately 89% of doctors and nurses actually use alarms.
Saria says this is an important metric because it can show whether the tool is timely or useful to clinicians.
“If you use something that is not timely, it will sound an alarm, but usually after the provider treats the patient, it is not very effective,” she said. “Or, if it is sending out an alarm, but there are a lot of false alarms. …If the number is really high, the supplier is really busy. They don’t have time to do this.”
Currently, most decision support tools used by hospitals, including sepsis alerts, have not yet been approved by the Food and Drug Administration. This makes the hospital have to listen to the developer’s statement about the accuracy of the model, hoping to verify it through their own evaluation.
Some models that tested these algorithms found that they were not as useful as advertised. A recent study of Epic Systems’ sepsis model found that Performance is “much worse” than claimedAlthough a large number of alarms have been generated, only a small number of cases of sepsis that clinicians have not yet discovered have been discovered.
In addition to its work on sepsis, Bayes Health is also developing models for clinical deterioration, care transition, and stress injury. These are not only important quality measures for hospitals, but can also change the lives of patients. After losing her nephew with sepsis, Saria knew this very well.
“The difference between correct and incorrect can mean a person’s life,” she said.
Photo credit: Gremlin, Getty Images
