Sujay Kulshrestha1, Dmitriy Dligach2, Cara Joyce3, Marshall S Baker4, Richard Gonzalez5, Ann P O'Rourke6, Joshua M Glazer7, Anne Stey8, Jacqueline M Kruser9, Matthew M Churpek10, Majid Afshar11. 1. Burn and Shock Trauma Research Institute, Loyola University Chicago, CTRE Building 115, Room 315, 2160 South 1st Avenue, Maywood, IL, United States; Department of Surgery, Loyola University Medical Center, EMS Building 110, Room 3210, 2160 South 1st Avenue, Maywood, IL, United States. Electronic address: sujay.kulshrestha@lumc.edu. 2. Center for Health Outcomes and Informatics Research, Health Sciences Division, Loyola University Chicago, CTRE Building 115, Room 126, 2160 South 1st Avenue, Maywood, IL, United States; Department of Public Health Sciences, Stritch School of Medicine, Loyola University Chicago, 2160 South 1st Avenue, Maywood, IL, United States; Department of Computer Science, Loyola University Chicago, 1052 West Loyola Avenue, Chicago, IL, United States. 3. Center for Health Outcomes and Informatics Research, Health Sciences Division, Loyola University Chicago, CTRE Building 115, Room 126, 2160 South 1st Avenue, Maywood, IL, United States; Department of Public Health Sciences, Stritch School of Medicine, Loyola University Chicago, 2160 South 1st Avenue, Maywood, IL, United States. 4. Department of Surgery, Loyola University Medical Center, EMS Building 110, Room 3210, 2160 South 1st Avenue, Maywood, IL, United States; Veterans Affairs Hospital, 5000 South Fifth Avenue, Hines, IL, United States. 5. Burn and Shock Trauma Research Institute, Loyola University Chicago, CTRE Building 115, Room 315, 2160 South 1st Avenue, Maywood, IL, United States; Department of Surgery, Loyola University Medical Center, EMS Building 110, Room 3210, 2160 South 1st Avenue, Maywood, IL, United States. 6. Department of Surgery, University of Wisconsin, 600 Highland Avenue, MC 3236, Madison, WI, United States. 7. Department of Emergency Medicine, University of Wisconsin, 800 University Bay Drive, Suite 310, MC 9123, Madison, WI, United States. 8. Division of Trauma and Surgical Critical Care, Department of Surgery, Northwestern University, 76 North St. Clair Street, Suite 650, Chicago, IL, United States. 9. Division of Pulmonary and Critical Care, Department of Medicine, Northwestern University, 633 North St. Clair Street, 20th Floor, McGaw M-335, Chicago, IL, United States; Department of Medical Social Sciences, Northwestern University, 633 North St. Clair Street, 19th Floor, Chicago, IL, United States. 10. Department of Medicine, University of Wisconsin, 8007 Excelsior Drive, Madison, WI, United States. 11. Center for Health Outcomes and Informatics Research, Health Sciences Division, Loyola University Chicago, CTRE Building 115, Room 126, 2160 South 1st Avenue, Maywood, IL, United States; Department of Health Informatics and Data Science, Loyola University Chicago, 2160 South First Avenue, Maywood, IL, United States.
Abstract
INTRODUCTION: Trauma injury severity scores are currently calculated retrospectively from the electronic health record (EHR) using manual annotation by certified trauma coders. Natural language processing (NLP) of clinical documents in the EHR may enable automated injury scoring. We hypothesize that NLP with machine learning can discriminate between cases of severe and non-severe injury to the thorax after trauma. METHODS: Clinical documents from a trauma center were examined between 2014 and 2018. Severe chest injury was defined as a thorax abbreviated injury score (AIS) >2 and served as the reference standard for supervised learning. Free text unigrams and concept unique identifiers (CUIs) from the Unified Medical Language Systems (UMLS) were extracted from clinical documents collected at one hour, four hours, and eight hours after patient arrival to the emergency department. Logistic regression models with elastic net regularization were tuned to maximize area under the receiver operating characteristic curve (AUROC) using 10-fold cross-validation on the training dataset (80%) and tested on a hold-out 20% dataset. RESULTS: There were 6,891 traumas that met inclusion criteria. The complete data corpus consisted of 473,694 documents. Models trained using the first hour of data had a mean AUROC of 0.88 (95%CI [0.86, 0.89]); model discrimination and reclassification from the first hour significantly improved after eight hours with a mean AUROC of 0.94 (95%CI [0.93, 0.95]). Performance of models using CUIs were similar to unigrams (p>0.05). Models demonstrated excellent clinical face validity. CONCLUSIONS: Both CUIs and unigrams demonstrated excellent discrimination in predicting severity of chest injury using the first eight hours of clinical documents. Our model demonstrates that automated anatomical injury scoring is feasible and may be used for aggregation of data for trauma research and quality programs.
INTRODUCTION: Trauma injury severity scores are currently calculated retrospectively from the electronic health record (EHR) using manual annotation by certified trauma coders. Natural language processing (NLP) of clinical documents in the EHR may enable automated injury scoring. We hypothesize that NLP with machine learning can discriminate between cases of severe and non-severe injury to the thorax after trauma. METHODS: Clinical documents from a trauma center were examined between 2014 and 2018. Severe chest injury was defined as a thorax abbreviated injury score (AIS) >2 and served as the reference standard for supervised learning. Free text unigrams and concept unique identifiers (CUIs) from the Unified Medical Language Systems (UMLS) were extracted from clinical documents collected at one hour, four hours, and eight hours after patient arrival to the emergency department. Logistic regression models with elastic net regularization were tuned to maximize area under the receiver operating characteristic curve (AUROC) using 10-fold cross-validation on the training dataset (80%) and tested on a hold-out 20% dataset. RESULTS: There were 6,891 traumas that met inclusion criteria. The complete data corpus consisted of 473,694 documents. Models trained using the first hour of data had a mean AUROC of 0.88 (95%CI [0.86, 0.89]); model discrimination and reclassification from the first hour significantly improved after eight hours with a mean AUROC of 0.94 (95%CI [0.93, 0.95]). Performance of models using CUIs were similar to unigrams (p>0.05). Models demonstrated excellent clinical face validity. CONCLUSIONS: Both CUIs and unigrams demonstrated excellent discrimination in predicting severity of chest injury using the first eight hours of clinical documents. Our model demonstrates that automated anatomical injury scoring is feasible and may be used for aggregation of data for trauma research and quality programs.
Authors: Tabitha Garwe; Linda D Cowan; Barbara Neas; Timothy Cathey; Brandon C Danford; Patrice Greenawalt Journal: Acad Emerg Med Date: 2010-11 Impact factor: 3.451
Authors: Philipp Fueglistaler; Felix Amsler; Marcel Schüepp; Ida Fueglistaler-Montali; Corinna Attenberger; Hans Pargger; Augustinus Ludwig Jacob; Thomas Gross Journal: Am J Surg Date: 2010-03-15 Impact factor: 2.565
Authors: M Kit Delgado; Michael A Yokell; Kristan L Staudenmayer; David A Spain; Tina Hernandez-Boussard; N Ewen Wang Journal: JAMA Surg Date: 2014-05 Impact factor: 14.766
Authors: Stephen T Wu; Hongfang Liu; Dingcheng Li; Cui Tao; Mark A Musen; Christopher G Chute; Nigam H Shah Journal: J Am Med Inform Assoc Date: 2012-04-04 Impact factor: 4.497
Authors: Sujay Kulshrestha; Dmitriy Dligach; Cara Joyce; Richard Gonzalez; Ann P O'Rourke; Joshua M Glazer; Anne Stey; Jacqueline M Kruser; Matthew M Churpek; Majid Afshar Journal: JAMIA Open Date: 2021-03-01