Inclusion and exclusion criteria

Inclusion criteria were:

• prospective study design;
• follow-up period of at least six months;
• musculoskeletal injury of any type (if mixed aetiology, the majority of participants has sustained a musculoskeletal injury);
• at least 18 years of age (for majority of participants);
• study aimed to determine prognostic factors associated with an outcome, or to assess the effect of an intervention with compensation related factors included as covariates;
• measurement of one or more compensation related factors associated with an outcome;
• at least one validated health related outcome measure was reported; and
• inclusion of a predictive model with multivariate statistical analysis.

Exclusion criteria were:

• participants with dementia or significant pre-existing cognitive impairment;
• participants with a moderate or severe traumatic brain injury, spinal cord injury, psychological or other organ and body system injuries;
• studies involving only children; and
• studies where the only outcome assessed is return to work with no other validated health related outcome.

Due to the diverse injury definitions, three approaches were used: definition and context (mechanism or insidious onset); diagnosis; and/or duration (acute or chronic). Only prospective studies were included to reduce the risk of bias [20]. A follow up period of six months was given to allow for injury recovery. Return to work was excluded because there is no standardised measure although it is recognised that return to work is correlated with health status.

Search strategy

Searches were conducted using Medline, CINAHL, Embase, Informit and Web of Science for studies published up to October 2012. Complete search strategies are available in S1 Appendix. The strategy was based on recommended guidelines to maximise search sensitivity [21]. Key elements involved exploding terms related to cohort studies, compensation and musculoskeletal injury. MeSH headings and text words were used in conjunction with Boolean operators and wildcards. For Informit health, law and social science subjects with key words (compensation, health and outcome) were used. Web of Science and Informit provided access to grey literature. A medical librarian was consulted to assist in developing the search strategies, which were reviewed by the authors.

Articles were initially screened by two authors (DM and PC) based on title and abstract. The full text of short listed papers was retrieved. Three investigators (DM, PC and IM) conducted a two stage screening process with two authors reviewing all papers in the second stage. Articles were not excluded based on methodological quality; this was taken into account in the quality assessment.

Data extraction, quality assessment and synthesis

The characteristics of each study were tabulated to address the aims of the review [22–24]. Statistical information, including reported effect sizes, for all compensation related factors associated with outcome(s) was recorded. Associations were considered significant if the 95% confidence intervals of the odds, hazard or relative risk ratios did not include 1 and/or the p-value was less than 0.05. Compensation related factors were categorised as follows:

• compensation (Yes/No)—having an open claim or having made a claim versus no open claim or no claim made;
• lawyer involved (Yes/No)—having sought or obtained legal representation versus having none;
• claim type—having an open claim or having made a claim under a specific scheme jurisdiction (Workers Compensation (WC), traffic injury (including Compulsory Third Party (CTP)), public health coverage, private health insurance, other (such as disability insurance, public liability, victims compensation);
• number of sick days in prior three years;
• prior claim (Yes/No);
• fault (Yes/No)—making a claim under tort (fault) or no fault insurance arrangements; and
• compensation at two years (Yes/No)—whether the claim was open or closed/settled at two years.

Outcomes were categorised based on measurement constructs. Similar classifications have been used in previous publications [12, 13, 25]. The categories were:

• physical function—generic and specific measures including recovery and disability, and physical health components of health related quality of life measures;
• psychological function—diagnostic based measures and mental health components of health related quality of life measures; and
• pain.

Unlike intervention studies there is no agreed quality assessment methodology for systematic reviews of prognostic studies [24, 26–28]. However, there is some guidance on assessing study quality and risk of bias [21–23, 26, 27, 29]. Aspects such as scoring remain controversial, especially for assessing the effect size of an intervention [23, 30–32]. For pragmatic purposes and to provide a meaningful conclusion we followed the methodology used in similar prognostic systematic reviews where a summary score was used [11, 14].

The quality assessment criteria address six areas of potential bias: study participation; study attrition; prognostic factor measurement; outcome measurement; confounding measurement; and analysis [23]. Each criterion in Table 1 specifies a bias and is assigned “Yes” or “No” with “Yes” scores being totalled (maximum score is 18). Further details are available in S2 Appendix. All papers were reviewed by two authors (DM and PC) independently. Discrepancies were resolved by consensus and/or consultation with two other authors (IC and IH). A score of 15 or over was deemed high quality, moderate quality was 12 to 14, and low was 11 or below. Although arbitrary, this division provided a fairly even distribution of scores and reflected the study quality.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Quality assessment criteria.

https://doi.org/10.1371/journal.pone.0117597.t001

Grading quality of evidence

Data analysis was based on recommendations from the GRADE (Grading of Recommendations Assessment, Development and Evaluation) working group. GRADE classifies strong, moderate and limited evidence based on: the number of papers; study design and quality; and the consistency and directness of results [28]. The levels are illustrated in Table 2. This methodology has been used in similar systematic reviews [11, 12, 14, 33]. Inconsistent evidence refers to the negative effect of a factor in one study with a positive effect in another study regardless of study quality. For example if high quality studies showed findings in one direction and low quality studies in another; this would be considered inconsistent. In setting out this paper the authors referred to the PRISMA statement to ensure reference to all relevant reporting items [24].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Levels of evidence.

https://doi.org/10.1371/journal.pone.0117597.t002

Study selection

The search results and study selection process are illustrated in Fig. 1. Initially, 391 papers were independently reviewed by one investigator (DM, PC or IM). Full texts of the remaining 89 papers were independently examined by two investigators (DM, PC or IM). Reasons for exclusions are explained in S3 Appendix. In summary, they were: no predictive statistical model and/or multivariate analysis (n = 10); compensation related factor not measured as a predictor (n = 15); retrospective studies (n = 22); compensation only cohort without additional compensation related factor for comparison (n = 4); no validated health outcome (n = 6); and/or majority of cohort without musculoskeletal injuries (n = 2). Often ‘prospectively collected data’ were used but the study hypothesis and design were initiated post hoc after routine baseline data collection during the follow up period; these were by definition retrospective. Hand searching of reference lists and personal communication with experts minimised the potential for missing papers. Ultimately, 29 papers met the inclusion criteria.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Retrieval of studies for the systematic review.

https://doi.org/10.1371/journal.pone.0117597.g001

In addition, ten papers reported results from overlapping cohorts. Only one paper from each cohort was included to avoid over representation of one population by taking into account: the range of compensation related factors and outcomes measured; injury type/s; sample size; and study quality. The studies all measured compensation status [4, 8, 9, 34–40] but the included ones measured a greater range of outcomes and/or with more applicable and comprehensive results [4, 9, 35, 36].

Quality assessment

Following independent assessment, two authors (DM and PC) scored in agreement 91% of the time for each criterion. To resolve discrepancies: reasons for individual scores; consistent criterion interpretation; text explanations; and other referenced papers were considered. Areas of disagreement were: study participation—potential baseline measurement error and poor representative sampling (criteria S2, S3); and prognostic factor and outcome measurement—inadequate justification for each measure (criteria P2, O2). The grading of the evidence was primarily conducted by the first author (DM) with consensus review by the remaining authors (PC, IC and IH).

There were seven papers referred to other authors (IC and IH) to reach consensus. These were intervention studies, and/or had complex statistical analysis [41–47]. Statistical pooling was not possible due to heterogeneity of compensation related factors and outcome definitions including constructs, and follow up time periods.

Overall, 11 studies rated as high quality, 10 as moderate and eight as low. Complete scoring can be obtained from the first author.

Summary of included studies

Key study characteristics are illustrated in Table 3. Of the 29 included studies 13 were from a primary care setting or surgical clinic and 10 involved hospital recruitment. Several included both settings [44, 45, 48]. A further three recruited via administrative databases [43, 49, 50].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Characteristics of included studies.

https://doi.org/10.1371/journal.pone.0117597.t003

Injury definitions were often incomplete. Acute trauma with a hospital inception source were best described, with baseline data often collected within two weeks [3, 4, 9, 35, 44, 45, 48, 51–53]. Soft tissue injuries with an outpatient inception source were not always clearly documented [42, 47, 54–57]. Furthermore, even if the inception time was stated it was not always obvious when baseline measures were conducted [46, 47, 58, 59]. This was taken into account in the quality assessment (criteria S1, S2). However, if researchers had followed their own criteria it was difficult not to score this positively. Scores are shown in Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Results from included studies.

https://doi.org/10.1371/journal.pone.0117597.t004

Sample size ranged from 65 to 3232 [43, 45]. Age range was not always explicit. In 19 studies the starting age was 14–18 years, whilst in 10 studies no range was stated or it was ambiguous. There were 13 intervention studies, seven surgical and the remaining offering rehabilitation or physiotherapy services.

Follow up was a minimum of six months and a maximum of 10 years [46], the majority (15/29) being 12 months. Loss to follow up ranged from 0% to 52% from baseline [43, 60]; this was difficult to interpret because the periods varied and/or were not reported for each outcome. Only 14 studies achieved less than 20% attrition. Most studies (n = 23) did not account for missing data but recorded loss to follow up (criterion F2). In 22 studies there was a significant difference in baseline variables between participants and those lost to follow up, or it was not explained. This was the lowest scoring criterion (F3).

Summary of compensation related factors

The studies were mostly from the United States of America (nine studies including 18 states) and Australia (nine studies from five states). There were four Canadian studies from five provinces, three Danish, two English, and one each from New Zealand and The Netherlands. The compensation schemes were predominantly WC (11/29) or a combination of WC and road traffic injury schemes (6/29). Only five studies were a road traffic injury scheme alone and one paper was for a universal accident compensation scheme. In six studies it was not stated.

A description of compensation related factors and outcomes including statistics are shown in Table 4. The most common prognostic factor was compensable status (compensation Y/N) measured in 22 studies followed by legal representation (lawyer involved Y/N) measured in six. Claim type was only measured distinctly three times. The least common measures were sick leave, fault and prior claim. Compensation at two years (Y/N) is more akin to claim duration than compensable status that is: making or having made a claim, therefore it was listed separately [49].

Overall, compensation related factors were measured simply. Some specific constructs such as: fault versus no fault; eligibility; entitlements; and/or any restrictions to access entitlements were rarely mentioned. The interpretation of compensation status is potentially ambiguous and may depend on scheme design. Does it mean claim lodged or claim lodged and accepted? Furthermore, claim lodgement with or without claim acceptance and litigation (meaning legal proceedings are underway) are separate factors [36]. Finally, baseline measures of compensation related factors are likely to vary. In certain schemes legal representation can be retained at any time and/or six to 12 months is given to lodge a claim [4, 9, 35, 44, 45, 49, 53, 61]. The timing and duration of exposure to compensation related factors was usually not documented. However, scoring for criteria (P1–3) was inclusive of compensation related and other prognostic factors. The latter were generally well justified, standardised measures with defined constructs; hence many studies (20/29) attained full scores.

Summary of health related outcome measures

Generally, studies selected more than one relevant health related outcome. Pain was the most common (14/29) usually the Visual Analogue Scale (VAS) or Numerical Rating Scale (NRS), although pain is an intrinsic component in many measures. Health related quality of life measures, namely the Short Form Medical Outcomes Study Questionnaires (SF36/12), were next in frequency (6/29). Otherwise, there was a mixture of disability/functional recovery measures such as the Roland Morris Disability Questionnaire (RMDQ), Sickness Impact Profile (SIP) or Neck Disability Index (NDI). In addition, Post Traumatic Stress Disorder (PTSD) questionnaires were used in two studies [36, 44].

Time to claim closure was used as a proxy health outcome in one study with other health and compensation related measures as predictors [43]. This study was included because time to claim closure represented a measure of recovery. Further, incorporating this study did not alter any conclusions. Taking into account the inclusion criterion of a ‘validated health related outcome measure’, most studies scored well (criteria O1–3) with 22/29 studies receiving full marks. Although two studies measured outcomes with face validity, rather than construct and/or criterion validity [50, 57].

Summary of other prognostic factors

Our search strategy was designed to only include studies that measured compensation related factors alongside other prognostic factors; therefore it was beyond the scope to report on all significant prognostic factors (these are listed in Table 3). Nevertheless, it is pertinent to provide some commentary.

The most common were socio-demographic factors such as age, gender, education and occupation, which often had conflicting associations across studies. This could be dependent on societal and population differences [4, 9, 35, 43, 49, 51, 52]. Factors that were frequently associated with poorer outcomes were: psychological such as depression, anxiety, and low self-efficacy [9, 48, 51, 53, 59, 61–63]; and high initial pain scores [3, 36, 41–43, 45, 46, 48, 50, 52–54, 57, 59, 61, 63].

Blame was a potential compensation related factor but it was described as ‘external attributions of responsibility’ or ‘blaming’ someone including themselves or work for their injury, which would not automatically mean access to compensation [4, 53, 64]. Hence, blame was excluded.

Summary of statistical analysis

All studies used a multivariate statistical model to adjust for confounding, and mostly (n = 22) the model was appropriate (criterion A3). Only seven papers received full scores for analysis (criteria A1–5) [3, 35, 36, 43, 47, 51, 61]. Many failed to provide an explanation of their power calculation [9, 42, 45, 49, 52, 55, 60, 62, 63]. On occasion this could be determined from: sample size; number of variables in the multivariate model; and/or loss to follow up [41, 46, 48, 57, 65]. Limited explanations were often given for the final model (criteria A4, A5). For example: which baseline variables were in the univariate analysis; significance level of each variable; and why variables were included/excluded [41, 42, 46, 52, 54–60, 62, 65]. In addition, not all studies reported measures of association and/or p-values [59, 60] especially when there was no association [4, 9, 35, 36, 47, 52, 62, 63]. Other studies mentioned significant results without reporting statistics; these were excluded [56, 63]. Relevant statistics are shown in Table 4.

Grading of evidence

The association between each compensation related factor and health outcome is presented in Table 5. There was either a negative association or no association between a compensation related factor and the outcome measured. There were no reported positive associations, that is: no studies reported that compensation related factors were associated with improved health outcomes. The grades of evidence are determined with reference to Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Compensation factors and outcomes extracted.

https://doi.org/10.1371/journal.pone.0117597.t005

A number of studies measured the association between two compensation related factors and an outcome; in most cases one predictor was significant and the other not significant [4, 9, 48, 49, 54]. Compensation related factors have the potential to be highly correlated. One of main objectives of this review was to determine the effect of each compensation related factor independently on an outcome. To avoid collinearity the non-statistically significant predictors were not considered and excluded from Table 5. Furthermore, the association varied depending on the outcome measured in seven studies [9, 35, 44, 47, 53, 54, 59, 63].

Compensation related factors

Compensation status. The association between compensation status (Y/N) and poorer physical function was statistically significant in eleven studies (four high quality studies, three moderate quality studies and four low quality studies), and not statistically significant in seven studies (three high quality studies, three moderate quality studies and one low quality study). The association between compensation status (Y/N) and poorer psychological function was statistically significant in four studies (two high quality studies, one moderate quality study and one low quality study). The association between compensation status (Y/N) and increased pain was statistically significant in eight studies (two high quality studies, three moderate quality studies and three low quality studies), and not statistically significant in four studies (two high quality studies, one moderate quality study and one low quality study).

Legal representation. The association between lawyer involved (Y/N) and poorer physical function was statistically significant in five studies (three high quality studies and two moderate quality studies), and not statistically significant in two studies (one high quality study and one moderate quality study). The association between lawyer involved (Y/N) and poorer psychological function was statistically significant in three studies (two high quality studies and one moderate quality study).

Other compensation related factors. The association between receiving compensation at two years and poorer physical function was statistically significant in one high quality study.

The association between number of sick days in the three years prior to injury and poorer physical function was statistically significant in one low quality study. The association between number of sick days in prior three years and increased pain was statistically significant in one low quality study.

The association between claim type (having a claim under a specific scheme jurisdiction) and poorer physical function was not statistically significant in one moderate quality study. The association between claim type and increased pain was statistically significant in one moderate quality study and not statistically significant in one moderate quality study.

The association between prior claim and poorer physical function was statistically significant in one moderate quality study. The association between prior claim and increased pain was statistically significant in one moderate quality study.

The association between tort insurance arrangements (as compared to no fault arrangements) and poorer physical function was statistically significant in one high quality study.

Strength of evidence recommendations

There is limited guidance to interpret these mixed results. GRADE refers to the inconsistency of relative treatment effects in binary/dichotomous outcomes following quantitative analysis. Inconsistency is described as a combination of negative and positive associations [66]. Following a review of the literature and consultation with experts, the level of evidence was downgraded for compensation related factors that showed both associations with poorer outcomes and no associations with an outcome [26, 27, 66]. Therefore, the evidence was downgraded for compensation status and poorer physical function; and compensation status and increased pain.

There is moderate evidence of an association between compensation status (having a claim) and poorer physical function. There is strong evidence of an association between compensation status and poorer psychological function. There is limited evidence of an association between compensation status and increased pain.

There is strong evidence of an association between legal representation (having a lawyer) and poorer physical function. There is moderate evidence of an association between legal representation and poorer psychological function.

There is moderate evidence of an association between receiving compensation at two years and poorer physical function. There is limited evidence of an association between number of sick days in prior three years, prior claim, and poorer physical function. There is limited evidence of an association between number of sick days in prior three years, prior claim, and increased pain. There is moderate evidence of an association between tort insurance arrangements and poorer physical function.

There is limited evidence of no association between claim type and poorer physical function. There is inconsistent evidence between claim type and increased pain. The evidence levels are summarised in Table 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Evidence levels^*.

https://doi.org/10.1371/journal.pone.0117597.t006

Limitations

An important strength of this review was its conduct according to current guidelines and recommended methods of reporting [22–24, 26–28]. Notwithstanding that, potential studies could have been missed because our search strategy focused on compensation wording in the abstracts. This was mitigated by hand searching of references, personal communication with experts, plus the authors’ existing knowledge of papers to increase the likelihood of including of all relevant papers.

Another limitation was potential measurement error, which is likely when the timing of exposure to a compensation related factor does not occur at baseline. Possible reasons for this include: legislated time periods to lodge a claim; people choosing to submit a claim only if they are not recovering; timing of legal representation; and the interaction between eligibility to claim and different follow up periods. Some authors have chosen not to include compensation status because of the difficulty defining it as a baseline measure [13]. We felt it was impractical to exclude certain compensation related factors and/or studies on this basis. Moreover, definitions of baseline tend to vary between studies.

Interpretation of statistical results was also hindered by selective reporting, particularly poor explanations for final predictive models. Although this would not have changed our conclusions we were not able to explore the reasons behind particular associations.

Implications for policy and future research

Considering the number of studies investigating outcomes following musculoskeletal injury it is of concern that many do not include compensation related factors as a potential confounder given the evidence available. Compensation schemes are diverse and contextual which makes interpreting the evidence based on existing data classifications challenging. The development of a compensable reporting framework would be valuable and has been recommended by others [10, 18, 75, 76]. Minimum reporting should include claim lodgement, claim acceptance, claim type, legal representation, entitlements, claim duration, litigation, sick leave, and weekly benefits paid for time off work if applicable. The timing of measures should be documented. For example: when legal representation or claim acceptance was obtained. A description should be provided of the legislative framework. Collaboration between researchers and the legal profession may also assist to untangle the complexities of scheme design particularly for future policy relevant research between and within jurisdictions [76, 77].

It is imperative for researchers to consider reverse causality bias [18, 78]. If present, this could be mitigated by risk assessments to identify triggers for poor recovery and facilitate early intervention. Furthermore, reducing compensation related psychological stressors such as: poor claims information and management; claim delays; perceived injustice; and numerous medico-legal assessments could improve injury recovery [74, 79, 80]. These stressors have also been linked to increased legal representation, delayed claim settlement and increased health care utilisation [15, 71, 73, 81].