Anand Avati, professor of computer science at Wu Enda and Stanford University, Kenneth Jung, Lance Downing and Nigam H., Stanford University Center for Biomedical Informatics. Shah, and Stanford University School of Medicine Stephanie Harmon, a team of six Stanford scientists, are investigating how artificial intelligence can be used to predict human deaths, thereby improving the level of palliative care, or providing specialized treatment for patients with severe illnesses. Care.
Research shows that about 80% of Americans want to spend the last hours of their lives in their own homes, but only 20%. In fact, more than 60% of deaths occur in the hospital's emergency ward, and patients receive invasive treatment for the last time before dying.
Hospitals that can provide palliative care have been increasing over the past 10 years. In 2008, 53% of all hospitals in the United States had palliative care teams, and in 2015 this proportion has climbed to 67%. Although there are more and more hospitals that can provide palliative care, according to the National Palliative Care Registry, among all patients requiring palliative care (7% to 8% of all inpatients), only Less than half of the people actually accepted this treatment.
This is closely related to the doctor's often optimistic about the patient's survival time. In addition, the relevant nursing staff and resources for palliative care are also limited. Therefore, in order to help as many patients as possible for this kind of comfort therapy, Stanford's research team hopes to use artificial intelligence technology to find objects with only three to twelve months remaining.
The basis for this time period is that if the patient will die within three months, the palliative care team will not have enough time to prepare. However, if the patient will die after twelve months, it is difficult to accurately predict the specific time of death.
In the past, each case list was examined by a doctor to determine if the patient was eligible for palliative care. But the whole process is time consuming and the doctor's personal bias may have an impact on the final care decision.
In response, the report states: "The results of this prediction will help the palliative care team to reach out to these patients and use the deep learning techniques to provide objective treatment recommendations based on the patient's EHR (ie, electronic health record) rather than relying on the referral of the attending physician. Or take the time to study all patient cases."
Specifically, the algorithm automatically evaluates the EHR data of inpatients and helps the palliative care team determine which patients may need palliative care. In fact, it is a neural network trained with the patient's previous HER data.
The report describes several types of methods that can make patients' prognostic information (prognosis refers to predicting the likely course and outcome of the disease) more objective and intelligent, including prognostic methods for palliative care, prognostic methods for intensive care unit ICU, early Identify prognostic methods and detail prognostic methods in the era of big data.
In an interview with CNBC, Shah said that although the use of AI technology may still result in some patients who should be treated without the successful completion of the application, the actual effect is better than manual analysis.
“At the moment, we missed most patients who should undergo palliative care because clinicians are too optimistic about the survival time... Only less than 1% of patients can receive palliative care for more than six months before they die. With this in mind, although AI-assisted methods will inevitably miss half of eligible patients, their effectiveness is far superior to the current situation."
To conduct the study, the team used 2 million adult and child electronic medical records from Stanford Hospital and Lucille Packard Children's Hospital as data samples.
Of course, Avati also stressed: "The predictions of this model are only used to recommend partial eligible patients when the palliative care team conducts case reviews (and automatic referrals). Human doctors are still responsible for the lead of the entire review process. The results obtained from the project are only used as a reference for palliative care conditions, not a direct prediction of the time of death."
As a way of judging advanced disease, death prediction can help identify eligible candidates. But it is important to emphasize that palliative care and end-of-life care are not the same thing.
Harman said in an interview: "Pulmonary bone transplantation (healing treatment) in hospitals often have to face some serious side effects, such as severe pain caused by treatment programs. For such patients, doctors tend to Take palliative care to relieve side effects and help patients complete the treatment process."
At the same time, the report also found that time of death is one of the effective indicators of directionality. For example, for patients who were predicted by AI to have a 90% chance of dying within three to twelve months, the team randomly selected 50 of them for manual review. The results showed that all 50 patients were “suitable for referralâ€. In other words, the effect of the AI ​​solution is completely in line with expectations.
Non Contact Thermometer Gun,Infrared Thermometer,Safety Infrared Thermometer
Dongguan Keyutai Mask Co., Ltd. , https://www.maskkytai.com