A team at OSF HealthCare in Illinois is using artificial intelligence (AI) to assist doctors in determining which patients have a higher likelihood of dying during their hospital stay. The primary aim is to initiate critical end-of-life discussions with these patients. Research has shown that only 22% of Americans have written down their end-of-life preferences, making it challenging to understand the care they desire when faced with critical medical decisions. This AI model seeks to address this gap by predicting patient outcomes.
How the AI Model Works
The team developed an AI model that predicts a patient’s risk of death within a window of five to 90 days after admission to the hospital. This prediction is based on various patient-related factors and data points.
Early Planning for End-of-Life Care
One of the key motivations behind this initiative is the realization that, in cases where patients become unconscious or rely on life-supporting measures like ventilators, they may be unable to convey their preferences for medical care. Early planning and discussions can prevent situations where patients miss out on the appropriate end-of-life care they might have chosen if their wishes had been documented earlier. To ensure a sense of urgency and relevance, the AI model begins predicting patient outcomes from the fifth day of their hospital stay and continues for up to 90 days.
Research Findings
The AI model’s effectiveness was tested using data from over 75,000 patients from diverse backgrounds. The results, published in the Journal of Medical Systems, revealed that while the mortality rate among all patients was approximately one in 12, for those identified as high-risk by the AI model, the mortality rate increased significantly to one in four, which is three times higher than the average.
Model Training and Information Used
The AI model was trained on 13 different types of patient information, including clinical trends, organ function, healthcare visit frequency, and patient age. It then utilizes this information to make predictions about a patient’s likelihood of passing away within the specified timeframe. The AI model not only provides a probability or “confidence level” but also offers explanations as to why a patient may have a higher risk of death. This data equips clinicians with valuable insights and helps streamline their decision-making process.
Inspiration from NYU Langone
The OSF researchers drew inspiration from a similar AI model developed at NYU Langone. While their patient population differed, they successfully adapted the AI model to meet their performance goals. Dr. Jonathan Handler, lead study author, acknowledges that the AI model isn’t flawless. Just because it identifies an increased risk of mortality doesn’t guarantee it will happen. Nevertheless, its primary purpose is to encourage clinicians to initiate end-of-life discussions.
The Human Touch in Healthcare
While AI can be a valuable tool, it’s essential to maintain a balance between technology and human interaction in healthcare, especially for sensitive conversations about end-of-life care. Continuous monitoring and ethical considerations are crucial to ensure AI’s responsible use in healthcare, particularly when it involves life and death predictions.
