Can AI be Better than Doctors in Intensive Care Units?
Artificial intelligence (AI) is revolutionizing the field of medicine, and its potential impact is becoming evident in intensive care units (ICUs). At TU Wien in Vienna, researchers have developed an AI system that provides treatment suggestions for patients requiring intensive care due to sepsis, a life-threatening condition. By analyzing extensive data from various hospital ICUs, the AI system has already demonstrated its ability to surpass the quality of human decisions in treatment recommendations.
The Power of Data
ICUs gather a wealth of data around the clock to monitor patients’ conditions closely. The research team aimed to explore how this data could be better utilized. Traditional medical decision-making relies on well-founded rules and parameters. However, AI can consider a multitude of factors beyond human capacity, leading to potentially superior decisions.
Reinforcement Learning and Temporal Progression
The researchers utilized a form of machine learning called reinforcement learning. Unlike simple categorization tasks, this approach focuses on temporal progression, predicting the likely development a patient will experience. The AI system acts as an agent that makes decisions, aiming to maximize its “reward” when the patient’s condition improves and avoiding “punishment” when deterioration or death occurs. By using extensive medical data, the system autonomously determines strategies with a high probability of success.
The impact of the AI system on sepsis treatment is significant. Sepsis is a leading cause of death in intensive care medicine, emphasizing the need for effective early detection and treatment. The AI system has demonstrated higher cure rates compared to purely human decisions. In one study, the 90-day mortality rate increased by approximately 3% to reach about 88%. While medical decisions should not be solely entrusted to AI, it can serve as a valuable additional tool for medical staff, enabling them to consult and compare their assessments with the AI system’s suggestions.
Despite the promising outcomes, implementing AI in clinical practice raises important legal questions. Liability for mistakes made by AI systems and the potential consequences of disregarding AI recommendations must be addressed. A comprehensive discussion about the social framework and clear legal rules is essential to harness the full potential of AI in healthcare.
The research project at TU Wien highlights the success of AI in intensive care units. While technological capabilities continue to advance, it is crucial to establish guidelines and regulations to ensure responsible and effective integration of AI systems in medical decision-making processes.
Source: Razvan Bologheanu, Lorenz Kapral, Daniel Laxar, Mathias Maleczek, Christoph Dibiasi, Sebastian Zeiner, Asan Agibetov, Ari Ercole, Patrick Thoral, Paul Elbers, Clemens Heitzinger, Oliver Kimberger. Development of a Reinforcement Learning Algorithm to Optimize Corticosteroid Therapy in Critically Ill Patients with Sepsis. Journal of Clinical Medicine, 2023; 12 (4): 1513 DOI: 10.3390/jcm12041513