Home
|
Inbox
|
Search
|
View Abstract
Biosignal Based Pattern Predicts Mortality in Critically Ill Children
Description
.abstract img { width:300px !important; height:auto; display:block; text-align:center; margin-top:10px } .abstract { overflow-x:scroll } .abstract table { width:100%; display:block; border:hidden; border-collapse: collapse; margin-top:10px } .abstract td, th { border-top: 1px solid #ddd; padding: 4px 8px; } .abstract tbody tr:nth-child(even) td { background-color: #efefef; } .abstract a { overflow-wrap: break-word; word-wrap: break-word; }
A4283 - Biosignal Based Pattern Predicts Mortality in Critically Ill Children
Author Block: S. Kim1, S. Lee2, S. Choi1, I. Sol1, Y. Kim1, H. Kang2, S. Lee3, K. Kim1, M. Sohn1, K. Kim1; 1Department of Pediatrics, Institute of Allergy, Yonsei University College of Medicine, Seoul, Korea, Republic of, 2Research Center in DS-eTrade Inc., Seoul, Korea, Republic of, 3Yonsei University Health System, Seoul, Korea, Republic of.
RATIONALE: Physiologic data are routinely gathered in intensive care unit (ICU), however, their use for potential risk assessment or long term morbidity prediction has been limited. We applied machine learning algorithms for model development and validation using biosignal data to identify early subclinical signatures of in-hospital mortality in critically ill children. METHODS: In total, 453,778 physiologic monitoring data and clinical information collected from the first week of 1,729 ICU admissions of patients under age 19 from year 2011 to 2016 were analyzed. We adopted machine learning algorithms, k-Means and Decision Tree, to discover distinct biosignal based patterns and develop a mortality prediction model. RESULTS: Overall mortality rate was 21.0%. We categorized patients into three age groups: 0-12 months old, 1-6 years old, and above 7 years old, and identified five biosignal based patterns in each age group. Association analysis revealed these patterns had significant associations with mortality. Specifically, three patterns turned out to be high-risk patterns (aOR [95% CI], 2.299 [1.088-4.632], 2.623 [1.568-4.637], and 4.641 [2.491-8.561], respectively). Accuracy of mortality prediction model was 0.860, 0.864, and 0.812 for each age group, which were similar or superior than Pediatric Risk of Mortality III (accuracy of 0.794).CONCLUSIONS: Data-driven machine learning algorithm discovered specific physiologic signatures associated with mortality, and provided a reasonable prediction model. This methodology can be used as a risk assessment tool in the ICU which may allow clinicians to make an early decision for their patients’ upcoming illness.
Author Block: S. Kim1, S. Lee2, S. Choi1, I. Sol1, Y. Kim1, H. Kang2, S. Lee3, K. Kim1, M. Sohn1, K. Kim1; 1Department of Pediatrics, Institute of Allergy, Yonsei University College of Medicine, Seoul, Korea, Republic of, 2Research Center in DS-eTrade Inc., Seoul, Korea, Republic of, 3Yonsei University Health System, Seoul, Korea, Republic of.
RATIONALE: Physiologic data are routinely gathered in intensive care unit (ICU), however, their use for potential risk assessment or long term morbidity prediction has been limited. We applied machine learning algorithms for model development and validation using biosignal data to identify early subclinical signatures of in-hospital mortality in critically ill children. METHODS: In total, 453,778 physiologic monitoring data and clinical information collected from the first week of 1,729 ICU admissions of patients under age 19 from year 2011 to 2016 were analyzed. We adopted machine learning algorithms, k-Means and Decision Tree, to discover distinct biosignal based patterns and develop a mortality prediction model. RESULTS: Overall mortality rate was 21.0%. We categorized patients into three age groups: 0-12 months old, 1-6 years old, and above 7 years old, and identified five biosignal based patterns in each age group. Association analysis revealed these patterns had significant associations with mortality. Specifically, three patterns turned out to be high-risk patterns (aOR [95% CI], 2.299 [1.088-4.632], 2.623 [1.568-4.637], and 4.641 [2.491-8.561], respectively). Accuracy of mortality prediction model was 0.860, 0.864, and 0.812 for each age group, which were similar or superior than Pediatric Risk of Mortality III (accuracy of 0.794).CONCLUSIONS: Data-driven machine learning algorithm discovered specific physiologic signatures associated with mortality, and provided a reasonable prediction model. This methodology can be used as a risk assessment tool in the ICU which may allow clinicians to make an early decision for their patients’ upcoming illness.
|
Home
|
Inbox
|
Search
|