Google's newly developed AI algorithm predicts people's death time, according to researchers, with a prediction accuracy of 95%.
The study looked at a range of clinical issues in hospitalized patients and was recently published in the Nature Collaborative Journal (npj) Digital Medicine. Google applies AI technology to a large amount of data from 21,600 patients in two medical centers, each of whom stayed in the hospital for at least 24 hours.
The data used in the study came from the patient's electronic medical record
The researchers explained in a magazine article: "We want to know if deep learning can provide effective predictions in the face of a wide range of clinical problems. So we have selected predictive projects in a wide range of clinical areas, including one Important clinical outcomes—death, a standard medical quality metric—are re-admitted, a resource utilization metric—the length of hospital stay, and an indicator of the patient's condition—the diagnosis.
The results of this proof-of-concept study show that the algorithm can accurately predict the patient's risk of death, whether it will be re-admitted, whether it will prolong hospital stay, and the patient's discharge diagnosis. Moreover, in each of the above cases, the prediction of this new algorithm is more accurate than the previously published model.
According to the study, the accuracy of patient mortality predictions using the data from the University of California San Francisco Medical System was 95%, and the accuracy of death predictions using data from the University of Chicago Medical System was 93%.
This accuracy is more accurate than the traditional “augmented Early Warning Score” predictive model, which uses a variety of factors to help doctors determine the patient's condition. According to the research report, the accuracy of the traditional method at the University of California, San Francisco is 85%, and the accuracy of the medical system at the University of Chicago is 83%.
The current emergence of Google's research coincides with a heated discussion about the potential benefits and risks of using AI technology. From cybersecurity risks and AI, perhaps the so-called “doomsday machine” that can bring disasters, to the potential drivers of AI technology for economic growth, experts are weighing the long-term impact of AI technology in every respect.
Because of the large amount of information that is relied upon, healthcare is increasingly seen as a suitable area for applying AI technology. FDA Commissioner Scott Gottlieb discussed the prospects of AI technology in the medical industry in a speech earlier this year.
This technology also faces enormous challenges
On Tuesday, on a Fox TV show, Dr. Mikhail Varshavski, a family medicine doctor, said: "It is true that connecting a large amount of medical information may help patients, but data privacy is still key. I am worried about who is using it. How do you use this data? As a doctor, I want to use this data to help patients, not just to make money."
"The machine also makes mistakes. Sometimes the wrong data can cause the machine to make mistakes, so we still need to supervise the process," Dr. Varshavski added.
Even within Google, the use of AI is controversial. Recently, Google quit a military project called Maven, which aims to improve the accuracy of the drone's aiming. The Maven project has always been the focus of Google's internal debate. In April of this year, more than 3,100 Google employees collectively signed the letter company CEO Sundar Pichai, hoping to withdraw from the project.