What If Algorithms Could Sense Death?
When it comes to the internet, we are all fairly aware of how companies such as Facebook use algorithms to filter and select the type of posts we see on our news feeds, based on tracking our internet browsing and search history in order to provide targeted information. Sheer demand leads to ever-increasing investment and innovation into the technology as we generate more data than ever before with almost a gluttonous, insatiable appetite that poor, fallible humans are finding increasingly difficult to satisfy. This is where artificial intelligence has come to the rescue, helping to make sense of the innumerable data and our activities through identifying trends and patterns and policing content.
Algorithms that can detect death?
However, what about if algorithms could not just identify trivial matters such as the likelihood of you wanting to see a quiz on what type of bread you most identify with or what Amazon purchase you will make next, but actually predict the rather weighty issue of death? Well, it may very well be possible.
Predicting death through algorithms
At the end of 2016, a student from Stanford’s computer-science department, Anand Avati as well as a team from the university’s medical school put creating a ‘dying algorithm’ to the test could anticipate which patients would die within the next three to 12 months.
How did the team do it? The key determinant was using the hospital’s medical records, through analyzing the medical information prior to the three-to 12 month period and trying to anticipate the way in which you could teach an algorithm to make a prediction of imminent death (Source: NY Times) . The algorithm was then applied via a trial to over 40,000 living patients. The results were, depending on how you view using technology to predict death, either amazingly accurate or completely terrifying. The algorithm managed to successfully predict the deaths of nine out of ten patients, who were anticipated to die within three to 12 months, whilst also being able to correctly foresee 95% of patients assigned with a low likelihood of death continuing to live past the 12-month window. Given that this was a trial, and with the intention of inputting additional data in the future, such as the doctor’s assessment, scan results or lab values which will improve the accuracy of the algorithm even further, the results are impressive.
Avati and the team used the hospital record’s of over 160,000 deceased patients, absorbing this information that had already been coded by the hospital doctors, including the following:
- The number of medical prescriptions written for each patient
- How many scans had been ordered
- How many days the patients had spent in hospital
- The patient’s diagnosis
- The types of procedures the patient had
This information was then put into software, titled a ‘deep neural network’ named due to the way in which it works in a similar way to how brain neurons do. The algorithm would then take this data to create a prediction and a score of the likelihood of a patient dying within the next 12 month period.
Why create a ‘dying algorithm’?
You may well feel that the creation of a dying algorithm feels a particularly morbid way to use this type of technology, however, Avati and his team believe that such identification of an algorithm could be a breakthrough when it comes to providing palliative care. The benefits include:
- Identifying accurately whether patients were likely to die within the next 12 months meant that medical resources could be used appropriately and prevent unnecessary allocation of resources too soon
- The ‘dying algorithm’ could also prevent resources being given too late too, as anticipating whether death would be likely to occur with three to 12 months gives medical professionals enough time to thoroughly prepare for a patient’s death and give the appropriate care
- By accurately predicting a terminally ill patient’s death, and allocating resources accordingly, hospitals could target their palliative-care resources that could save hospitals a considerable amount of money
- If the algorithm could reliably predict deaths, this could also help palliative-care teams productivity levels and help to use their time more effectively. This is because such care teams wouldn’t have to manually look at patients individual charts and decide who would be most likely to benefit from care. In the UK the end-of-life nursing charity Marie Curie reported in 2017 that 57,000 terminally ill patients or those progressing to such a stage are not receiving the appropriate end-of-life care support that they need. Furthermore, it has also been reported that less than a third (28 percent) of NHS clinical groups were hitting the targets for fast-track healthcare support within a 48 hour period. Taking into account these statistics, a ‘dying algorithm’ could be advantageous
- The algorithm could also help doctors to use medical interventions in a more effective, humane manner
- Accurate algorithms could radically change how we approach palliative care and may lead to patients dealing with death differently if they have an exact quantitative prognosis, which can make decision making easier and with less ambiguity involved in doing so
The disadvantages of using a ‘death algorithm’
Whilst the trial has clearly shown to yield accurate results about predicting death, it is not able to tell you how it has been able to determine why is patient is likely to die. The algorithm is unable to express the reasoning behind its prediction, which means that it cannot help doctors determine why a patient will die or understand the patterns of mortality.
Follow our website for more information about prepaid funeral plans for you or your loved ones.
In addition, critics have put forward their concerns on relying heavily on technology rather than doctors to determine predictions of death. Others are concerned we may lose the humanity inherent in dealing with palliative care if we let algorithms lead the way when it comes to death and dying, which could be a huge disadvantage given the emotional and sensitive nature of any conversations regarding death.