Taking into account the wishes of patients at the end of life with artificial intelligence
Is it possible to develop an automated tool that would alert physicians when hospitalized patients have a high probability of dying within the next 12 months? This is the task that Dr. Ryeyan Taseen is working on. He is a respirologist who did his residency at the CIUSSS de l’Estrie-CHUS and a master’s student in health informatics at the GRIIS at the Université de Sherbrooke. Under the direction of Profs. Jean-François Ethier and Luc Lavoie, he is developing a tool that would motivate physicians to start a discussion with their patients about the care they wish – or do not wish – to receive.
To complete this research project, Dr. Taseen was awarded a joint grant from the Fonds de recherche du Québec – Santé and the ministère de la Santé et des Services sociaux. This grant will allow him to train in cutting-edge research while continuing his profession as a physician, and to aspire to an academic career as a clinician-scientist.
Dr. Taseen recounts his background: “I grew up with computers. I’m from the social media generation. I take it for granted that IT can benefit our daily routines. When I entered the medical world, I was shocked by the fact that we still use paper! Doctors have to write the same information over and over again!”
He decided early on not to be discouraged by archaic health systems. He would help propel health into the computerized 21st century.
A visionary and realistic scenario
To understand Dr. Taseen’s proposed project, let’s take the fictive case of Jean-Pierre. Jean-Pierre is 72 years old and has heart failure along with a chronic lung disease requiring him to breathe bottled oxygen at all times. He has just been hospitalized with pneumonia and acute kidney failure.
Jean-Pierre’s doctor opens the medical file and enters routine data concerning his patient’s state of health. Automatically, he receives an alert: the computer tool warns him that the probability of Jean-Pierre’s death within the next year is high.
Jean-Pierre and his doctor had only partially acknowledged the extent of his prognosis and the consequences of his state of health. After all, Jean-Pierre’s dependence on oxygen is a sign of the advanced stage of his disease. Rather than continuing his patient rounds as he had planned, the doctor turns his attention to the alert he has just received. He analyzes it in detail.
He decides to give priority to Jean-Pierre’s situation and to have a discussion with him about the goals of care. The doctor thus discusses with Jean-Pierre his prognosis and the care he would prefer to receive in the event of his health deteriorating over the next few months.
Jean-Pierre is certain he doesn’t want to be resuscitated or intubated. He’s reassured that the health care team is respecting his preferences. He also specifies that he only wants to be treated for maximum comfort and that he does not want invasive treatments, even if he is not sure what this means. The doctor suggests that Jean-Pierre meet with the palliative care team. Jean-Pierre agrees.
Dr. Ryeyan Taseen argues in his research work that this scenario is realistic. “Patients who are at a high risk of mortality are not benefitting from a goals-of-care discussion as often as they should be, explains Dr. Taseen. Doctors are faced with organizational pressures; they have to finish patient rounds, review results and write notes. These constraints result in fewer opportunities for goals-of-care discussions. Patients receive care that they may not have wanted had they been included in the care-planning process sooner.”
One of the most common cases is that of patients receiving treatments that are excessive relative to the improvement that could be expected. In these cases, a last-minute discussion focuses on futile care when it should have occurred sooner and concentrated on the patient’s care preferences.
Discussing goals of care to improve quality of life
The decision-support and alerting tool Dr. Taseen is developing will be compatible with data from the CIUSSS de l’Estrie–CHUS and will take advantage of artificial intelligence.
“This decision-support tool,” says Dr. Taseen, “is based on the patient’s health data at the time of admission. It tells physicians that the risk of mortality is high and suggests that they have a discussion about goals of care. It has to be automated for it to succeed.”
In his master’s thesis, Dr. Taseen took on the challenge of developing a predictive model that uses real-time health data available at the CIUSSS de l’Estrie–CHUS. This computerized model uses the random forest method, which is a type of machine learning in the field of artificial intelligence.
With this method, predictions will be more precise than those produced by conventional approaches. One of the challenges of artificial intelligence is to deliver results that are interpretable, rather than opaque, as if they came out of a “black box”. With the method proposed by Dr. Taseen, the tool will produce an alert that is accessible to clinical reasoning.
In the next phase of his project, Dr. Taseen plans to deploy his automated prediction tool in a hospital setting. This will be an opportunity to conduct a concrete test of the model’s promising clinical impact.