AI determines the individual time of death

Artificial intelligence can predict the time of death

Researchers have now developed a method to calculate the exact time of death of patients. An artificial intelligence is able to analyze important medical records and electronic health records, allowing accurate prediction of the remaining life expectancy.

Stanford University scientists have developed an artificial intelligence that can predict the death of patients with cancer and other incurable diseases. The so-called deep learning system could in the future lead to pioneering developments in palliative medicine. The experts published the results of their study on the document server for Preprints "Arxiv".

Many people die every year from the effects of cancer or other incurable diseases. An artificial intelligence is now able to predict the time of death of such sick people. (Image: mdennah / fotolia.com)

New development could improve palliative care in the future

Stanford University researchers have tested a new artificial intelligence algorithm to help hospitals improve palliative care for cancer patients and those with incurable diseases.

The algorithm, which is based on a deep neural network learning machine, can analyze important medical records or electronic health records of terminally ill patients. It can then be calculated whether the patients benefit more from a so-called end-of-life care or palliative care.

Predictions have an accuracy of three to twelve months

The algorithm can predict the mortality of patients with an accuracy of three to twelve months, and on the basis of this prediction affected patients can be referred to a palliative care.

The predictions could allow hospital physicians and caregivers to proactively address such patients instead of relying on referrals from treating physicians or conducting time-consuming examinations.

The wishes of incurably ill people are rarely discussed

Previous studies have already shown that around 80 percent of Americans want to spend their last days at home. However, only 20 percent of those affected are capable and many die in hospitals. In fact, terminally ill patients often receive aggressive medical care in their last days, rather than their needs being met at the end of their lives, the experts explain.

These problems are common in hospitals in palliative care

In recent years, the ability of hospitals to provide palliative care has improved. However, only seven to eight percent of patients receive such care, the doctors say. The lack of palliative care professionals analyzing all the patient's data and the frequent over-optimism of physicians in predicting the course of the disease are points that contribute to this problem. This is where the so-called deep learning AI algorithm comes into play.

Predictive model of mortality is based on the analysis of large amounts of data

Based on the amount of data available, we were able to produce a predictive model of all-cause mortality, explain the researchers. The learning technique known as the deep learning algorithm uses neural networks to filter and analyze a large amount of data.

The prediction of mortality is independent of the type of disease, the age of the patients and other factors. The algorithm used the data from the previous year's patients from the first contact to determine their mortality within twelve months.

The datasets of two million people were analyzed for the study

For the study, the researchers analyzed two million records of adults and children who were admitted to the Stanford Hospital and the Lucile Packard Children's Hospital. The physicians thus identified potential 200,000 patients for their study. Participants' electronic medical records were then analyzed by the system to predict their mortality.

The algorithm was then to predict the mortality of 160,000 patients within 12 months of a given date. The system was able to improve and learn. Finally, the algorithm was able to predict the mortality of patients within the next three to twelve months.

Mortality data were accurate in nine out of ten cases

The algorithm then evaluated the data of the remaining 40,000 patients. He was able to accurately predict mortality over a period of three to twelve months in nine out of ten cases. It should be ensured that the most seriously ill patients get a chance to discuss with their families how and where they would like to spend their last days before they become so seriously ill that they need to be admitted to intensive care, say the authors. (As)