Behind Epic’s algorithm to detect sepsis before it progresses

You’re reading the web edition of STAT Health Tech, our guide to how tech is transforming the life sciences. Sign up to get this newsletter delivered in your inbox every Tuesday and Thursday. 

How Epic’s sepsis algorithm went off the rails

EPIC late alerts

Epic Systems’ algorithm to give doctors early warning of sepsis is failing to reliably detect the condition in advance. Casey reports in a new investigation that part of the reason is that the algorithm uses active orders for antibiotics, a treatment for sepsis, as a prediction variable. While that helps improve accuracy, it also means the model is often “alerting” clinicians to something they already know. The model also uses variables such as marital status and whether ethnicity is documented in the patient’s medical record, but Casey found that Epic has not audited it for bias across demographic groups. None of this means the model can’t still be helpful, as a recent randomized study shows, but experts said it points to the need for more rigorous evaluation before models are sold to hospitals.


Using faces to screen for genetic disorders

It can take years for children with a genetic disorder to find a definitive diagnosis — an odyssey that can grow even longer for patients in low- and middle-income countries with limited access to genetic tests. So a new study in The Lancet Digital Health explored the potential for facial images to screen children for genetic disorders, training a machine learning model on a set of photos of under-21 patients. Overall, the model correctly identified images of patients with a genetic disorder in 88% of cases, which could help drive specialized care in low-resourced areas. But critically, it didn’t work equally well across populations: Accuracy was around 90% for white and Hispanic populations, and about 83% for African and Asian populations.


Access isn’t everything 


In a brief from the Office of the National Coordinator for Health Information Technology, researchers dove into how patients are currently using patient portals and apps to manage their health and health care — a relationship that may change significantly now that information blocking rules are in effect. But the results of the survey reveal that merely providing access to electronic health records won’t enable their use across the board: In 2020, 12% of those surveyed didn’t have access to a mobile device, and even among those who had downloaded a health app, 15% didn’t use it.

FDA takes a step toward AI transparency

In an era of rapid AI innovation in health care, even knowing what algorithms have been approved by regulators, and how they were evaluated, is extremely challenging. But the FDA is taking a new step to address that by publishing a database of AI devices it cleared in recent years. Despite being the body that gave the green light, the FDA compiled the database using outside sources, including a database STAT constructed during an investigation of the agency’s process for reviewing products that incorporate AI. The investigation found that approvals were based on widely divergent amounts of clinical data and without public disclosure of how they performed across patients of different genders, races, and geographies. The agency’s new database lists 343 AI products approved since 1997.

A health care ‘front door’ gets better signage

Google has updated its search and maps tools in the U.S. to make it easier to find local health care services. When a user searches for a “doctor near me,” Google’s results will now include the types of insurance accepted by nearby providers, language services available, and other information. It’s a small change that could make a big difference given the volume of people using the search engine every day to look for health care services.

A big deal on the horizon 

  • EHR company Athenahealth is approaching a sale or IPO that would value the company at $20 billion or more, according to Bloomberg. Private equity firm Veritas Capital and Elliott Investment Management, which acquired the company in 2019 for $5.7 billion, are considering options for a deal that could close by the end of the first quarter of 2022.
  • Lab automation company Opentron Labworks raised $200 million in a series C led by SoftBank, valuing the company at $1.8 billion on the back of its work using robots to speed diagnostic test processing during the pandemic.
  • Diagnostics company Cue Health completed its IPO, raising $200 million largely on the back of its portable Covid-19 test. Google reportedly uses Cue’s tests for its employees, and the companies together plan to build an “advanced respiratory biothreat detection system,” according to Cue’s prospectus.
  • Another digital mental health company that has grown during the pandemicMeru Health, raised $30 million in equity in a round led by Industry Ventures, along with $8 million in debt funded by J.P. Morgan.

Personnel file

  • The Biden administration has tapped Lisa Pino to direct the Office of Civil Rights, the part of the S. Health and Human Services Department that enforces the HIPAA privacy rule. Pino previously served as deputy commissioner of the New York State Department of Health.
  • The digital health clinic PeerWell has appointed Jonathan Slotkin, a neurosurgeon and chief medical officer of Contigo Health, to serve on its board of directors.
  • Omar Nagji, co-founder of Lyft Healthcare, has joined Memora Health as a senior vice president for growth.
Source: STAT