Article Text

Download PDFPDF
Original article
Prediction of all-cause occupational disability among US Army soldiers
  1. D Alan Nelson1,
  2. Vickee L Wolcott2,
  3. Lianne M Kurina1
  1. 1Division of Primary, Preventive, and Community Medicine, Department of Medicine, Stanford University School of Medicine, California, USA
  2. 2Army-Baylor University Graduate Program in Health and Business Administration, Joint Base San Antonio (JBSA), Ft Sam Houston, Texas, USA
  1. Correspondence to Dr Lianne M Kurina, Stanford University 450 Serra Mall, Bldg 20, Stanford, CA 94305-2160, USA; lkurina{at}


Introduction Long-term occupational disability rates associated with eventual discharges from military service have risen sharply among active-duty US Army soldiers during the last three decades, with important implications for soldier health and national security alike. To address this problem, we built predictive models for long-term, all-cause occupational disability and identified disability risk factors using a very large, multisource database on the total active-duty US Army.

Methods We conducted a cross-temporal retrospective cohort study and used mixed-effects logistic regression models to derive and validate disability risk assignments. The derivation cohort included 510 616 US Army soldiers on duty in December 2012, and the validation cohort included 483 197 soldiers on duty in December 2013.

Results The predictive model yielded an overall c-statistic of 85.97% (95% CI 85.61% to 86.32%). Risk thresholds at the population's 75th and 95th centiles identified 80.53% and 42.08%, respectively, of the disability designations that occurred population wide during the subsequent 9 months. Frequent work excusals, high outpatient care utilisation and psychotropic medication use were the strongest independent predictors of later disability.

Conclusions These findings indicate that predictive models using diverse data types can successfully anticipate long-term occupational disability among US Army soldiers and could be used for disability risk screening.

  • predictive analytics

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.