Thursday, March 28, 2024

Algorithms for predicting fatal infections are often flawed

Must Read

A complication The infection called sepsis is Number one killer In the American hospital. Therefore, it is not surprising that more than 100 health systems use the early warning system provided by Epic Systems, a major supplier of electronic health records in the United States. The system will issue an alert based on a proprietary formula, tirelessly observing signs of illness in the patient’s test results.

But a new study using data from nearly 30,000 patients at the University of Michigan Hospital showed that Epic’s system performed poorly. The author said it missed two-thirds of sepsis cases, rarely found cases that medical staff did not notice, and frequently sent false alarms.

Karandeep Singh, an assistant professor at the University of Michigan who led the research, said these findings illustrate a broader problem with patents. algorithm It is increasingly used in medical care. “They are very widely used, but there are very few publications about these models,” Singh said. “For me, this is too shocking.”

The research was published on Monday JAMA Internal MedicineAn Epic spokesperson disputed the conclusions of the study, saying that the company’s system “helped clinicians save thousands of lives.”

Epic is not the first widely used health algorithm. It raises concerns that the technology that should improve healthcare is not provided or even harmful at all. In 2019, a system for millions of patients was discovered to prioritize special care for people with complex needs Underestimating the needs of black patients Compared with white patients.that Prompted some Democratic senators Require federal regulators to investigate biases in health algorithms.A kind Learn A study published in April found that statistical models used to predict suicide risk in mental health patients performed well for white and Asian patients, but did not perform well for black patients.

The way that sepsis sneaked into the hospital ward made it a special target for medical staff’s algorithm assistance. guide Health service providers from the Centers for Disease Control and Prevention to sepsis are encouraged to use electronic medical records for monitoring and prediction. Epic has several competitors that provide commercial warning systems, and some American research hospitals have Build your own tools.

Singh said the automatic sepsis warning has great potential because the key symptoms of the disease, such as low blood pressure, may have other causes that make it difficult for staff to detect early.Start sepsis treatment such as antibiotics one hour early Have a lot to do For the survival of the patient.Hospital administrators are often particularly interested in the sepsis response, partly because it helps U.S. Government Hospital Rating.

Singer runs a laboratory in Michigan, researching Machine learning To patient care. After being asked to serve as the chair of the university’s health system committee, he became curious about Epic’s sepsis warning system, which aims to oversee the use of machine learning.

As Singh learned more about the tools used in Michigan and other health systems, he began to worry that most of these tools came from vendors who rarely disclosed their work or how they performed it. His own system was licensed to use Epic’s sepsis prediction model, and the company told customers that the model was very accurate. However, its performance has not been independently verified.

Singh and colleagues in Michigan tested Epic’s predictive model on records of nearly 30,000 patients in 2018 and 2019, who covered nearly 40,000 hospitalizations. The researchers pointed out that Epic’s algorithm marks the frequency of people with sepsis as defined by the CDC and the Centers for Medicare and Medicaid Services. They compared the alarms triggered by the system with the sepsis treatments recorded by the staff, and they did not see the Epic sepsis alarms for the patients included in the study.

The researchers stated that their findings indicate that Epic’s system will not make hospitals better at infectious sepsis, and may cause unnecessary alarms to staff. The company’s algorithm failed to identify two-thirds of the approximately 2,500 sepsis cases in Michigan data. It will remind 183 patients who have sepsis but have not received timely treatment by staff.


Source link

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Latest News

Amid legal woes, Brazil’s Bolsonaro seeks passport return for Israel trip | Jair Bolsonaro News

Brazil’s embattled former President Jair Bolsonaro has requested the return of his passport in order to visit Israel,...

More Articles Like This