IN BRIEF

People Outperform Computers at Assigning Injury Severity in Trauma Patients

August 7, 2019
Written by

Injury severity scores determined by specially trained staff were better at predicting length of stay and mortality than the computer generated scores.

The success of trauma quality improvement and research is based upon detailed information maintained in trauma registries. Stratification of injury severity is critically important when collecting patient data for these registries.

By compiling large collections of patient information into a single, quantifiable value, researchers and clinicians can compare hospitals, regions, practice environments, instances of trauma team activation, and modes of therapy. The Injury Severity Score (ISS) summarizes the severity of multi-system injuries from defined body regions into a single numerical value and functions as an important predictor of morbidity and mortality. The ISS is widely used in injury research and trauma care quality improvement efforts. Hospitals can either use manually assigned or computer software assigned ISS in their registries.

Maintenance of registries requires specially trained staff and creates significant personnel costs. A computer generated ISS could decrease costs and increase efficiency and may be a useful alternative in some centers. A program was developed to use International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes to generate ISS: ICDPIC. The ICDPIC was developed by David Clark and colleagues and used by the Agency for Healthcare Research and Quality in HCUP data.

A recent study from the Center for Pediatric Trauma Research and the Center for Injury Research and Policy at Nationwide Children’s Hospital compared the ISS manually assigned by hospital personnel and those generated by the ICDPIC software for value agreement and predictive power of length of stay and mortality. Researchers found that manually and computer assigned ISS had strong comparative agreement for minor injuries but did not correlate well for critical injuries (ISS ranging from 25 to 75).

“This study demonstrated that the ICDPIC software did not capture the same depth and accuracy of information as trained manual coders for critically injured patients,” says Kathy Haley MS, BSN, RN, Trauma Program manager at Nationwide Children’s. The ISS generated by ICDPIC software had a statistically lower length of stay and mortality predictive power when compared with ISS assigned by manual coders.

“These findings have important implications for hospital resource allocation, trauma care quality improvement and trauma research based on existing medical records,” says Jonathan Groner, MD, Trauma medical director and medical director of the Center for Pediatric Trauma Research at Nationwide Children’s. “The findings are especially salient when utilizing trauma databases that rely upon ICD-9-CM codes to generate the ISS.”

“We are glad that this study has provided important scientific evidence for continuing the support of trauma registrars by hospital trauma programs across the nation,” says Henry Xiang, MD, MPH, PhD, senior author of the study, professor of Pediatrics, and director of the Center for Pediatric Trauma Research at Nationwide Children’s. “As artificial intelligence becomes increasingly popular, some people have wondered whether computers can replace human registrars to assign injury severity scores in trauma registries. Our findings suggest that we are not there yet.”

Based on the results of the study, the authors advise that hospitals should be cautious about transitioning to computer assigned ISS, specifically for critically injured patients.

 

Citations:

  1. Otto L, Wang A, Wheeler K, Shi, J, Groner J, Haley K, Nuss K, and Xiang H. Comparison of Manual and Computer Assigned Injury Severity Scores. Injury Prevention. 2019 Jul 12. [Epub ahead of print]
  2. Clark DE, Osler TM, Hahn DR. ICDPIC: Stata module to provide methods for translating International Classification of Diseases (Ninth Revision) diagnosis codes into standard injury categories and/or scoresStatistical Software Components. 2009; S457028, Boston College Department of Economics, revised 29 Oct 2010. 

Image credit: Adobe Stock