Introduction

In the last decade, there has been an increased focus on understanding the determinants of optimal return-to-work (RTW) processes. While physical, social and interpersonal factors in the RTW process have been studied and their importance has been recognized, consistent weaknesses noted about RTW research have been the absence of a theoretical framework integrating the multiple factors involved in RTW and the absence of valid theoretically-derived self-report measures of constructs affecting RTW outcomes [1, 2].

While some models of work disability are available, they are not integrative. The biomedical model, due to its narrow focus on physical illness and physical factors, is insufficient to explain work disability, a phenomenon involving complex social and psychological elements. In response to a growing dissatisfaction with the medical model, a descriptive model emerged, which identified the structural components of RTW, and led to a recognition of the multifactorial nature of work disability [3–5]. The main players involved in the RTW process—the employer/workplace, healthcare provider, insurer, and employee—were identified [3], and the importance of their collaboration was highlighted [6]. Other researchers added a psychiatric component [5], and economic, social and legislative factors [4, 7].

In addition to the structural description of work disability, its temporal aspect was incorporated in phase-specific models of disability. Phase-specific models address the developmental character of work disability [4, 8]. These models highlight the phase specificity of risk factors for work disability, of interactions with the social environment, and of the impact of interventions.

A model is now needed, which integrates both the structural and temporal elements of work disability, and which addresses explanatory mechanisms. As a first step, we propose to apply the Readiness for Change model [9–14] to RTW. The Readiness for Change model is an evidenced-based model identifying social and individual factors impacting on an individual’s ability to initiate and maintain behavior change, in this case the behavior of returning to work after an injury or illness. This model conceptualizes the individual as progressing through stages of change, shifting from the intention to not engage in the given behavior in the foreseeable future to the intention and ability to engage in the behavior in a sustainable fashion.

The Readiness for Change model has been applied to various behaviors. It has received strong empirical support relative to smoking cessation and drug addiction [12–14]. More recently, it has been validated with non-addictive behaviors including pain management [15–17]. The model has excellent predictive validity, particularly with regards to readiness for self-management of pain [18] and for smoking cessation [14]. In the area of injury prevention, a scale of stages of change has been validated regarding safety behavior for farm work [19].

The following aspects of the model are well suited for its application to the behavior of RTW: (1) The model is less focused on individual factors than biopsychosocial models as it can integrate external factors such as the workplace, healthcare, and insurer systems, and recognizes the role of the worker in decision-making (2) It addresses the temporal aspects of work disability (3) It specifies hypotheses regarding relationships among constructs that can be tested empirically (4) It can lead to stage-based interventions, tailored to characteristics of each stage—as an example, based on what is known about the readiness model, individuals in the Precontemplation stage would be best assisted by simply discussing with them the role and meaning of work in their life, without providing any direct advice or intervention to actually have them return to work immediately, while for individuals in the Prepared for Action stage, they would most benefit from concrete and directive interventions to help them return to work (e.g. facilitate contact with employer, provide a work accommodation) (5) It is well suited for research transfer as it is accessible to healthcare providers, insurance providers, and human resource personnel. The Readiness for Change model offers a promising conceptual framework to identify, integrate, and communicate about determinants of optimal RTW stemming from the main players of the RTW process—the workplace, healthcare provider, the insurer, and the worker.

Based on previous applications of the model to other human behaviors, [13, 15–17, 20–27], we developed a scale to assess the stage of readiness for RTW in a cohort of workers who have been absent from work due to a work-related musculoskeletal (MSK) disorder. We report in this paper on the development and the initial psychometric properties of the instrument.

Applying the Readiness for Change Model to Return-to-work Behavior

The Readiness for change model proposes that relative to a given behavior, individuals progress from one stage to the other. Each stage is determined by individuals’ decisional balance, self-efficacy, and change processes about RTW. Individuals can “relapse back” to a previous stage at any point. The five stages have been explained as they would apply to RTW in detail previously [28, 29]. They are briefly described below:

Precontemplation. Workers absent from work due to injury or illness are not yet thinking about initiating behaviors to support a RTW. For a severe injury or illness, it may be appropriate for individuals to temporarily put work issues aside in order to focus exclusively on the physical/mental recovery process.

Contemplation. Workers begin to consider returning to work in the foreseeable future. Although workers are thinking about pros’ and cons’ of returning to work, they are not actively engaged in making concrete plans to do so. The defining characteristic of this stage is ambivalence, where workers do not yet initiate change because they perceive that the benefits of a RTW fail to outweigh possible negative experiences or outcomes, for the time being.

Preparation for action. Workers are actively seeking information regarding a RTW in the near future, testing their abilities to do so, and making a concrete plan to return to work. Workers in this stage are responsive to action-oriented help from external sources, such as the workplace, insurance case managers, or healthcare providers.

Action. Workers are putting a RTW plan into action and going back to work in some capacity. Workers in this stage are responsive to help, and motivated to initiate and maintain RTW. They are at high risk for a recurrence of work absence as they negotiate potential barriers. To the extent that they perceive themselves as being successful in returning to work, they increase their sense of self-efficacy.

Maintenance. Workers use specific skills and social support to identify and face high-risk situations that can trigger a relapse and interfere with successful RTW. They maintain preventive strategies. Self-efficacy increases as workers continue to stay at work, and the ability to do so is integrated in self-identity.

Steps in Development of the Stage Model with New Behaviors

Four steps for the development of the stage model have been proposed by Velicer and colleagues to guide the adaptation of the model to other health behaviors [30]: Theory building, Operationalization, Empirical Testing, and Development of Intervention. As a first step, in this study, we have followed the first three steps proposed:

Step 1—Theory building: defining constructs and the relations between them. Our Readiness for RTW model has been described in a theoretical paper which summarizes how the model integrates the role of workplace, healthcare, insurer, and individual factors as they relate to stages of readiness for RTW, decisional balance, self-efficacy and RTW outcomes [28]. The constructs used to establish the scale’s concurrent validity are based on that model.

Step 2—Operationalization: development of measures for constructs and assessment of psychometric properties, starting with a staging scale, and followed by decisional balance, self-efficacy, and change processes scales. In this study, we focused on the staging scale only. We used two approaches to determine and validate the stages of readiness: The stage allocation approach and the multidimensional approach. In the stage allocation approach, individuals are allocated to one of the stages based on the highest score obtained on the stage scales [31, 32]. It is limited by the fact that in the event of a tie between two stages, it is unclear to which stage the individual should be allocated. However, it offers the advantage of having one stage per person, which facilitates the design and delivery of stage-based interventions. In the multidimensional approach, all individuals have a score on each factorially-derived stage dimension of readiness, rather than allocating only one stage per worker. This approach has been used for readiness for self-management of pain [15, 16, 33]. The disadvantage of this approach is that individuals are not allocated to a specific stage, which limits the practical applications of the approach. However, the multidimensional approach is in line with the emerging literature which suggests that readiness may be best conceptualized as a multidimensional construct [34] and may be a more cautious approach to adopt in the early phases of application of the readiness model to a new behavior.

Step 3—Empirical testing: testing of concurrent and predictive validity. In this paper, we report on the concurrent validity only of a staging scale. Its predictive validity will be addressed in a subsequent paper.

Step 4—Intervention: development of stage-based interventions and testing their stage-specificity. This step will be addressed in subsequent studies.

Objectives and Hypotheses of the Study

Our objective is to address the steps described above of operationalization and empirical testing of the stage model, by conducting exploratory and confirmatory factor analyses, and examining psychometric properties of the newly developed subscales with alpha coefficients and correlations between subscales. Furthermore, concurrent validity is examined using (1) a multidimensional approach where correlations between subscales and relevant constructs are examined and (2) a stage allocation approach where stage-based groups of participants are compared with MANCOVAs, ANCOVAs, and multiple comparisons.

Our hypotheses regarding concurrent validity are that less advanced stages of change would be associated with:

  1. (1)

    More impaired levels of physical health, mental health (including depressive symptoms), and functional ability: Previous research on readiness for self-management of pain shows that individuals in less advanced stages of change report higher levels of pain-related disability and higher levels of depressive symptomatology [16].

  2. (2)

    Higher levels of pain: Previous research on readiness for self-management of pain shows that individuals in less advanced stages of change report more severe pain [16].

  3. (3)

    Higher levels of fear-avoidance about work and physical activity: Previous research with work-disabled workers indicates that fear-avoidance of both work and physical activity is significantly associated with work absence [35].

Methods

Study Design

In this prospective cohort study, a sample of 632 lost-time claimants with work-related back or upper extremity MSK disorders were recruited in cooperation with the Workplace Safety and Insurance Board (WSIB) of Ontario, Canada. Over 70% of workers making lost-time claims in Ontario have a back or upper extremity MSK disorder [36]. Data was obtained from two separate sources: participant telephone interviews (self-report) and the WSIB administrative database. The participants were interviewed by phone at one month (baseline) and six months post-injury. They provided information about their RTW process, their workplace, healthcare provider, insurer, as well as their physical and mental health. Routinely captured claim information, such as site of injury, time on benefits, and claim status, was extracted from the WSIB database and linked to the interview data when written consent for linkage was provided by the participants.

Participants

All 632 participants were Ontario workers who were employed by firms that had workers’ compensation coverage. In Ontario, 69% of the workforce is covered by the WSIB [37, 38]. Eligible participants in the study had filed a lost-time claim for back or upper extremity (UE) work-related MSK disorders. Only accepted or pending WSIB claims registered within 7 days post-injury were included. If the claim status changed to “denied” during the course of the study, we continued to follow the participant. Participants had to be absent from work for a minimum of 5 days within the first 14 calendar days post-injury (based on self-report), and were at least 15 years old at the time of their injury.

Respondents were not eligible to participate if they reported having a fracture, amputation, burn, concussion, electrocution, head injury, hernia, cut, crush injury (without broken bones), or had difficulty speaking or understanding English. In addition, claimants flagged as a security concern in the WSIB database (i.e., had a previous history of violent or harassing behaviour), and those incarcerated or receiving institutional care were also deemed ineligible.

Procedure

Potential participants were identified by running a weekly computer program on the WSIB database of registered claims. Recruitment and eligibility screening occurred in a three-stage process: the WSIB tracking level, the WSIB recruiting level (for details see Bültmann [39] & Franche [40]), and the university-based survey unit level. WSIB staff “trackers” reviewed the files of claimants as a preliminary screen. The “trackers” verified claim status, injured body part(s), nature of injury, and flagged any security concerns. Due to privacy protection standards, the WSIB staff “recruiters” made initial telephone contact with the potential participants selected by the “trackers” to further determine their eligibility.

The recruitment screen involved a language screen, injury screen, and a preliminary work absence duration screen. If the potential participant was deemed eligible to take part in the study, the WSIB “recruiter” asked for permission to provide their contact information to an interviewer from a university-based survey unit administratively and geographically separate from the WSIB. Claimants agreeing to be contacted were sent an information sheet about the study and a consent form. Potential participants were assured that agreement or refusal to be further contacted by the survey unit or to participate in the study would not affect how they were treated by the WSIB, their employer, or healthcare provider.

Once the potential participant agreed to survey unit contact, they were telephoned by an interviewer who explained what participation in the study involved and asked for verbal consent over the phone to complete the questionnaire. Participants were then asked to sign and return the consent form and for consent to link their questionnaire data with their routinely collected WSIB data. The study was approved by the University of Toronto Ethics Review Board.

Measures

Development of Readiness for RTW Staging Scale Items

A pool of stage-specific items were generated by a group of researchers composed of two clinical psychologists, an organizational psychologist, a psychometrist, a graduate student in community research, and one public health researcher, all with expertise in occupational health. Given that according to the model, individuals in the first stages of Precontemplation, Contemplation, and Prepared for action are not actively engaged in the targeted behavior, and that individuals in the later stages of Action and Maintenance are engaged in the targeted behavior, we generated items to be given to two samples of individuals—those working and those not working. Our generation of items was guided by the results of a previous factor analysis conducted as part of a cross-sectional pilot study of injured workers. This previous factor analysis had shown the presence of the three factors—Precontemplation, Contemplation, and Prepared for Action—for workers not working, and of two factors for workers who were working—Uncertain Maintenance and Proactive Maintenance. Alpha coefficients were, however, low, so that in this current study, additional items were generated by the same group of researchers to be re-tested—12 items for those working and 22 items for those not working.

In the previous pilot study, the “Action” items were loading on the two Uncertain and Proactive Maintenance factors. However, it may be difficult to distinguish between Action and Maintenance in the case of RTW. The Maintenance stage has typically been determined for other types of behaviors by having been engaged in the given behavior for more than 6 months, but this duration distinction between Action and Maintenance may not be appropriate in the context of RTW. We therefore retained the concepts of Uncertain and Proactive Maintenance to categorize the items we generated.

Sociodemographic Characteristics

Age and gender were based on self-report. Participants reported their marital status using five categories: married; single; widowed; living with a partner or common-law; separated/divorced. The variable was dichotomized [living with partner; living without partner] for the purpose of the analyses. Participants reported whether or not they had any children/grandchildren under age 18 (yes/no). Level of education and annual personal and family income were assessed through self-report as categorical variables.

Workplace Characteristics

The following information was based on self-report: Full-time/part-time status, temporary vs. permanent employment, number of employees at worksite, hours worked at pre-injury job. The following information was extracted from the WSIB administrative database: industrial sector, occupational classification, coded as white collar, pink collar, blue collar-indoor, or blue collar-outdoor, using the system devised by Gaudette and colleagues [41]; firm size of pre-injury workplace based on the number of full-time equivalent (FTE) workers employed at their firm.

Health and Injury Characteristics

General health was assessed with the Short Form-12 (SF-12), a 12-item version of the SF-36 [42] to measure physical (PCS12) and mental (MCS12) health. The scores range from 0 to 100, a higher score indicating better health. The psychometric properties of the SF-12 are good: coefficients for test-retest reliability, measured over two weeks, are 0.89 (PCS12) and 0.76 (MCS12) [42]. Good internal consistency, validity, and responsiveness have been reported with patients with low back pain [43]. In the present study, the internal consistency was 0.89 (PCS12) and 0.86 (MCS12) at baseline. Functional status was measured with the Roland–Morris Disability Questionnaire and the Quick-DASH. The Roland–Morris is a 24-item scale used to assess functional status in individuals with low back pain [44], with good convergent validity (α = .84 to .92) [45–49]. The Quick-DASH is an 11-item scale that assesses functional ability for upper extremity disorders, with good test–retest reliability (.94–.97) [50]. Scores from both functional status instruments were converted into a z-score. Fear-avoidance was measured with an 11-item scale assessing a person’s beliefs about how physical activity and work affect his/her pain with two factorially-derived subscales of fear-avoidance about work (α = .74 to .92) and about physical activity (α = .52 to .77), with good convergent validity [35, 51, 52]. Primary pain site for participants who reported having pain both in the back and upper extremities was determined by choosing the site associated with the functional status scale with the highest z-score. Current pain level was assessed with one item from the intensity subscale of the Von Korff Pain Scale [53]. Participants were asked to rate their pain “right now” from their workplace injury on a scale from 0 (no pain) to 10 (pain as bad as could be) at the present time. To determine the most severe pain site for participants who reported having pain both in the back and upper extremities, the site pertaining to the scale with the highest z-score was used as the index for functional status.

Work Absence and Compensation Characteristics

The number of work days missed due to the injury at the time of interview was collected by self-report. As well, number of calendar days receiving wage replacement benefits (100% benefits) was obtained from the WSIB administrative database.

Statistical Analyses

First, socio-demographic characteristics of the sample, total and stratified by working status (working and non-working group) were described. Second, selection bias analyses to examine the representativeness of the sample were conducted.

Third, each working status group (working and non-working) was randomly divided in two subsamples, to allow for the cross-validation of a first exploratory factor analysis with a second and different subsample, using confirmatory factor analysis. Two principal component analyses (Exploratory Factor Analysis), with orthogonal and oblique rotations, were conducted on items from the two first subsamples (not working sample = 149 and working sample = 166) to identify the number of emerging dimensions. Both rotations, Varimax (orthogonal) and Promax (oblique), were used because the correlations between emerging factors are initially unknown. If the pattern of correlation is similar using the two types of rotation, the solution from the exploratory factor analyses with Varimax rotation is retained as this method tends to spread the variance equally between factors [54].

Two confirmatory factor analyses (CFA) were performed with the second subsamples (not working sample = 150 and working sample = 167) to test the factor solution stemming from the first exploratory factor analyses. In parallel to the exploratory factor analyses, two models were tested, the non-correlated dimension model and the correlated dimension model. Using the EQS Software [55] for the CFA, the Maximum Likelihood-Robust (estimation method), was utilized to evaluate the two different above mentioned models. The adjustment indices such as the χ2/degrees of freedom ratio, the “NonNormed Fit Index” (NNFI), the “Comparative Fit Index”, the “Bollen Incremental Fit Index” (IFI), as well as the “Root Mean Standard Error of Approximation” (RMSEA) were used to measure the fit of the models. Regarding the cutoff criteria of these indices, the χ2/degrees of freedom ratio had to be close to 5 [56, 57], the NNFI, CFI, and IFI higher than .90 and the RMSEA close to .05, indicating a good fit for the model [58, 59].

Fourth, the internal consistency (Cronbach’s alpha) of the dimensions extracted from the exploratory and confirmatory factor analyses results was computed to examine the internal validity of each dimension.

Fifth, concurrent validity was examined with Pearson correlations between subscales and theoretically relevant constructs. We also created mutually exclusive groups reflecting stages of change and compared these groups on relevant constructs to examine concurrent validity using MANCOVAs, ANCOVAs, and multiple comparisons.

Results

Participation Rates, Timing of Interviews, and Description of the Sample

A total of 632 claimants completed the baseline telephone interview with a participation rate of 61% (for further information on flow of participants, see Franche et al. [40]), consistent with participation rates of other cohort studies of adults with MSK conditions which range between 55% [60] and 63% [61]. Average time between injury date and the baseline interview date was 29.6 days (SD = 6.2; range 15–46 days). The socio-demographic characteristics of the sample, total and stratified by working status, are found in Table 1. WSIB data stratified by working status is only reported for participants who provided written consent for linkage of questionnaire data and WSIB dataFootnote 1.

Table 1 Baseline (1 month post-injury) socio-demographic characteristics from interview and WSIB data for the study sample and by work status
Full size table

Selection Bias Analyses

Selection bias analyses conducted with WSIB data and described in detail elsewhere [39] revealed that participants and non-participants were generally comparable on firm size, industrial sector, and income. However, participants were more likely to be older and female, and participants with accepted claims were more likely to be receiving wage replacement benefits for a longer duration and to have a higher rate of re-instatement of wage replacement benefits at six months post-injury than non-participants, but not at one month post-injury, suggesting that participants had longer work absences.

Exploratory and Confirmatory Factor Analyses

The principal components analyses with Varimax and Promax rotations were conducted on the two subsamples (Not working sample = 149 and Working sample = 166) using an initial pool of 22 items for the Not working sample, and of 12 items for the Working sample. For both samples, Not working and Working results from both rotations were similar, and consequently confirmed the stability of the readiness questionnaire factor solution. Items which did not meet our criteria for saturation on one factor (>.40) and of non-saturation on another factor (<.40), were removed from the scale, resulting in a scale of 13 items for the Not working sample and a scale of 9 items for the Working sample. For the Not working sample, the solution with Varimax rotation explained 60% of the variance for the four emerging factors (Table 2). The first factor—Prepared for Action—Self-evaluative comprised four items, and the three other factors—Contemplation, Precontemplation, and Prepared for Action—Behavioral—included three items each. For the Working sample, 58% of the variance was explained by two factors—Uncertain Maintenance (5 items) and the Proactive Maintenance (4 items) (Table 2). All items loaded significantly on one factor ranging from .64 to .86 (Table 2). Variance explained by each factor was roughly equivalent, ranging between 14 to 18% for the Not working sample, and between 24% and 34% for the Working sample.

Table 2 Factor structure of the Readiness for Return-to-Work scale (RRTW) scale
Full size table

During the development of the scale, items were created to reflect one stage—Precontemplation, Contemplation, Prepared for Action, Uncertain Maintenance, or Proactive Maintenance. This a priori categorization of items is found in Table 2. Twenty of the 22 items (91%) loaded on the appropriate a priori factor (previously hypothesized factors).

Confirmatory factor analyses were carried out on the second subsamples (Not working sample = 150 and Working sample = 167), using the same items stemming from the exploratory analyses (13 items for the Not working group, and 9 items with the Working group), to test the factor structure obtained i.e. four factors for the Not working sample and two factors for the Working sample. For both samples, the models with correlation between factors fitted the empirical data and were more satisfactory (For Not working sample: χ2 = 90.6, df = 59, χ2/df = 1.54, NNFI = .90, CFI = .92, Bollen IFI = .92, RMSEA = .06; for Working sample: χ2 = 57.9, df = 26, χ2/df = 2.2, NNFI = .86, CFI = .90, Bollen IFI = .90, RMSEA = .08 ) than the other models with no correlation between factors (For Not working sample: χ2 = 140.8, df = 65, χ2/df = 2.17, NNFI = .77, CFI = .81, Bollen IFI = .82, RMSEA = .09; for Working sample: χ2 = 67.5, df = 27, χ2/df = 2.5, NNFI = .83, CFI = .98, Bollen IFI = .88, RMSEA = .10 ) . However, the fit indices from the model with ‘correlation between factors’ in the Working sample presented borderline coefficients. Considering the Lagrange Multiplier Test, an error correlation was found between the items #6 and #9. By adding this error item correlation between the two items (re9e6), all fit indices were satisfactory (χ2 = 45.2, df = 25, χ2/df = 1.81, NNFI = .91, CFI = .94, Bollen IFI = .94, RMSEA = .07).

Descriptives and Intercorrelations of Dimensions

Each factor was scored by taking the mean of all items creating that factor. Intercorrelations between these dimensions were examined and the dimensions were significantly and modestly correlated (r = −.34 to r = .39), except for some correlations with the Contemplation subscale which were nonsignificant (r = −.03 with Pre-contemplation; r = −.04 with Prepared for Action—Self-evaluative; r = .10 with Prepared for Action—Behavioral).

Internal Validity

Considering the number of items for each dimension (from three to five items), Cronbach’s alphas were satisfactory: 0.65 for Pre-contemplation, 0.69 for Contemplation, 0.75 for Prepared for Action—Self-evaluative, 0.67 for Prepared for Action—Behavioral, 0.82 for Uncertain Maintenance, 0.67 for Proactive Maintenance.

External Validity

To examine the external validity of the scale, we adopted two approaches to organize the readiness dimensions:

Multidimensional. We used the factorially-derived readiness dimensions described above—four dimensions for the Not working participants, and two dimensions for the Working participants.

Stage allocation. We grouped the participants in one of four Readiness stage-based groups based on their highest scores on the factorially derived readiness dimensions—Contemplation (n = 97), Prepared for action—Self-evaluative (n = 75), Prepared for action—Behavioral (n = 119), Uncertain Maintenance (n = 34), Proactive Maintenance (n = 299). Given that only five individuals had their highest score on the Precontemplation dimension, we excluded these individuals from any readiness group created. For the 18 individuals with a tie between two subscales for the highest score (one between Contemplation and Prepared for Action—Self-evaluative, eight between the two Prepared for Action subscales, nine between Uncertain Maintenance and Proactive Maintenance), we placed them in the least advanced group as in all cases this increased the sample size of the smallest group. Three participants were excluded from the stage allocation approach as their highest score was equivalent on three subscales.

Concurrent Validity of Multidimensional Subscales

Correlations with theoretically relevant constructs were examined cross-sectionally to evaluate concurrent validity of the factorially-derived dimensions (see Table 3). Higher scores on the Precontemplation subcale were significantly but modestly associated in the hypothesized direction with poorer levels of mental health (SF-12), higher levels of depressive symptoms (CES-D), and higher levels of fear-avoidance of work. Higher scores on the Contemplation subscale were significantly but modestly associated with higher levels of depressive symptoms. Higher scores on the Prepared for Action subscales, both Self-evaluative and Behavioral, were associated with lower levels of depressive symptoms, functional disability, fear-avoidance of work and physical activity, and current pain, with more highly significant associations for the Self-evaluative subscale. Higher scores on the Prepared for Action -Self-evaluative subscale were also highly associated with improved physical health (SF-12).

Table 3 Correlations between readiness subscales and relevant constructs (n = 632)
Full size table

Relationships between the Uncertain Maintenance subscale and concurrent validity constructs were of a higher magnitude than those observed for the Proactive Maintenance subscale, and in the same direction. Higher scores on both subscales were significantly associated with poorer levels of physical health, higher levels of functional disability, and higher levels of fear-avoidance for work and physical activity. In addition, higher scores on the Uncertain Maintenance subscale were significantly and strongly associated with poorer levels of mental health, higher levels of depressive symptoms, and higher levels of pain.

Concurrent Validity of Readiness Stage-based Groups

Preliminary Results to Examine Potential Group Differences in Baseline Sociodemographic Variables

Based on their scores on the factorially-derived readiness dimensions, participants were grouped in five stage-based groups, corresponding to one of the following groups: Contemplation (C), Prepared for Action—Self-evaluative (PA-S), Prepared for Action—Behavioral (PA-B), Uncertain Maintenance (UM), Proactive Maintenance (PM). Prior to examining group differences in theoretically relevant constructs, we examined the following variables as potential confounding variables using either χ2s or ANOVAs: Age, hours worked at job of injury, full-time vs. part-time status, gender, living arrangement, presence of children or grandchildren, temporary versus permanent employment, number of employees at worksite, personal income, family income, and education level. Only two variables differed significantly between stage-based groups: marital status and self-reported number of employees at work site. Individuals living alone and individuals working at a worksite with fewer employees were more likely to be in the earlier stages of change (e.g., Contemplation) than those living with someone (χ2 = 10.34, p = .016) and those working at worksites with more employees (χ2 = 24.31, p = .004). These two variables were therefore used as covariates in the subsequent MANCOVAs and ANCOVA to examine concurrent validity of stage-based groups.

To examine concurrent validity of the five Readiness stage-based groups, two MANCOVAs and one ANCOVA were conducted, with living arrangement and number of employees at worksite used as covariates. The two MANCOVAs examined (1) physical and mental health and (2) fear-avoidance. The ANCOVA examined the current pain level. The multivariate F-test (Wilks’ Lambda) for the first MANCOVA was significant (F = 10.42, p < .0001) and the univariate F-tests for the dependent variables were all significant: SF-12 Physical score (F = 12.72, p < .0001), SF-12 Mental score (F = 9.22, p < .0001), CES-D score (F = 18.46, p < .0001) and functional disability score (F = 34.84, p < .0001). The multivariate F-test was significant for the second MANCOVA (F = 11.83, p < .0001) and univariate F-tests for the dependent variables were all significant: Fear-avoidance of work (F = 16.35, p < .0001) and fear-avoidance of physical activity (F = 18.41, p < .0001). The univariate F-test for the ANCOVA for current pain was significant (F = 34.20, p < .0001)

Given the uneven cell sizes of the groups, Tukey–Kramer multiple comparisons were conducted, with the two covariates (for adjusted group means and 95% confidence intervals, see Table 4). The Proactive Maintenance group showed significantly better levels of physical health than the Contemplation (t-test = −5.06, p < 0.0001) and Prepared for Action—Behavioral (t-test = −6.04, p < 0.0001) groups. The Proactive Maintenance group also reported significantly less functional disability than all other four groups—compared to Contemplation (t-test = 10.09, p < 0.0001), to Prepared for Action—Self-evaluative (t-test = 5.91, p < 0.0001), to Prepared for Action—Behavioral (t-test = 7.17, p < 0.0001) and to Uncertain Maintenance (t-test = 5.20, p < 0.0001) groups. The Contemplation group also reported significantly more functional disability than the Prepared for Action—Self evaluative (t-test = 2.75, p < 0.05) and Prepared for Action—Behavioral (t-test = 2.99, p < 0.05) groups. The Proactive Maintenance group showed the most optimal levels of mental health, differing significantly with the Contemplation (t-test = −4.50, p < 0.0001), the Prepared for Action—Self-evaluative (t-test = −3.84, p < 0.0001), and with the Uncertain Maintenance (t-test = −3.79, p < 0.001) groups. On the CES-D, both the Contemplation and the Uncertain Maintenance groups demonstrated more impaired mental health, with the Contemplation group being significantly different from the Prepared for Action—Self-evaluative (t-test = 2.72, p < 0.05), the Prepared for Action—Behavioral (t-test = 3.11, p < 0.05) and the Proactive Maintenance (t-test = 7.51, p < 0.0001) groups. The Proactive Maintenance group was also consistently significantly different from the three other groups in levels of depressive symptomatology, with the lowest CES-D scores—compared to the Prepared for Action—Self-evaluative (t-test = 3.60, p < 0.01), the Prepared for Action—Behavioral (t-test = 4.22, p < 0.001), and to the Uncertain Maintenance (t-test = 4.66, p < 0.0001) groups. In multiple comparisons of current pain levels, again the Proactive Maintenance group showed lower levels of pain than the four other groups—compared to the Contemplation (t-test = 10.18, p < 0.0001), the Prepared for Action—Self-evaluative (t-test = 4.91, p < 0.0001), the Prepared for Action—Behavioral (t-test = 7.30, p < 0.0001), and the Uncertain Maintenance (t-test = 5.07, p < 0.0001) groups. In addition, the Contemplation group reported significantly higher levels of current pain than the Prepared for Action—Self-evaluative (t-test = 3.67, p < 0.01) and the Prepared for Action—Behavioral (t-test = 2.96, p < 0.05) groups. For fear-avoidance of work, the Proactive Maintenance group reported significantly lower levels than all other four groups—compared to the Contemplation (t-test = 6.85, p < 0.0001), the Prepared for Action—Self-evaluative (t-test = 3.17, p < 0.01), the Prepared for Action—Behavioral (t-test = 4.88, p < 0.0001), and the Uncertain Maintenance (t-test = 4.23, p < 0.001) groups. For the fear-avoidance of physical activity, the Proactive Maintenance group reported significantly lower levels than all other four groups—compared to the Contemplation (t-test = 8.10, p < 0.0001), the Prepared for Action—Self-evaluative (t-test = 3.56, p < 0.01), the Prepared for Action—Behavioral (t-test = 4.25, p < 0.001), and the Uncertain Maintenance (t-test = 2.91, p < 0.05) groups. In addition, the Contemplation group reported significantly higher levels of fear-avoidance of physical activity than the Prepared for Action—Self-evaluative group (t-test = 3.21, p < 0.01) and the Prepared for Action—Behavioral group (t-test = 3.60, p < 0.001).

Table 4 Means of dependent variables of readiness groups
Full size table

Discussion

Our findings support the internal validity and concurrent validity of a newly developed 22-item measure of Readiness for RTW stages—the Readiness for Return-to-Work (RRTW) scale—as applied to a sample of injured workers with work-related MSK disorders who made a claim for lost time at work.

To address internal validity of the scale, exploratory and confirmatory factor analyses confirmed the presence of (1) four factors—Precontemplation, Contemplation, Prepared for action—Self-evaluative, Prepared for action—Behavioral, in a sample of workers who were not back at work, and of (2) two factors—Uncertain Maintenance and Proactive Maintenance—in a sample of workers who had returned to work. The amount of variance explained by the factor analysis was significant (>50%), alpha coefficients were equal to or superior to .65, and correlations between factor subscales were low to moderate (between −.34 and .39).

Previous research has had mixed results concerning the presence of five mutually exclusive stages [15, 62]. Similarly, in our study, we did not find the five original stages. When using the stage allocation approach for stage categorization, we found that very few workers endorse Precontemplation items one month after their injury. This may reflect the fact that when adapting the model to a new behavior such as RTW, adjustment needs to be made for the timing of the assessment and for the type of behavior considered. In this study, the assessment may have occurred too late to capture individuals in the Precontemplation stage. As well, workers with more prolonged and life-threatening or degenerative health conditions than MSK conditions, may be more likely to be in the Precontemplation stage than the workers of our sample. A second discrepancy with the original five stages is that in our study, two stages are “split” in two—the Prepared for Action stage and the Maintenance stage. In the discussion of concurrent validity which follows, differences and similarities between the “split” stages factors are highlighted.

Concurrent validity of the RRTW scale was examined using two approaches to determine level of readiness of participants: Multidimensional and Stage allocation. Readiness levels were examined as they relate to the theoretically relevant constructs of physical health, functional disability, pain, mental health, depressive symptoms, and fear-avoidance of work and physical activity.

Using the multidimensional approach, hypothesized relationships between constructs examined and readiness dimensions are generally in the expected direction. The two Prepared for Action dimensions show similar relationships with constructs, with the Self-evaluative dimension showing stronger relationships than the Behavioral one.

Our findings indicate that maintenance of RTW is hard work. Indeed, high levels of both types of Maintenance—Proactive and Uncertain—are associated with poorer physical health, more functional disability, and more fear-avoidance. In addition, individuals engaging in high levels of Uncertain Maintenance also report higher levels of pain and poorer mental health than those engaging in lower levels. Results regarding the Proactive Maintenance are surprising—they indicate that even in the presence of a proactive and positive stance towards going back to work, individuals who endorse high levels of proactive maintenance experience challenging levels of poor physical health and fear-avoidance.

Interestingly, Proactive Maintenance is significantly associated with constructs related to workers’ physical status and physical demands only—SF-12 physical, functional disability, and fear-avoidance of work and physical activity. The impediments to proactive maintenance may therefore be primarily of a physical nature—related to workers’ physical health, and how they are challenged by work and physical activity demands. The absence of a relationship with psychological health appears to be due to the low variability in psychological health observed in individuals scoring high on this dimension, as they are the group in the best psychological health—variability in their CES-D scores and SF-12 mental health, as indicated by standard deviations, is low.

Using the stage allocation approach, as hypothesized, we observe an overall improvement in physical health from less advanced to more advanced stages of readiness. Workers in Contemplation show impaired levels of mental health and of functional disability, and unexpectedly, so do workers in Uncertain Maintenance. Associations with current pain levels are in the hypothesized direction. For fear-avoidance of work, workers in both the Contemplation and Uncertain Maintenance stages show high levels of fear-avoidance. Overall, workers in the Proactive Maintenance stage consistently demonstrate better levels of physical and mental health than other groups, while the Uncertain Maintenance and Contemplation groups are compromised in their mental health, pain, functional disability, and fear-avoidance.

Both Prepared for Action scales are quite similar when using the stage allocation approach, however the two Maintenance scales are not—workers in the Uncertain Maintenance stage show more impaired mental health, more functional disability, more fear-avoidance, and more current pain, than workers in Proactive Maintenance.

Taken together, our findings support the concurrent validity of the newly developed RRTW scale. However, due to the nature of our sample—workers with MSK conditions one month post-injury—few workers are categorized in the Precontemplation group, but the factor analysis nevertheless supports the presence of that factor.

The most striking findings are first that, based on results from the multidimensional approach, the most compromised group in terms of overall health are individuals who return to work and are uncertain about their ability to maintain their return to work (the Uncertain Maintenance dimension). Conversely, the most advantaged group in terms of overall health are individuals who are preparing to return to work and who evaluate that they are ready (the Prepared for Action—Self-evaluative dimension). Based on results from the stage allocation approach, it can also be seen that individuals in the Proactive maintenance group are in relatively good health and show low levels of fear-avoidance. Second, the stages of Contemplation and Uncertain Maintenance are associated with high depressive symptomatology. Regarding Contemplation, this finding is consistent with previous findings demonstrating that depressive symptomatology is associated with longer work disability duration [63–65]. Third, our findings suggest that in workers who return to work, there are two very distinct sub-groups in the Maintenance group—workers who are “struggling successfully” (Proactive), and workers whose struggle is marked by high mental distress and physical pain (Uncertain). Future research should investigate what determines if a RTW occurs in Proactive or Uncertain maintenance conditions.

The “best” approach to determine readiness levels is largely dependent on the reason for assessing it. For research purposes, the multidimensional approach retains the complexity of readiness characteristics, where one individual can score high on one dimension but still have some characteristics of other dimensions. For practice use, the stage allocation approach may present more advantages as it allocates individuals to one readiness stage, which can facilitate subsequent stage-based intervention. Stage-based interventions for RTW have not yet been developed, but general stage-based therapies such as Motivational Enhancement Therapy [66] could be considered for adaptation to the behavior of returning to work. In addition, in our study, while the multidimensional approach highlighted the relationships between readiness dimensions and relevant constructs, the stage allocation made it possible to look at differences in absolute levels of these constructs between the stage-based groups.

Limitations of our study include the fact that factors for the five stages of readiness for change were not found, with two stages being split into two. As mentioned before, this is not uncommon when adapting the Readiness model to a new behavior. In addition, the number of individuals in the Uncertain Maintenance stage was small (n = 34), representing only 5.4% of the entire sample. In addition, for individuals having a tie for highest score on a stage subscale (3%), we placed them in the less “advanced” group as the sample size of the group was smaller. Finally, the cross-sectional design of the study prevents the assessment of how individuals progress from one stage to another over time, an important feature of the theoretical readiness model. Longitudinal studies examining the presence of this progression are needed for further validation of both the readiness model and of the RRTW scale.

A strength of our study is the use of exploratory and confirmatory factor analysis with two separate samples, which lends strong support to the validity of our findings. In addition, the fact that 91% of the a priori items loaded on the correct factor suggests that our scale is consistent with the theory of readiness of change and also reflects a solid face validity of the construct. A methodological strength is that comprehensive selection and consent bias analyses show that there does not appear to be systematic biases present in our analyses. However the generalizability of our study results remains limited with regards to injured workers with shorter duration receiving wage replacement benefits and to younger males.

From a practice and policy point of view, this new measure of readiness for RTW can assist in identifying workers in need of intervention. Indeed, the Uncertain Maintenance group is clearly compromised in physical and mental health, and could potentially benefit from specialized services. Interestingly, this group, representing 5% of the overall sample may correspond partially to the often quoted small proportion of individuals (approximately 10%) with MSK disorders which is responsible for the largest proportion of incurred direct costs (80%) for employers and compensation systems [8, 67–69]. Our future analyses will examine if this group might be more likely to have recurrences of work absence and negative RTW outcomes when examined longitudinally.

Much can be learned by focusing on the groups/dimensions associated with an improved health and fear-avoidance profile—the Proactive Maintenance group and the Prepared for Action—Self-evaluative dimension. It may be fruitful to examine in these groups how potential determinants of work absence, such as type of workplace disability management received, organizational climate, co-worker support, and supervisor attitude with respect to the injured worker are associated with work absence and health. This new measure of readiness for RTW can therefore assist in evaluating the effectiveness of stage-based interventions for RTW.

The next steps in the validation process of the RRTW scale will be to consider longitudinally the predictive validity of the instrument, and to expand instrumentation development to other readiness constructs—decisional balance, self-efficacy, and change processes for RTW. In addition, the validity of the measure should be evaluated with workers presenting with physical and mental conditions other than MSK disorders. Finally, the validity of the Readiness for RTW Scale needs to be explored within the context of intervention research for RTW.

The readiness model has previously been criticized for not integrating sufficiently the social determinants of behavior change [70], resulting in a tendency to over-emphasize the role of the individual in the change process. It is now well accepted in the field that RTW is a highly socially determined process, influenced by interactions with the workplace, insurer, and healthcare systems process [3, 71–73]. It is therefore important to avoid over-emphasizing individual determinants of RTW. The readiness levels of workers needs to be conceptualized as being the result of their complex interactions with the various systems involved in RTW.