Behavioral,Psychological,Educational,and Vocational Interventions to Facilitate Employment Outcomes for Cancer Survivors: A Systematic Review

1.1.1 How the Intervention Might Work

It was anticipated that the literature would identify a broad set of interventions with behavioral, psychological, educational, or vocational components. Pathways of effects to the outcomes were expected to vary widely along with the setting, training of the facilitator, and whether delivered to individuals or groups by individuals or teams.

Approaches to addressing strain on individual and interpersonal resources---one of the domains Mehnert, de Boer, and Feuerstein (2013) identified---would include vocational components. Survivors are four times more likely to be employed when they receive employment assistance and support, such as job-hunting services or on-the-job training (Strauser et al., 2010). Basic components of interventions that United States-based vocational rehabilitation (VR) agencies use generally include diagnosis and vocational assessment, counseling, training, provision of accommodations, job placement, and post-employment services (Waddell, Burton, & Kendall, 2008). This review was designed to include studies of interventions that measured outcomes related to participants' search for new jobs but often dealt with issues of job retention. In an example of how VR, as a multi-component intervention, can be tailored to meet the needs of these two groups, Chiu and colleagues (2013) studied recipients of state VR services who had cancer, and their chief finding was that those who were employed used different VR services than those who were not employed: “While some services were indeed used by a large portion of all clients such as assessment, diagnosis and treatment of impairments and VR counseling and guidance, which were major needs by over half applicants in the current study---other services were contingent on the clients' employment status” (Chiu et al., 2013, p. 7). Employed clients needed services to support work adjustment and accommodation, such as diagnosis, treatment, “rehabilitation technology, disability-related augmentative skills training, technical assistance services, on-the-job supports and basic academic, remedial or literacy training” (Chiu et al., 2013, p. 7). Especially because there were significantly lower educational levels compared to the employed group, unemployed clients needed a wide variety of vocational rehabilitation services such as the search and placement process to obtain a job, and then job readiness training. A pilot randomized controlled trial of VR services among women with breast cancer following surgery is currently underway in the United Kingdom, though in these countries VR is more heavily oriented toward interventions that focus on health (Kyle et al., 2011) and psychological counseling.

Approaches to addressing health and well-being include components targeting behavioral change and/or alleviation of physical symptoms or emotional issues, with a focus on symptom reduction and improvement in related quality of life. A review of psycho-social interventions in oncology noted that treatment options for cancer patients vary due to the diversity among types of cancer and their treatment options, but that they included “counseling, cognitive-behavioral methods, information and educational treatments and complementary therapies” (Whatley & Milne, 1998, p. 1). Similar to the studies located in a meta-analysis of psychosocial interventions with adult cancer survivors conducted by Meyer and Mark (1995), these intervention studies did not measure employment. Another intervention targeting barriers in this domain is a tool currently being developed and evaluated at the University of Wisconsin-Madison called “Work ability Improvement through Symptom management and Ergonomic strategies” (WISE). The tool is a website that guides breast cancer survivors through questions that help determine how to address on-going symptoms, particularly shoulder pain, and improve ergonomics at their offices (Garrett, 2012; National Rehabilitation Information Center, 2012, p. I-37). Educational interventions also seek to overcome barriers in this domain, such as a group educational and discussion group intervention for men with prostate cancer (Lepore, Helgeson, Eton, & Schulz, 2003). Another example is the cancer-related fatigue intervention trial CAN-FIT, that sought to reduce severity of fatigue among survivors who have had radiotherapy by sharing a handbook, presentations, a goal-setting sheet, and progress diary that aimed to increase participants' knowledge about radiotherapy side effects and strategies to reduce fatigue (Purcell, Fleming, Burmeister, Bennett, & Haines, 2011).

Approaches to addressing barriers to employment that express themselves in work environments are primarily educational. Several organizations, for example including some that belong to the United States-based National Cancer Legal Services Network (www.nclsn.org), provide training and information to both employers and survivors. Publications and other information on various Web sites offer information about cancer-related employment issues and how to address them through greater awareness of cancer survivors' needs for accommodation. Examples of such informational resources are published by Cancer and Careers in the “At Work” section of their web site (see www.cancerandcareers.org/en/at-work) and Hoffman (2012). The American Cancer Society also has an employer-oriented publication (American Cancer Society, 2014). While increasingly common, to date the efficacy of this information provision has not been studied formally.

Given that the typical intervention includes more than one component targeting barriers to employment that straddle the domains outlined above, it is not surprising that a common approach to return-to-work interventions for cancer survivors (a) emphasizes the involvement of multidisciplinary teams (Fleischman, Retkin, Brandfield, & Braun, 2006; Retkin, Antoniadis, Pepitone, & Duval, 2013; Vonk Noordegraff et al., 2012) or (b) at least seeks to improve communication among employers, survivors, and medical providers (Nieuwenhuijsen, Bos-Ransdorp, Uitterhoeve, Sprangers, & Verbeek, 2006; Tamminga, de Boer, Verbeek, & Frings-Dresen, 2010). Many interventions seek to meet a diverse set of survivors' needs, using a variety of intersecting mechanisms.

3.2.1 Participants

The participant samples were required to be (a) adults aged 18 years or older and (b) cancer survivors (i.e., had a past or present cancer diagnosis which occurred while the individual was aged 18 years or older). Studies of populations that included, but were not limited to, cancer survivors were not excluded if the employment outcomes of the participants who were cancer survivors were reported independently from those of other participants. Studies of adults who were survivors of pediatric cancer were excluded, since these individuals may have participated in interventions as children, such as high school transition-to-work programs to which adult-onset populations could not have participated in, but which could affect employment outcomes. Study participants eligible for inclusion were individuals diagnosed with any type of cancer. Study participants with co-morbidities were not excluded.

Participants not employed at the time of the study intervention were the focus of this review as return-to-work and gainful employment were primary outcomes; however, employment maintenance is an important concern and individuals who were employed prior to an intervention study were not excluded in this review. Reviewers did not exclude studies in which the participant pool included both participants who had an employment history and those who did not.

3.2.2 Intervention

Included studies addressed the effectiveness of a behavioral, psychological, educational, or vocational intervention or component of an intervention for cancer survivors that facilitated their employment outcomes, including employment initiation, return-to-work, or decreasing absenteeism and use of work disability or sick leave.

Eligible interventions addressed social, behavioral, cognitive, or specific employment skills related to the acquisition and maintenance of employment among the study participants. Interventions could involve relatively specific and structured experiences of VR designed to support employment placement, for example, providing guidance in completion of applications, résumés, and engaging in interviews; shaping of work skills and appropriate employment setting social skills; or teaching of appropriate work-related communication skills. Furthermore, interventions could address more general psychological or behavioral aspects of functioning that can facilitate employment, such as coping skills. Examples are behavioral treatments such as self-care behaviors to reduce fatigue; behavioral therapies that help survivors cope with their scars or other issues associated post-surgery; psychological interventions through individual or telephone counseling that assist with adjusting to a cancer diagnosis; lecture-based educational interventions on cancer survivorship; and, vocational interventions through supported employment from a VR agency that provide job supports, interview training, etc.

Interventions could be multidisciplinary, incorporating physical and medical treatment and educational training. For example, an intervention could include multiple components such as a vocational training workshop, exercise, and yoga.

Interventions to be included in this review were often divergent in their scope and area of focus, e.g., including components that promoted psychological, physical, and work-oriented outcomes because of the particular needs of the subject group. This was expected due to the wide variation in health-related outcomes related to improving cancer survivors' overall well-being. To accommodate this need and appropriate variation in interventions, studies for inclusion addressed psychological, behavioral, medical and/or skill-oriented interventions with an employment outcome. Therefore, if there were two studies that evaluated the impact of VR, one with an employment outcome and the other with a quality of life outcome, only the first was included, and the latter excluded. Interventions must have included an element apart from medical or physical treatment (e.g., exercise, surgery, pharmaceutical treatment). One reason for excluding interventions with solely medical components is our focus on VR agencies as our primary audience and, as a result, focused on programs that rehabilitation organizations can feasibly implement. Interventions of any length or duration were included provided adequate description was given.

3.2.3 Research Design

Included studies used a randomized controlled trial (RCT) design, quasi-experimental equivalent and non-equivalent comparison design, or quasi-experimental design that employed regression discontinuity. Quasi-experimental designs with equivalent groups formed by matching or equating, or non-equivalent groups without matching or equating, involved, for example, comparing the employment rate of a treatment group compared with the employment rate of a general population of cancer survivors that did not receive the intervention.

3.2.4 Outcome Measures

Eligible studies provided evidence for the effect of the intervention on employment status and/or related outcomes such as disability onset; of time out-of-work (i.e., number of leave days taken, including sick, disability, or vacation); and/or differences in rates of employment between the intervention and comparison groups. Various other measures were used to estimate the rate of employment such as wage-earning or hours worked. These employment-related outcomes are commonly studied in the return-to-work literature and provide a more fine-grained understanding of the impact of cancer on employment (Murphy, Markle, Nguyen, & Wilkinson, 2013).

Consistent with many aspects of the definition that United States federal agencies such as the Bureau of Labor Statistics and United States Census use, this review considered study participants to be ‘employed’ if they have done at least one hour of work per week. This job could be as a paid employee or “in their own business, profession, or on their own farm.” Also considered ‘employed’ are those who were not working but who “had jobs or businesses from which they were temporarily absent because of vacation, illness, bad weather, childcare problems, maternity or paternity leave, labor-management dispute, job training, or other family or personal reasons, whether or not they were paid for the time off or were seeking other jobs” (Bureau of Labor Statistics, 2008). Because this definition includes a rather attainable threshold for “gainful” employment, the included studies operationalized employment in this regard.

3.2.5 Publication Status

Published and unpublished studies were included in the evidence pool.

3.2.6 Country of Origin and Language of Publication

Studies that were conducted in any country were eligible. We did not exclude studies reported in languages other than English, but we did not specifically search for non-English literature; however, we did search selected international databases. Non-English language studies that were retrieved or reviewed required the reviewers to obtain translation assistance from native speakers (e.g., Bottcher, 2013).

3.3.1 Electronic Searches

Studies were identified using electronic search techniques of 27 computerized databases (last search occurred in November 2013). We consulted database thesauri, where they were available, to assure that the universe of appropriate synonyms had been included in the intervention and outcome search term categories. Search terms and search strategies were modified to fit individual databases.

Databases searched included:

Academic One File

3.3.2 Search Terms

We included literature published between 1973 through August 2013. The rationale for the start date was that, during 1973, the United States Congress approved the Rehabilitation Services Act, catalyzing the development of VR programs and efforts to accommodate people with disabilities in the workplace. There has been some attention to occupational health internationally for over a century; for example, the International Commission on Occupation Health was founded in 1906. This date was well before scholarly work on disability management began to develop. For example, the first conference of the International Disability Management Standards Council was not held until 2002. Similarly, the field of cancer survivorship studies is relatively recent because the population of cancer survivors who had employment-related issues was small until medical advances of the late 20^th century (Hewitt, Greenfield, & Stovall, 2006).

The search of databases used four sets of keywords that pertain to the population, intervention, outcomes, and study design. Search strings' keywords were customized to the particular thesaurus of each database. Keywords were connected with “and”/“or” when searching titles and abstracts. Search terms were truncated to include variations in word endings, spellings, and database indices.

The following is an example of the types of terms. Terms from the four categories were connected with “or” within each category and by “and” between categories.

Population: cancer, cancer survivor, neoplasm, leukemia

A sample search strategy is:

((cancer* OR “cancer survivor” OR neoplasm* OR leukemia*) AND (interven* OR model* OR program* OR practice* OR train* OR “vocational rehabilitation” OR accommodat* OR “occupational therapy”) AND (employ* OR “return-to-work” OR job OR wages OR salar*) AND (experiment* OR “control group*” OR random OR effect*))

For more information on the search strategy, see Appendix A: Documentation of Search Strategies for the Systematic Review.

3.3.3 International Contacts

Our efforts to find studies from outside the United States included searching in several non-United States and international databases. This yielded studies that were reviewed in Stage 1 of the title/abstract review procedures as well as one included study (Granstam-Bjorneklett et al., 2013).

3.3.4 Grey Literature

Grey literature that is identified through electronic searches was submitted to the same inclusion criteria as other studies. A time range for these types of studies was not specified in order to maximize consideration of all relevant grey literature. Reference lists from other systematic reviews and individual studies were searched for potential studies to consider for inclusion.

We inquired with two prominent researchers in the field of vocational rehabilitation of cancer survivors about any unpublished reports or completed research activity that may have a pending report. We searched the reference lists of identified articles (ancestry search), which might have been helpful, for example, in locating dissertations and theses not identified by our database searches. Also, a search of popular search engines was conducted. Using the search strategy described earlier, we used Google, Google Scholar, and Yahoo! to uncover any relevant web materials or unpublished studies not accessible through electronic databases. In addition, searching ProQuest Dissertations and Theses allowed for opportunities to uncover relevant unpublished doctoral dissertations and masters theses.

3.3.5 Cross-referencing of Bibliographies

The references in relevant journal articles, systematic reviews, and other reports of research results were reviewed for new additions to our study pool.

3.3.6 Conference Programs and Relevant Associations

Recent conference programs and conference syntheses from relevant associations and conferences were used to identify unpublished studies eligible for review inclusion. Programs and organizational resources were searched in 2012, using the most recent conference programs or those available online. Professional organizations/events that were reviewed included:

American Cancer Society

3.4.1 Citation and Abstract Stage

At Stage 1, the decision for advancing the retrieved citations and abstracts to the full text stage retrieval was made independently two reviewers, discussing discrepancies to reach consensus. Decisions were based on meeting two criteria from the following questions (items a, b, c, or d) or a designation by a reviewer of ‘unsure’ (item e):

Are the participants identified, described, and defined as cancer survivors?

3.4.2 Full-Text Level

At the Full-Text Stage 2 level, full texts of all citations advanced from Stage 1 were obtained and coded for an inclusion/exclusion decision. The decision for advancing the retrieved full-text studies to an inclusion status was made by two reviewers for each study, independently evaluating each study. An inclusion decision for advancement to the coding stage of the process required that a study met all the criteria presented earlier. Inter-rater reliability was established prior to initiating coding activities, minimizing coding disagreements. Coders demonstrated agreement in an initial set of 2 studies at a 95% rate of agreement. When differences did arise, resolution occurred through discussion and agreement of the two reviewers.

At the Full-Text Stage 2 level, the two reviewers also recorded all excluded studies and the reason for exclusion independently. For more information see Appendix B: Reasons Stage 2 Studies were Excluded from the Systematic Review.

When multiple studies used the same sample or outcome data, the study providing the most complete information focusing on our desired intervention outcome was selected for inclusion.

Other data for extraction and coding from the primary studies included: publication source, subject characteristics, sample source, employment setting, intervention characteristics, type of employment, and outcome measurement. See Appendix C for the coding form.

3.6.1 Within-study Synthesis

The measures of effect were the rate of employment, mean number of leave days taken since disability onset, and mean number of hours worked. If these were not reported directly, they were calculated from the outcomes that were reported.

We employed a shifting unit of analysis approach (Cooper, 1998), which involves coding as many effect sizes from each study as exist as a result of variations in characteristics of the manipulation, sample, setting, and outcomes within the study. However, when calculating the overall effect size, the multiple effect sizes were averaged to create a single effect size for each study. The shifting unit of analysis approach maximizes the amount of data from each study without violating the assumption of independent data points.

3.6.2 Across-study Synthesis

The synthesis of effect sizes across conceptually similar constructs were conducted in order to determine the magnitude of the effect when combining similar outcome effects from several studies. However, all cross study synthesis utilized only one comparison effect size per study for any summary synthesis so that no single study outcome is represented more than one time in any analysis. For any group study (e.g., RCT, QED), any Cohen's d effect sizes or means and standard deviations were converted to a log odds ratio effect size so that all effect sizes for all group studies will be presented in the same metric. The following conversion formula was used:

L o g O d d s R a t i o = d π 3 ,

The log odds ratio was synthesized in the final analysis (with results translated back into odd ratio for interpretability). We used a random-effects inverse variance weighting method to synthesize the effect sizes. Confidence intervals were calculated and reported as well.

The magnitude of the intervention effect was calculated using the commonly accepted statistical formulae and dedicated programs available.

3.6.3 Heterogeneity Analysis

For the analysis of dichotomous and continuous data, an assessment of heterogeneity was conducted using τ² (tested against the null hypothesis of τ² = 0) and I² for ease of interpretability (see Higgins, Thompson, Deeks, & Altman, 2003). It is suggested that 50% is a moderate level of heterogeneity. We used a random effects model for heterogeneity analysis.

3.6.4 Sensitivity Analysis

A sensitivity analysis was conducted to assess the impact of a single study on the magnitude of an overall observed effect size. The sensitivity analysis was also to be conducted for the impact of moderating variables (e.g., attrition, type of treatment, missing data, sample size, study design). However, due to lack of data and reporting among the included studies, only study design was tested as a moderator.

3.6.5 Publication Bias

Publication bias was not assessed for published vs. unpublished included studies because there were no unpublished studies in this synthesis. However, we provide a funnel plot to assess the potential for small study bias (see Figure 4.1 and 4.2).

3.6.6 Incomplete Reporting of Study Data

For studies reporting incomplete outcome data (k = 3), we first contacted the senior author of the study and requested the missing information to include in the analysis. Study authors did not respond to our inquiries, so studies with incomplete data where imputation was not possible were excluded.

3.6.7 Subgroup and Moderator Analyses

Subgroup/moderator analyses were not conducted, except for study design (due to interest to a priori theoretical considerations). Since the approach to the moderator analyses were dependent on the available data, our included study sample did not allow us to conduct moderator or subgroup analyses given the small number of studies (<10) available within any category of moderator.

4.2.1 Research Design

Eight of the 12 included studies were randomized controlled trials (Berglund, 1994; Granstam-Bjorneklett, 2013; Hubbard, 2013; Lepore, 2003; Maguire, 1983; Maunsell, 1996; Purcell, 2011; Tamminga, 2013). The other four were quasi-experimental studies, (Bottcher, 2013; Capone, 1980; Gordon, 1980; Sachs, 1980) which compared the treatment groups to a non-equivalent yet comparable comparison group. No discussion of matching procedures was included in the quasi-experimental study reports.

All experimental studies allocated conditions at the level of the individual participant. Two studies used a form of stratification in the randomization process. One study stratified by adjuvant chemotherapy (Granstam-Bjorneklett, 2013) and another by age and cancer type (Tamminga, 2013). No included studies used wait-listed control groups. Five included studies reported data at multiple follow-up points (Berglund, 1994; Capone, 1980; Granstam-Bjorneklett, 2013; Hubbard, 2013; Maunsell, 1996).

4.2.2 Sample Sizes

The number of unique participants from both treatment and comparison groups within the 12 included studies was 2151. The average sample size per study was 139.5 participants (Treatment: 65.7 participants; Comparison: 73.8 participants).

4.2.3 Setting and Recruitment

All included studies recruited their participants through a hospital or clinic. Recruitment often occurred through referrals after chemotherapy or surgery.

A hospital or clinic was typically the primary treatment setting. The few exceptions were studies that incorporated treatments at the hospital as well as the home such as phone interviews/consultations or home visitations (Lepore, 2003; Maunsell, 1996). Another study incorporated the home setting with a vocational rehabilitation agency (Hubbard, 2013). One study administered treatment at a resort (Granstam-Bjorneklett, 2013). One study did not specify the treatment setting in the research report (Purcell, 2011).

4.2.4 Participants

4.2.5 Interventions

Interventions varied widely in program components and duration. Three of the included interventions consisted solely of an educational or training component on topics ranging from physical information such as cancer biology, how to control side effects, and future health concerns, coping and stress, and goal setting. (Berglund, 1994; Lepore, 2003; Purcell, 2011). One study incorporated follow-up phone calls to reinforce the information (Purcell, 2011). Other interventions consisted of a therapeutic or counseling component on issues related to emotional support, behavioral change, holistic self-concept, and information-processing (Capone, 1980; Maguire, 1983). Two interventions paired education and counseling interventions together (Gordon, 1980; Maunsell, 1996). Two interventions combined physical components with education or counseling-related components such as integrating informational sessions with relaxation, qi-gong, and “liberating dance” (Granstam-Bjorneklett, 2013) or combining information, exercise, discussion, and group therapy (Sachs, 1980). In a similar way, another study evaluated a case-management intervention that tailored individual plans to integrate both physical and counseling components (Hubbard, 2013). Only two interventions directly addressed return-to-work or job-related issues by establishing return-to-work plans for cancer survivors combined with informational sessions (Tamminga, 2013) or therapy on job-related issues with physical therapy (Bottcher, 2013). See Figure 2 for a checklist of intervention components across studies.

4.2.6 Outcomes

Employment status or return-to-work was the primary outcome of the review, assessed by 10 studies (Berglund, 1994; Bottcher, 2013; Capone, 1980; Gordon, 1980; Lepore, 2003; Maguire, 1983; Maunsell, 1996; Purcell, 2011; Sachs, 1980; Tamminga, 2013). Two of these studies also included the hours worked as a secondary outcome (Purcell, 2011; Maunsell, 1996). A related primary outcome was the amount of sick leave taken by the participants. Two studies only included sick leave (Hubbard, 2013; Granstam-Bjorneklett, 2013), and two studies included sick leave as well as employment status (Berglund, 1994; Tamminga, 2013).

4.3.1 Selection bias

The majority of randomized studies had relatively low risk of selection bias due to specifying a method of generating an allocation sequence including the use of a biased coin design (Berglund, 1994), envelope (Granstam-Bjorneklett, 2013; Lepore, 2003), Bernoulli probability distribution (Hubbard, 2013), random numbers table (Maguire, 1983), or random number generator (Purcell, 2011). Other randomized studies did not specify the generation of the allocation sequence (e.g., Maunsell, 1996). Other studies (e.g., Capone, 1983; Gordon, 1980) have high risk of selection bias due to lack of randomization.

4.3.2 Performance bias

Only one included study specified that the researcher was blinded to the allocation of subjects (Purcell, 2011). Many studies were judged to have unclear risk of performance bias due to lack of specification of researcher blinding.

4.3.3 Detection bias

Risk of detection bias was low for three studies that specified that assessors were blinded to participants' allocation (Hubbard, 2013; Lepore, 2003; Purcell, 2011). The remaining studies were judged to have unclear risk of detection bias.

4.3.4 Attrition bias

Attrition was generally well-reported (low risk of bias) in the included studies, which often explained attrition due to one or a combination of the following: cancer recurrence, misdiagnosis, decline, refusal to participate or death (Berglund, 1994; Gordon, 1980; Hubbard, 2013; Lepore, 2003; Maguire, 1983; Tamminga, 2013). Some studies acknowledged the presence of attrition in their studies but did not discuss reasons (Bottcher, 2013; Capone, 1983; Granstam-Bjorneklett, 2013). Lastly, two studies did not discuss any attrition (Purcell, 2011; Sachs, 1980), resulting in a rating of unclear risk of attrition bias.

4.3.5 Reporting bias

About half of the included studies had low risk of reporting bias and provided adequate information to derive effect sizes. Some studies did not include data on all follow-up time points such as all outcome data (Berglund, 1994) and sample sizes (Bottcher, 2013; Capone, 1980; Gordon, 1980; Lepore, 2003; Maguire, 1983; Sachs, 1980).

4.4.1 Employment status

Ten studies (Berglund, 1994; Bottcher, 2013; Capone, 1980; Gordon, 1980; Lepore, 2003; Maguire, 1983; Maunsell, 1996; Purcell, 2011; Sachs, 1980; Tamminga, 2013) measured employment status (employed vs. unemployed). Two main measures were used: employment status or return-to-work. Since studies that measured employment status did not distinguish whether participants had prior employment before cancer diagnosis or treatment, we combined these measurements as a single measure of gainful employment status. The results of the studies were synthesized in a random-effects meta-analysis using the odds ratio effect sizes. The weighted mean effect size for employment status was OR = 1.71 (95% confidence interval (CI) = 1.26 to 2.32); P = 0.001), favoring the intervention groups. To better interpret the odds ratio, we converted the mean odds ratio to percentages. We first calculated a baseline employment rate for all the comparison groups across studies, which was 60%. The adjusted odds ratio of 1.71 translates to an employment rate of about 71% for intervention participants, a non-trivial change. See Table 2 for a summary of intervention effects and Table 3 for effects on employment status. See Figure 5.1 for a forest plot. We assessed heterogeneity using the Q-statistic (Q(10) = 15.89, P = 0.10), I² (37.06%), and τ² (0.09). These statistics collectively indicate that the distribution of effect sizes is relatively homogeneous. This homogeneity and the small number of included studies suggest that the examination of moderators is not warranted.

OPEN IN VIEWER

5.1

Forest Plot for Studies with Employment Outcome.

Although we did not locate a large enough number of eligible studies to conduct moderator analyses, we performed an exploratory analysis to distinguish effects of the randomized controlled trials from the quasi-experimental studies. For RCTs, the weighted mean effect size for employment status was OR = 1.44 (CI = 0.99 to 2.09); p = 0.058. The mean odds ratio of 1.44 translates to an employment rate of about 68% for intervention participants, compared to the baseline 60% for comparison group participants described previously. For QEDs, the weighted mean effect size for employment status was OR = 2.18 (CI = 1.32 to 3.60); p = 0.002. The mean odds ratio of 2.18 for the quasi-experimental studies translates to an employment rate of about 77% for intervention participants, compared to the baseline rate of 60% for the comparison group participants. See Figure 2 for a forest plot. Although the mean effect size from QEDs was larger than that from the RCTs, there is no significant difference between the two types of experimental designs (p = 0.19).

OPEN IN VIEWER

5.2

Forest Plot for Studies with Employment Outcome by Research Design (RCT and QED).

This result provides a potentially troubling indication that QED studies may be over-estimating intervention effects on employment whereas the RCT studies, when examined alone, result in a non-significant mean effect size. Therefore, caution needs to be exercised when interpreting the effectiveness of these interventions for increasing employment.

4.4.2 Hours worked

The number of hours worked was measured as an outcome in two studies (Maunsell, 1996; Purcell, 2011). Both studies were randomized controlled trials. The weighted mean effect size for hours worked was OR = 0.89 (CI = 0.22 to 1.52; p = 0.67), favoring the comparison groups. This non-significant result indicates there was no evidence of an effect of the interventions on working hours. See Table 4 for effects on number of hours worked.

OPEN IN VIEWER

5.3

Forest Plot for Studies with Hours Worked Outcome

4.4.3 Sick Leave

Four studies (Berglund, 1994; Granstam-Bjorneklett, 2013; Hubbard, 2013; Tamminga, 2013) measured sick leave. All four studies were randomized controlled trials. Sick leave was measured as the average number of days taken by cancer survivors. The weighted mean effect size for sick leave was OR = 1.18 (CI = 0.81 to 1.71; p = 0.39), favoring intervention participants. This non-significant result indicates there was no evidence of an effect of the interventions on sick leave. See Table 5 for effects on sick leave.

OPEN IN VIEWER

5.4

Forest Plot for Studies with Sick Leave Outcome