Algorithms for predicting the mortality in critically ill patients

A research team from Stanford University is striving to improve the life care of critically ill patients during last days of their life. Basically, the team is using an artificial intelligent algorithm which predicts the mortality of a patient. It is proved on various tests that the system accurately predicts the mortality in almost 90% of cases. This system only predicts that when a patient probably may die but it is unable to determine that how it comes to end.

The new research is going to help Doctors in decision making

To predict about the mortality of a person is quite difficult. Doctors have to consider an array of some serious factors, ranging from age, life style and family history. At time it creates a problem for exact scheduling of palliative care. Most of the time when a patient is not going to live more than a year the treatment is shifted to a palliative care team. The purpose of the palliative team is to make last days of patients as pain frees as possible.

To achieve that goal, the team takes care of patients basic life needs like loss of appetite, pain, nausea and also helps patient and his family to overcome psychological trauma. But if the patient is transferred to palliative care at a late stage they are going to miss out best of the care. But its a fact that admitting too early brings the undue burden on healthcare system. Most of the time illness of the patients end up in the critical care unit.

According to a research by Ken Jung a research scientist at Stanford medicine one of the preferences of the palliative team is to talk to patient to know about their needs and thoughts before getting to a terminally ill situation. See list of best hospital for liver cancer.

Close up of hand signing a Last Will and Testament document
Close up of hand signing a Last Will and Testament document (Credits: BrianAJackson/Getty Images under Standard License)

Jung describes that the meeds of patients were surveyed about a decade ago revealing that 80% of Americans want to die around their family in their home but only 35% do so. Though the situation is improving but it has to go a way long. To fulfil the wishes of people regarding end age it is important to have an idea of time the patient has left. Annand Avati and his team at Stanford University have developed an Algorithm system to improve the prognosis. This system is used as a tool to improve the palliative care and also helps to reduce the burden of doctors while predicting patient outcome. See how vaginal mesh is treated.

The problem lies with the palliative care are that only a small no of patients who can benefit from it can receive it. The major reason is that it is identified too late that patient now needs palliative care. Well partly the shortage of workers in palliative care is also a reason that patients lack the care they need.

The basic working principle of the AL system

The System is based basically on a form of artificial intelligence also known as deep learning. A neural data shows its findings based on large amount of data. The data is provided to the system with the help of electronic health record of patient admitted at Stanford Hospital. The researches find 200,000 cases suitable for the research after analyzing the data of two million patients. The researchers are basically using the principle of agnostic to disease stage, severity or type. They then decide accordingly that patient has to be admitted to ICU or should treated in non ICU wards. The case reports of these patients are associated with diagnosis, tests and scans numbers, detail of procedure to be used for treatment, duration of stay in hospital and medicines. See how cold and flu is treated.

The case reports of 160,000 patients were studied using deep learning algorithm which leads to mortality date. By analysis g the EHR data of a whole year and date of the given patient the mortality rate is determined that whether the patient will die in 12 months.

It is a sophisticated system specified to predict the mortality of patient within coming three to twelve months. The patients with a predicted life span of less than 3 months are not considered eligible to provide them the palliative care. It is believed that they have less time for palliative care. The algorithm is a tool to make access to palliative care a better procedure. Its basic intention is not only to predict the time of death rather a sophisticated tool to improve palliative care in patients population.

The system predicts the mortality rate with an accuracy of 90%

The remaining 40000 patients were also assessed with the new skills of this method. The average of predicting accurate life span from 3 to 12 months is accurate for 9 out 10 patients. There are about 95 percent patients who were predicted with low probability of dying were actually lived more than 12 months proving the system to be quite close to accuracy. This result in a hope among researchers that this system will now be used on a larger scale. As researchers said that this system is not only intended to calculate the death time rather to is helping the health care professionals to help the people in getting the palliative care by calculations that when they need it the most. See how to cope for anxiety and depression?

Limitations of the AL System

The researchers have also found many shortcomings in the system during the study of system. But it is believed that these do not have to be addressed before the system is applicable to use in future. According to Jung it is too difficult to have a good time and proper place for palliative care staff and doctor to have a meeting with hospital staff in a proper manner. Another reason is the detail surfaced at the time of case study is not enough and complete. Most of the data that is assumed to be available is missing from system record.

Most important is that while system prognosis the need of therapy from the given data therefore alerting the health care team for palliative care, the system is not able to justify the point that why the specific prognosis is made. Also, it cannot tell what kind of medical care, or treatment should be given to the patient. See list of top heart transplant centers in world.

AL system is proving to be a a Black Box

This is actually what scientist called a black box problem. It is the same with the AI developers at present. When a mechanical system predicts about the solution of a specific problem but cannot explain the methods of getting the results.

The physician Siddhartha Mukherjee writes in New York Times that studying and working on a specific case of a patient you find expected and sometimes unexpected patterns. So it is better to learn from the findings of the algorithm so that we should be able to get more précised results. For now we just have to do some more work on it.

According to Nigham Shah while talking to Gizmodo he certainly believes that black box model can help physicians in making good decisions regarding patient problem but only if they consider all the clinical, societal and personal context by using their human intelligence. The good news is that this system can get better by the use of more sophisticated models and having diverse data. The system predicts only about when we might die, but it’s a good thing to provide better end of life care.

2 thoughts on “Algorithms for predicting the mortality in critically ill patients”

Comments are closed.