Master of Computer Science - Research
School of Computer Science and Software Engineering
Mills, Ross Irwin, Predicting workers compensation return-to-work outcomes at claim lodgement, Master of Computer Science - Research thesis, School of Computer Science and Software Engineering, University of Wollongong, 2011. https://ro.uow.edu.au/theses/3472
When an injured worker returns to work, their return to work is influenced by a variety of factors including the injured worker’s true work capacity, the availability of work to return to, and the injured worker’s motivation to return to work.
There is strong evidence in the literature associating Absenteeism with receiving compensation payments or having a lawyer involved, with workplace climate factors including job satisfaction and job strain, increasing age and with the presence of co-morbidity (particularly mental health related). Worker expectations (which to a degree reflected that of their treating professionals) were a predictor of Return-to-Work outcome. There was weaker evidence associating Female Gender, Smoking and Short Service in the current job with Absenteeism.
No studies have attempted to predict Return-to-Work outcomes on the basis of the information available at the time of the Injured Worker lodging a worker’s compensation claim.
This thesis sets out to, using the data as collected by the commonly accepted practice, to determine if it is possible to identify which factors influence an Injured Worker’s Return to Work, and can we predict at the time of claim lodgement which Injured Worker’s claims will become Tail Claims?
Data was obtained from two sources within Australia, one being a NSW Self-insurer, and the other being a different Australian workers compensation jurisdiction (WorkSafe VIC). A combined total of 9048 cases were analysed. Study entry was having a claim notified to each of the two organizations as being a significant injury (by virtue of time lost or financial expenditure) in the four months following 01 April 2007. Study exit was at either being certified fit for pre-injury duties, or their medical certification at 4 months post insurer notification.
Both organisations providing data used the same Notification of Injury and Medical Certificate forms. Information was extracted by the organization, de-identified, and submitted for analysis. Although using common forms, and collecting common data, the data submitted differed between the organizations, with neither organisation being able to extract all of the information collected at the time of claim lodgment.
This research has found that although it is not possible to predict Return-to-Work outcomes by Multiple Regression analysis of existing data, there are multiple associations with Return-to-Work outcomes buried within the data. This research did find associations between (older) Age, Diagnosis, Mechanism of Injury, Bodily Location of Injury and Return-to-Work Outcome. Although present, these associations were either too small, or diluted by the large number of sub-groups such that they are unable to be used to build a predictive model with these Data Sets.
There was a consistent trend for Female gender to be associated with a prolonged Return-to-Work outcome. No significant relationship was found between Type of Employment, Occupation Income, Work Hours, Workplace Size, Industry and Return-to-Work outcomes.
In summary, although this researcher found that although it is not possible to predict return-to-work outcomes by multiple regression analysis of the existing data, there are multiple associations with return-to-work outcomes buried within the data.
Many of the short comings of this work could be solved if the work could be performed prospectively, with the information collected specifically designed for this purpose. Then through a combination of better data management (including cleaning of data, collapsing of categories, eliminating ambiguity and overlap of definitions, and improved accessibility), expanding the questions asked at the time of claim lodgement (to include potentially predictive data identified by the literature), and utilising different analytic tools (applying data mining techniques before developing regression models) it may possible to markedly improve the predictability of return-to-work outcomes at the time of claim lodgement. This would create the opportunity for early intervention for potential problem cases to be addressed early in the claims process, possibly significantly reducing time lost from work.