Data analysis

RTW baseline characteristics, including full/modified duties and full/part-time, were summarised using descriptive statistics. Outcomes were assessed using survival analysis with Cox proportional hazards regression models employed to determine the multivariate predictors of time to RTW. The Cox model is considered an appropriate approach to accounting for time to an event [31]. The variables selected for the model have been shown to be independent predictors of RTW and/or potential confounders of poorer outcomes in other research [7, 16–18]. Similarly, compensation-related factors were selected for the same reasons [2, 4, 7, 8].

Selection of variables for the Cox model was based on associations between baseline characteristics, including compensation related factors and time to RTW. These were assessed using the logrank test to detect differences in the survival distributions across categorical variables. All variables with p-value ≤ 0.20 were entered into the Cox regression model using a backward elimination process with an entry p-value < 0.05 and an exit p-value < 0.10. Variable selection was confirmed through explained variation and predictive accuracy using R-squared values calculated with the Cox and Snell R-squared approach and a concordance index [32]. The concordance index is a widely applicable measure with progressive addition of factors that improve discrimination of the model. When a variable is added and the c-index plateaus or decreases, that variable and additional variables can be regarded as noise and excluded to avoid over fitting in the model [31]. Furthermore, explained variance using R-squared describes the relative importance of adding each variable into the Cox regression model.

Data from participants where the endpoint (RTW) had not occurred or was unknown at 24 months were considered censured. In studies of survival, in which the outcome is death, Hazard Rate Ratios (HRR) greater than 1 indicates risk. However, in this study, the fewer cumulative days of time taken to RTW, the more positive the outcome, in terms of injury recovery/RTW, and the higher the HRR. Therefore, a HRR less than 1 indicates higher risk and a longer time taken to RTW. A test of proportionality was performed on all predictors, and claim made and legal representation. The assumption of proportionality was not violated (p > 0.05) [33].

A separate Cox proportional hazards regression analysis was done for compensation related variables (claim made and legal representation), as these variables were measured at the six month time point and only a portion of participants made a claim and sought legal representation. In this Cox regression, claim made and legal representation were added to the final variables in the baseline RTW model. All data analysis was performed using SPSS statistical software version 22 (SPSS Inc, USA).