Scouting Report-Bayesian Health:AI Tools to Reduce Physician Overload & Improve the Quality of Care
Bayesian Health, a spinoff from John Hopkins University raised $15 million following its emergence from stealth mode in a funding round led by Andreessen Horowitz, along with Health 2047 Capital Partners, Lifeforce Capital, and Catalio Investments. Founded by the director of the machine learning and health care lab at John Hopkins, Suchi Saria, the company will use the funding to commercialize its sepsis detection machine learning algorithm and develop other models to detect conditions earlier to improve care.
According to the Journal of Patient Safety, approximately 400,000 preventable deaths cost over $17 billion a year.
Almost 300,000 people per year die from sepsis, a complication from infection, accounting for 1 in 3 patients who die in a hospital according to the CDC.
One in every six patients are affected by diagnostic errors and one in every 1,000 primary care visits cause preventable harm.
Sepsis is treatable with early diagnosis and intervention yet the risk of dying from sepsis remains in the range of 10-30%.
Founded in 2018 by Dr. Suchi Saria, Bayesian Health has received honorary recognition and awards for its digital health platform, powered by AI, allowing clinicians to quickly provide a better quality of care. Dr. Saria spent over 5 years researching and developing machine learning models to detect early signs of sepsis when she lost her young nephew to sepsis. Her research attempts to demystify health AI data has gained the trust of many healthcare professionals; it provides clinicians with reliable tools to improve the quality of care. Bayesian Health makes the EHR system more proactive by allowing physicians to catch life threatening conditions earlier with the use of AI models which are continuously analyzing patient’s data. The company’s system alerts healthcare professionals of actionable clinical signals during critical moments in patient care, allowing for prompt intervention. Bayesian Health platform is powered by over two dozen studies resulting in precision care delivery, reduction in bias, and improvement in the quality of evaluation and reporting. Bayesian’s research-first approach created a trustworthy AI platform which transforms delivery of care. Bayesian Health deployed its sepsis detection model to five hospitals (over a 2-year period) and found that “the platform drove 1.85 hour faster antibiotic treatment for sepsis where timely treatment directly impacts mortality rate.” Additionally, the platform adoption was sustained at 89% by physicians and nurses “driven by the sensitivity and precision of the insights and user experience of the software”.
Given the tremendous demands on clinician’s time with most doctors and nurses feeling overworked and facing burnout, there is strong demand for healthcare interventions that help prioritize critical interventions. In addition, given the increased deployment of technology, physicians are often overloaded with data but not adequately trained or equipped to understand or rely on the output of AI technology so they don’t appropriately use the data or output of AI-based systems to drive clinical decision making. Bayesian Health offers one platform where the EHR system is integrated with clinical workflows, allowing for manageable and actionable alerts. This is especially noteworthy as exemplified by a recent JAMA article which found that “the Epic Sepsis Model poorly predicts sepsis” generating a large number of false positives and leading to alert fatigue among clinicians. EPIC's model was crafted from the hospital’s billing codes, while Bayesian’s model “pools data in real time from the electronic health records and other data systems [and then] stitches all the data together to create a comprehensive, longitudinal patient view. As sepsis can be fatal and is an indicator of the quality of care provided, false and overactive alerts can influence morale issues amongst healthcare providers who feel in essence that they have let patients down by allowing them to contract sepsis. Bayesian Health reported that their “technology accuracy is 10 times higher than other solutions” which will not only improve detection but will also help associated morale issues. In addition to increasing a healthcare provider’s confidence, Bayesian Health uses “cutting edge AI/ML strategies such as a wait and watch strategy and real-time feedback loops to increase precision, and strategies to make the models stronger”.
The Big Picture:
According to the book, Medical Error Reduction and Prevention, “the most common diagnostic errors that occur in primary care settings include failure to order appropriate tests, faulty interpretation, failure to follow-up, and failure to refer with one in every six patients affected and one in every 1,000 primary care visits causing preventable harm.” For example, in 2018, the Department of Human and Health Services (HHS) reported that sepsis hospitalizations cost Medicare $41.8 billion, and the costs are expected to increase 12-14% every 2 years. Sepsis models are particularly useful for hospitals to administer because it is a preventable condition; if left untreated, patients can undergo septic shock which costs more to treat. Bayesian Health provides healthcare systems with an optimal solution to promptly treat sepsis which fits into the workflow, without the possibility of alert fatigue. While the company has started by targeting sepsis it is also looking at applying its AI technology to other high priority areas for hospitals such as in-hospital deterioration and pressure injuries. As noted earlier, systems like Bayesian’s will be key to incorporating AI into the care process and transforming the delivery of care by making AI explainable and trustworthy thereby increasing clinician’s usage leading to a more direct influence on clinical decision making. Moreover, with the increasing use of sensors and technology in both at home and inpatient care, clinicians will increasingly have to rely on the computational power of AI to help in decision making. However, while AI models will undergo extensive training and testing, the human body is not a predictable system and AI models will always need human oversight and intervention. AI models should always be used to optimize and augment clinician performance, not as a replacement for their clinical assessment or skills.
Johns Hopkins Spinoff Building Risk Prediction Tools Emerges with $15M; Popular Sepsis Prediction Model Works ‘Substantially Worse’ than Claimed, Researchers Find; External Validation of a Widely Implemented Proprietary Sepsis Prediction Model in Hospitalized Patients