Effectiveness of employer financial incentives in reducing time to report worker injury: an interrupted time series study of two Australian workers’ compensation jurisdictions

In Australia, WC insurance is regulated by state, territory, and Commonwealth government agencies, each with its own policy settings and mixtures of public and private claims management systems [23]. There are nine main schemes: one for each of the six states and two territories, and one for workers in the Commonwealth government and interstate employers. In 2014, these schemes covered 10.8 million, or 94%, of Australia’s 11.5 million workers [24].

In 2007, scheme reviews for both SA [11] and TAS [12] recommended ERIs as a means of improving injured worker outcomes by reducing the time between injury and WC services. In SA, ERIs came into effect on 1 January 2009 as part of the Workers Rehabilitation and Compensation (Scheme Review) Amendment Act 2008 [13]. From this date, employers who lodged a WC claim within two working days of becoming aware of an injury were given a rebate on their insurance excess [13], which could amount to the first 14 calendar days of wage replacement payments, or up to $2335.60 per week for two weeks [25]. As implemented in SA, ERIs were intended as a bonus to employers who performed better than the minimum required. The statutory five-day period to lodge a WC claim remained in place, enforced with a fine of up to AUD $1000 for late lodgement [26]. Eighteen months later, TAS implemented a similar set of ERIs as part of the Workers Rehabilitation and Compensation Amendment Act 2009 [14]. From 1 July 2010, employers were required to report worker injuries to their insurer within three working days of becoming aware of them. Employers who exceeded this time became responsible for wage replacement payments for each day they were late. In TAS, ERIs were recommended partly in response to increases in the proportion of claims that were reported outside the existing statutory five-day period, reaching one-quarter by 2005/06 [12]. TAS retained a statutory five-day lodgement period with ERIs, though it was not enforced by fine [14, 27].

In both jurisdictions, ERIs were part of a larger package of WC legislative amendments. SA implemented these amendments in five waves over the course of two years beginning 1 July 2008, six months prior to ERIs coming into effect. TAS implemented all legislative amendments at one time. The features of ERIs in each jurisdiction are summarised in Table 1.

The data were derived from the National Dataset for Compensation-based Statistics, a collection of de-identified WC claim records from each Australian WC jurisdiction, compiled by Safe Work Australia for inter-jurisdictional comparisons and national data analysis [23]. To minimise bias from co-occurring events, changes in SA and TAS were evaluated in contrast to a comparator [20, 21, 28] composed of other Australian WC jurisdictions that had not implemented ERIs and adhered to the most up-to-date coding standards for the duration of the study. These included New South Wales, Victoria, Western Australia, and the Northern Territory. Claims were eligible for inclusion if they had been lodged with an insurer between July 2006 and June 2012 and subsequently accepted, resulting in N = 1,470,303 records.

We evaluated ERI impact with a multiple baseline ITS study design. ITS is considered one of the most powerful quasi-experimental designs for evaluating natural experiments like policy change or community interventions where data have been collected at regular intervals before and after a time-bounded event [19‐21, 28, 29]. ITS can be applied where randomised controlled trials are cost-prohibitive, impractical, or unethical, and can evaluate population-level impacts in real-world settings [21, 22, 28‐31]. Unlike other before-and-after analytical techniques such as difference-in-differences, ITS accounts for secular trends, minimising the likelihood that observed differences due to pre-existing trends are misattributed to the event [28, 31‐34]. ITS also detects trend changes, such as the progressive reduction of a surgical procedure following publication of evidence that it was ineffective [35]. Often used to evaluate count or rate changes, such as the incidence of acute coronary events [19] and gun homicides [36], ITS has been applied to analysis of changes in central tendency, such as the impact of publishing guidelines on median reporting quality in peer review journal articles [37] and reimbursement caps on the mean number of prescriptions dispensed to public assistance recipients [28]. In WC settings, ITS has been used to evaluate the impact of experience-rated premium discount programmes on claim rates [38]. The 18-month gap between SA and TAS’s ERI implementation allowed for the addition of a multiple baseline design, which minimises the biasing potential of co-occurring events by analysing the same yet staggered events in different populations [22, 29, 30].

Outcomes were monthly median days in three main time periods comprising the claim lodgement process, which were total time (injury/illness date to insurer decision date), claim reporting time (injury/illness date to employer report date), and insurer decision time (employer report date to insurer decision date). In SA, it was possible to evaluate two time periods that together comprised claim reporting time: worker reporting time (injury/illness date to worker report date) and employer reporting time (worker report date to employer report date). The time periods are illustrated in Fig. 1. We used the median for stability and to evaluate ERI impact on the “average” claim. After identifying some substantial changes to the upper range of the interquartile range (IQR) following ERI implementation, and in light of nearly 25% of claims in TAS exceeding the statutory five-day reporting period [12], we conducted sensitivity analyses with the 75th percentile as the outcome. Worker report dates were missing in 75.6% of TAS cases (41,388 of 54,765 cases), far below the recommended threshold of 80% complete data [31]. This date was also entirely missing in New South Wales, which we excluded from the comparator for these analyses. All other jurisdictions had at least 80% complete data for each outcome (see Additional file 1: Table S1 for a summary of data completeness).

Crude analyses describe the duration of the claim lodgement process before and after ERIs using median number of days with IQR.

For ITS analyses, we created an aggregate dataset of monthly median durations for SA, TAS, and the comparator, along with the 75th percentile for sensitivity analyses. SA and TAS were analysed in separate models, both in reference to the comparator. Changes were evaluated with a generalised least-squares regression [39]. Harmonic terms were added to the models to account for seasonality, if statistically significant [40, 41]. To account for additional autocorrelation, we fit the data to autoregressive-moving average (ARMA) models based on correlated residuals observed in Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) plots [28, 39, 42]. For each analysis, all possible models were compared on Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) with an analysis of variance (ANOVA). The final models, with harmonic and ARMA terms, are listed in R code in the additional files. To check for history threats to internal validity such as national-level events, we conducted an ITS analysis of monthly claim volumes. Volumes were indexed to the first point in the time series to account for large volume differences between the groups. Methods are reported against the Quality Criteria for ITS Designs [31] in Additional file 1: Table S1. Seasonally-adjusted ITS trend lines were plotted over monthly data points for each jurisdiction using code adapted from the ITS tutorial in Lopez Bernal et al. [19]. For visual clarity, trend lines for SA and TAS reflect regression models without reference to the comparator, and the comparator trend line includes both ERI event interruptions.

Analyses were conducted in R (v3.3.2) [43] using RStudio (v1.0.44) [44] and the nlme package for generalised least squares regression analyses [45]. R Markdown files containing code for analyses and plotting can be found in the Additional files 2, 3, 4, 5 and 6.

This study’s main strengths were the use of a powerful quasi-experimental research design, the ITS, with multiple baselines, a comparator, and adjustments for seasonality and autocorrelation, using a large number of claims from a dataset created for inter-jurisdictional comparisons and national data analysis [23]. The number of observations exceeded most recommended minimums and generated greater certainty about seasonal and autocorrelation adjustments. Additionally, we evaluated several time periods within the claim lodgement process, exploring both the ERI target, claim reporting times (via employer reporting times), and areas that could have been indirectly affected (insurer decision time).

The study had several limitations. The ITS design is vulnerable to confounding from co-occurring events, which can lead to misattributions of cause [29, 32]. Further, ITS assumes linear trends among data, which is likely to be violated with longer time series [32, 33], and with the use of aggregated data, does not allow for adjustment of individual-level characteristics [28, 33]. The staggered introduction of SA’s amendments likely confounded insurer decision time, as did the introduction of a new IT system.

Changes in monthly claim volumes raise the possibility of selection bias, but also demonstrate the strength of using a comparator within an ITS study design. While claim volumes significantly changed in SA and TAS following ERI implementation, both attenuated to non-significance when adjusting for the comparator (i.e., controlling for the national-level effect). The timing and direction of volume changes suggest they were the result of the Global Financial Crisis (GFC). Though the GFC did not have as large an impact on Australia as other developed economies [49], there was nevertheless a reduction in WC claims. This may be due to worker reluctance to make a claim and risk unemployment in the face of economic uncertainty, and/or employers favouring experienced, less risky workers in hiring/firing decisions, resulting in safer worker cohorts [50]. If the change in claim volumes was due to the GFC, use of the comparator likely reduced its potential to bias outcomes.

The comparator was unable to adjust for all co-occurring events, particularly those specific to each jurisdiction. While SA and TAS introduced ERIs as part of a larger set of legislative WC amendment packages, they likely varied in terms of their impact on the claim lodgement process. For instance, the staggered introduction of WC amendments in SA likely diffused administrative burden. The introduction of provisional liability in concert with ERIs in SA could have incentivised injury reporting among workers due to greater certainty of WC services. There is also the possibility that other co-occurring events, specific to either jurisdiction, could have confounded the associations. Further, the two ERI policies had different designs. It is unclear whether the number of days prescribed (two in SA, three in TAS) or the incentive type (limited rebate in SA, accruing penalty in TAS) modified ERI effectiveness. Lastly, in several cases baseline durations were different, which likely moderated ERI impact.