Citation

Mills R. Intern. Med. J. 2012; 42(8): 924-927.

Affiliation

RAN.

Copyright

(Copyright © 2012, John Wiley and Sons)

DOI

10.1111/j.1445-5994.2011.02639.x

PMID

22152007

Abstract

Aim:  The research question is: is it possible to predict, at the time of workers compensation claim lodgement, which workers will have a prolonged return to work (RTW) outcome? This paper illustrates how a traditional analytic approach to the analysis of an existing large database can be insufficient to answer the research question, and suggests an alternate data management and analysis approach. Methods:  This paper retrospectively analyses 9018 workers compensation claims from two different workers compensation jurisdictions in Australia (two data sets) over a four month period in 2007. De-identified data, submitted at the time of claim lodgement, was compared with RTW outcomes for up to three months. Analysis consisted of descriptive, parametric (ANOVA and Multiple Regression), survival (Proportional Hazards) and data mining (Partitioning) analysis. Results:  No significant associations were found on parametric analysis. Multiple associations were found between the predictor variables and RTW outcome on survival analysis, with marked differences being found between some sub-groups on partitioning - where diagnosis was found to be the strongest discriminator (particularly neck and shoulder injuries). There was a consistent trend for female gender to be associated with a prolonged RTW outcome. The supplied data was not sufficient to enable the development of a predictive model. Conclusion:  To build a predictive model to identify injured workers at risk of a prolonged RTW outcome at the time of workers compensation claim lodgement in Australia it is necessary to redesign the workers compensation claim lodgement questionnaires; improve data management, and utilise specific analytic techniques.

Language: en