Article Text

Download PDFPDF
How well do principal diagnosis classifications predict disability 12 months postinjury?
  1. Belinda J Gabbe1,2,
  2. Pam M Simpson1,
  3. Ronan A Lyons1,2,3,
  4. Suzanne Polinder4,
  5. Frederick P Rivara5,
  6. Shanthi Ameratunga6,
  7. Sarah Derrett7,8,
  8. Juanita Haagsma4,
  9. James E Harrison9
  1. 1Department of Epidemiology and Preventive Medicine, Monash University, Melbourne, Victoria, Australia
  2. 2Centre for Improvement of Population Health through E-records Research, Swansea University, Swansea, UK
  3. 3Public Health Wales NHS Trust, Cardiff, UK
  4. 4Department of Public Health, Erasmus MC, Rotterdam, The Netherlands
  5. 5Departments of Pediatrics and Epidemiology, University of Washington, Seattle, Washington, USA
  6. 6Section of Epidemiology and Biostatistics, School of Population Health, University of Auckland, Auckland, New Zealand
  7. 7Injury Prevention Research Unit, Department of Preventive and Social Medicine, Dunedin School of Medicine, University of Otago, Dunedin, New Zealand
  8. 8School of Health and Social Services, Massey University, Palmerston North, New Zealand
  9. 9Research Centre for Injury Studies, Flinders University, Adelaide, South Australia
  1. Correspondence to Professor Belinda Gabbe, Department of Epidemiology and Preventive Medicine, Monash University, The Alfred Centre, Commercial Rd, Melbourne, VIC 3004, Australia; belinda.gabbe{at}


Background The application of disability weights by nature of injury is central to the calculation of disability-adjusted life years (DALYs). Such weights should represent injury diagnosis groups that demonstrate homogeneity in disability outcomes. Existing classifications have not used empirical data in their development to inform groups that are homogeneous for disability outcomes, limiting the capacity to make informed recommendations for best practice in measuring injury burden.

Methods The Validating and Improving injury Burden Estimates (Injury-VIBES) Study includes pooled data from over 30 000 injured participants recruited to six cohort studies. The International Classification of Disease 10th Revision (ICD-10) diagnosis codes were mapped to existing injury burden study groupings and prediction models were developed to measure the capacity of the injury groupings and ICD-10 diagnoses to predict disability outcomes at 12 months. Models were adjusted for age, gender and data source and investigated for discrimination using area under the receiver operating characteristic curve (AUC) and calibration using Hosmer–Lemeshow statistics and calibration curves.

Results Discrimination and calibration of models varied depending on the outcome measured. Models using full four-character ICD-10 diagnosis codes, rather than groupings of codes, demonstrated the highest discrimination ranging from an AUC (95% CI) of 0.627 (0.618 to 0.635) for the pain or discomfort item of the EQ-5D to 0.847 (0.841 to 0.853) for the extended Glasgow Outcome Scale independent living outcome. However, gain over other groupings was marginal.

Conclusions Prediction performance was best for measures of function such as independent living, mobility and self-care. The classifications were poorer predictors of anxiety/depression and pain/discomfort. There was no clearly superior classification.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.