Improved Programs for Injury Categorization Using ICD-10
- 1Department of Surgery, Maine Medical Center, Portland, ME, USA
- 2Department of Surgery, Tufts University School of Medicine, Boston, MA, USA
Abstract
Background: Among the many efforts to convert International Classification of Diseases (ICD) codes into useful categories and scores for injury research has been the “ICD Programs for Injury Categorization” (ICDPIC).
Methods: Data from the American College of Surgeons Trauma Quality Program (TQP) and US National Inpatient Sample (NIS) were used to develop Version 2 of ICDPIC for the R language (ICDPIC-R2). Programming efficiency was increased and new injury categorizations were added. Effect estimates for each ICD-10 Clinical Modification (ICD-10-CM) code were derived using logistic ridge regression with hospital mortality as the outcome. These estimates were converted to Abbreviated Injury Scale (AIS), Maximum AIS (MAIS), Injury Severity Score (ISS), and New Injury Severity Score (NISS). ICD-based Injury Severity Score (ICISS) was also calculated using TQP, NIS, and previously published data from an International Collaborative Effort for Injury Statistics (ICE-ICISS). Mortality predictions from ICDPIC-R2 were compared to those from ICDPIC Version 1 for R (ICDPIC-R1). Predictions derived from TQP data were partially validated using NIS data, and vice versa.
Results: Among 1,178,107 patient records in TQP or NIS, there were 8,427 distinct ICD-10-CM anatomic injury diagnosis codes. Regression results or predictions of mortality derived from AIS in ICDPIC-R2 were satisfactory measures of severity (evaluated using area under a Receiver Operating Characteristic curve or mean square error) and improved slightly on predictions obtained with ICDPIC-R1. ICISS methods (new in ICDPIC-R2) were also quite accurate, including ICE-ICISS. Predictions derived from TQP data performed well with NIS data, and vice versa. Runtimes were much faster with ICDPIC-R2. Effect estimates from ICD-10-CM codes and ICE-ICISS were applied to an additional 12,124 possible ICD-10 codes, including all those published by the World Health Organization.
Conclusion: ICDPIC remains reasonably accurate for categorizing injured patients, especially when only ICD codes are available. ICDPIC-R2 adds features and may extend applicability, especially for international data.
Related
- Neoplastic and non-neoplastic astrocyte states converge on mesenchymal and injury-response programs during glioblastoma recurrence
- Empagliflozin attenuates ISO-induced myocardial injury through PGC-1α-mediated mitochondrial biogenesis and improved fatty acid metabolism
- Divergent Schwann cell-immune cell trajectories across vestibular schwannoma subtypes mirror nerve injury-repair programs.
- Deep visual proteomics reveals distinct proximal tubular and glomerular injury programs in experimental diabetic kidney disease
- Regulated cell-death programs across patient-resolved proximal-tubule states in acute kidney injury: ASPHD1-silencing increases apoptosis and lipid peroxidation
BibTeX
@article{clark2026improved,
title = {{Improved Programs for Injury Categorization Using ICD-10}},
author = {David E Clark},
journal = {Clinical Epidemiology},
year = {2026},
month = nov,
doi = {10.2147/clep.s628506},
url = {https://doi.org/10.2147/clep.s628506},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Improved Programs for Injury Categorization Using ICD-10
AU - David E Clark
PY - 2026
DA - 2026/11/01
JO - Clinical Epidemiology
DO - 10.2147/clep.s628506
UR - https://doi.org/10.2147/clep.s628506
AB - Background: Among the many efforts to convert International Classification of Diseases (ICD) codes into useful categories and scores for injury research has been the “ICD Programs for Injury Categorization” (ICDPIC). Methods: Data from the American College of Surgeons Trauma Quality Program (TQP) and US National Inpatient Sample (NIS) were used to develop Version 2 of ICDPIC for the R language (ICDPIC-R2). Programming efficiency was increased and new injury categorizations were added. Effect estimates for each ICD-10 Clinical Modification (ICD-10-CM) code were derived using logistic ridge regression with hospital mortality as the outcome. These estimates were converted to Abbreviated Injury Scale (AIS), Maximum AIS (MAIS), Injury Severity Score (ISS), and New Injury Severity Score (NISS). ICD-based Injury Severity Score (ICISS) was also calculated using TQP, NIS, and previously published data from an International Collaborative Effort for Injury Statistics (ICE-ICISS). Mortality predictions from ICDPIC-R2 were compared to those from ICDPIC Version 1 for R (ICDPIC-R1). Predictions derived from TQP data were partially validated using NIS data, and vice versa. Results: Among 1,178,107 patient records in TQP or NIS, there were 8,427 distinct ICD-10-CM anatomic injury diagnosis codes. Regression results or predictions of mortality derived from AIS in ICDPIC-R2 were satisfactory measures of severity (evaluated using area under a Receiver Operating Characteristic curve or mean square error) and improved slightly on predictions obtained with ICDPIC-R1. ICISS methods (new in ICDPIC-R2) were also quite accurate, including ICE-ICISS. Predictions derived from TQP data performed well with NIS data, and vice versa. Runtimes were much faster with ICDPIC-R2. Effect estimates from ICD-10-CM codes and ICE-ICISS were applied to an additional 12,124 possible ICD-10 codes, including all those published by the World Health Organization. Conclusion: ICDPIC remains reasonably accurate for categorizing injured patients, especially when only ICD codes are available. ICDPIC-R2 adds features and may extend applicability, especially for international data.
SN - 1179-1349
ER -