-
2023/10/15 AI Prediction of Sudden Cardiac Arrest in ICU Using Routinely Monitored Vital Signs
- Prof. 李建璋(Chien-Chang Lee)
- Current Position
- National Taiwan University Hospital
Biography：
Dr. Chien-Chang Lee, MD, ScD, serves as a professor of emergency medicine at National Taiwan University Hospital. Additionally, he plays a pivotal role as Deputy Director of the Center for Intelligent Healthcare at NTUH. Dr. Lee's educational journey commenced at NTU, where he completed his medical studies and later pursued an emergency medicine residency at NTU Hospital. His unyielding enthusiasm for clinical and population research led him to attain a doctorate in health data science from Harvard University. Recognizing the emergency department’s central role in diagnosing patients with acute undifferentiated symptoms, Dr. Lee is committed to enhancing patient care. His extensive experience in assessing novel laboratory tests has furthered this commitment. In the realm of Taiwan's emergency medicine, Dr. Lee's pioneering efforts introduced new biomarkers for sepsis and point-of-care multiplex PCR systems. At NTUH, Dr. Lee directs the Biomedical Data Science lab, comprising a diverse team of native students, and international students from around the world. His research spans translational diagnostic medicine, big data epidemiology and artificial intelligence in clinical contexts. Dr. Lee's significant contributions earned him recognition as a top 2% scientist worldwide by Stanford University in 2021 and 2022. His impressive body of work encompasses 250 scientific papers, over 10,000 citations, and an H-index of 54. Furthermore, he has received multiple awards for his excellence, including the Future Technology Innovation Award from Taiwan's National Science Council and the Kyorin Award from Taipei Medical Association. Dr. Lee's remarkable journey underscores his dedication to advancing medical knowledge and patient care.
Dr. Chien-Chang Lee, MD, ScD, serves as a professor of emergency medicine at National Taiwan University Hospital. Additionally, he plays a pivotal role as Deputy Director of the Center for Intelligent Healthcare at NTUH. Dr. Lee's educational journey commenced at NTU, where he completed his medical studies and later pursued an emergency medicine residency at NTU Hospital. His unyielding enthusiasm for clinical and population research led him to attain a doctorate in health data science from Harvard University. Recognizing the emergency department’s central role in diagnosing patients with acute undifferentiated symptoms, Dr. Lee is committed to enhancing patient care. His extensive experience in assessing novel laboratory tests has furthered this commitment. In the realm of Taiwan's emergency medicine, Dr. Lee's pioneering efforts introduced new biomarkers for sepsis and point-of-care multiplex PCR systems. At NTUH, Dr. Lee directs the Biomedical Data Science lab, comprising a diverse team of native students, and international students from around the world. His research spans translational diagnostic medicine, big data epidemiology and artificial intelligence in clinical contexts. Dr. Lee's significant contributions earned him recognition as a top 2% scientist worldwide by Stanford University in 2021 and 2022. His impressive body of work encompasses 250 scientific papers, over 10,000 citations, and an H-index of 54. Furthermore, he has received multiple awards for his excellence, including the Future Technology Innovation Award from Taiwan's National Science Council and the Kyorin Award from Taipei Medical Association. Dr. Lee's remarkable journey underscores his dedication to advancing medical knowledge and patient care.
Abstract：
The burden of cardiac arrest among patients within intensive care units (ICUs) is escalating on a global scale. The most effective strategy for mitigating this burden lies in the early detection of potential adverse events, allowing for timely interventions. Traditional tools such as the National Early Warning Score (NEWS) or Modified Early Warning Score (MEWS) have relied on weighted composite scores derived from routinely monitored vital signs, but their ability to predict cardiac arrest events early remains limited.
Addressing this challenge, deep learning algorithms, including recurrent neural networks and transformers, have emerged as promising tools for predicting adverse events well in advance of traditional methods. This presentation delves into our experience of harnessing novel machine learning algorithms to forecast the occurrence of cardiac arrest among ICU patients with a lead time of up to 12 hours before the event.
By leveraging the wealth of continuously collected vital sign data, our approach capitalizes on the capabilities of deep learning to identify subtle patterns and trends that precede cardiac arrest events. These algorithms exhibit the potential to recognize complex relationships within the data, enabling them to surpass the predictive capacity of traditional scoring systems. Our research highlights the critical role of early intervention in reducing the mortality and morbidity associated with cardiac arrest in the ICU.
Through case studies and empirical evidence, we showcase the feasibility and efficacy of our predictive model. The ability to anticipate cardiac arrest events holds tremendous promise for improving patient outcomes and resource allocation within ICUs. As we delve into the specifics of our deep learning approach, we invite discussion on the integration of such advanced algorithms into clinical practice and their transformative impact on critical care medicine.
The burden of cardiac arrest among patients within intensive care units (ICUs) is escalating on a global scale. The most effective strategy for mitigating this burden lies in the early detection of potential adverse events, allowing for timely interventions. Traditional tools such as the National Early Warning Score (NEWS) or Modified Early Warning Score (MEWS) have relied on weighted composite scores derived from routinely monitored vital signs, but their ability to predict cardiac arrest events early remains limited.
Addressing this challenge, deep learning algorithms, including recurrent neural networks and transformers, have emerged as promising tools for predicting adverse events well in advance of traditional methods. This presentation delves into our experience of harnessing novel machine learning algorithms to forecast the occurrence of cardiac arrest among ICU patients with a lead time of up to 12 hours before the event.
By leveraging the wealth of continuously collected vital sign data, our approach capitalizes on the capabilities of deep learning to identify subtle patterns and trends that precede cardiac arrest events. These algorithms exhibit the potential to recognize complex relationships within the data, enabling them to surpass the predictive capacity of traditional scoring systems. Our research highlights the critical role of early intervention in reducing the mortality and morbidity associated with cardiac arrest in the ICU.
Through case studies and empirical evidence, we showcase the feasibility and efficacy of our predictive model. The ability to anticipate cardiac arrest events holds tremendous promise for improving patient outcomes and resource allocation within ICUs. As we delve into the specifics of our deep learning approach, we invite discussion on the integration of such advanced algorithms into clinical practice and their transformative impact on critical care medicine.