OSH-ROM denotes Occupational Safety and Health-Read Only Memory. aOf hand-searched articles, all were from included articles and none were from relevant reviews.
CI denotes confidence interval; RR, relative risk.
Angela G. E. M. de Boer, Taina Taskila, Anneli Ojajärvi, Frank J. H. van Dijk, Jos H. A. M. Verbeek. Cancer Survivors and UnemploymentA Meta-analysis and Meta-regression. JAMA. 2009;301(7):753–762. doi:10.1001/jama.2009.187
Author Affiliations: Coronel Institute of Occupational Health, Academic Medical Center, Amsterdam, the Netherlands (Drs de Boer, Taskila, van Dijk, and Verbeek); Primary Care Clinical Sciences, School of Health and Population Sciences, The University of Birmingham, Birmingham, England (Dr Taskila); Finnish Institute of Occupational Health, Good Practices and Competence, Statistical Services, Helsinki, Finland (Dr Ojajärvi); and Finnish Institute of Occupational Health, Knowledge Transfer Team, Cochrane Occupational Health Field, Kuopio, Finland (Dr Verbeek).
Context Nearly half of adult cancer survivors are younger than 65 years, but the association of cancer survivorship with employment status is unknown.
Objective To assess the association of cancer survivorship with unemployment compared with healthy controls.
Data Sources A systematic search of studies published between 1966 and June 2008 was conducted using MEDLINE, CINAHL, EMBASE, PsycINFO, and OSH-ROM databases.
Study Selection Eligible studies included adult cancer survivors and a control group, and employment as an outcome.
Data Extraction Pooled relative risks were calculated over all studies and according to cancer type. A Bayesian meta-regression analysis was performed to assess associations of unemployment with cancer type, country of origin, average age at diagnosis, and background unemployment rate.
Results Twenty-six articles describing 36 studies met the inclusion criteria. The analyses included 20 366 cancer survivors and 157 603 healthy control participants. Studies included 16 from the United States, 15 from Europe, and 5 from other countries. Overall, cancer survivors were more likely to be unemployed than healthy control participants (33.8% vs 15.2%; pooled relative risk [RR], 1.37; 95% confidence interval [CI], 1.21-1.55). Unemployment was higher in breast cancer survivors compared with control participants (35.6% vs 31.7%; pooled RR, 1.28; 95% CI, 1.11-1.49), as well as in survivors of gastrointestinal cancers (48.8% vs 33.4%; pooled RR, 1.44; 95% CI, 1.02-2.05), and cancers of the female reproductive organs (49.1% vs 38.3%; pooled RR, 1.28; 95% CI, 1.17-1.40). Unemployment rates were not higher for survivors of blood cancers compared with controls (30.6% vs 23.7%; pooled RR, 1.41; 95% CI, 0.95-2.09), prostate cancers (39.4% vs 27.1%; pooled RR, 1.11; 95% CI, 1.00-1.25), or testicular cancer (18.5% vs 18.1%; pooled RR, 0.94; 95% CI, 0.74-1.20). For survivors in the United States, the unemployment risk was 1.5 times higher compared with survivors in Europe (meta-RR, 1.48; 95% credibility interval, 1.15-1.95). After adjustment for diagnosis, age, and background unemployment rate, this risk disappeared (meta-RR, 1.24; 95% CI, 0.85-1.83).
Conclusion Cancer survivorship is associated with unemployment.
Improvement in the treatment and prognosis of many forms of cancer has resulted in increasing numbers of cancer survivors.1 The prevalence of cancer survivors is expected to increase in most countries because of an aging population and continued improvements in early detection and treatment of cancer. Therefore, it is important to understand the adverse long-term effects of cancer survivorship on medical, psychological, and social outcomes.2
Quiz Ref IDA significant proportion of cancer survivors experience physical, emotional, and social problems such as fatigue, pain, cognitive deficits, anxiety, and depression, all of which may become chronic.3 These long-term medical and psychological effects of cancer or its treatment may cause impairments that diminish social functioning including the obtainment or retention of employment.4,5 Almost half of all cancer survivors are younger than 65 years. Thus, many cancer survivors are at an age at which cancer and its treatment could alter their employment opportunities.6,7
Many cancer survivors want and are able to return to work after diagnosis and treatment.8 Patients often regard returning to work as indicative of complete recovery4 and regained normalcy.9 Employment is also associated with a higher quality of life.4 The encouragement of cancer survivors to return to work also benefits aging societies economically.
Relatively few studies have assessed the association of cancer survivorship with unemployment. Several mechanisms may operate to promote unemployment after the diagnosis and treatment of cancer. Job discrimination,7,10 difficulty combining treatment with full-time work,11 and physical or mental limitations12 may be major causes of unemployment. Additional factors such as age, sex, and the prevailing unemployment rate in a specific country or region could further increase the risk of unemployment for cancer survivors. A prior study has shown that young adult survivors of childhood cancer in the United States are at a higher risk of unemployment compared with a like population in Europe,13 perhaps due to differences in social security systems and health care insurance.
The purposes of this meta-analysis are to quantify the risk of unemployment among adult cancer survivors compared with healthy control participants, to examine the influence of prognostic factors on unemployment, and to identify groups of cancer survivors at highest risk for unemployment.
Studies were included if they met the following criteria: (1) inclusion of a control group of healthy participants; (2) inclusion of patients diagnosed with cancer working at the time of diagnosis, mean age at diagnosis 18 years or older, and mean average age at time of study 18 to 60 years; and (3) assessment of employment status measured in a follow-up study design.
A series of literature searches was conducted using the electronic databases MEDLINE, CINAHL, EMBASE, PsycINFO, and OSH-ROM. Studies published from 1966 to June 2008 were retrieved, with no language restrictions.
Search terms included employment, unemployment, absenteeism, work, sick leave, vocational rehabilitation, occupational, vocational guidance, job satisfaction, occupation(s), rehabilitation, work disability, return-to-work, sickness absence, disability pension, work ability, or job performance and were combined with survivor(s), late effects or longevity and with neoplasm(s), cancer(s), carcinoma(s), oncology, leuk(a)emia(s), sarcoma(s), lymphoma, melanoma, radiotherapy, or chemotherapy. When available, subject heading terms such as Medical Subject Headings terms were added in all searches. Publications included in the meta-analysis and review articles on employment in cancer survivors were hand-searched for additional references.
The literature search was conducted independently and in duplicate by 2 investigators (A.dB. and J.V.). Each abstract was evaluated independently and in duplicate by 2 investigators (A.dB. and J.V.). Only abstracts reporting empirical studies were selected. Full reports of potentially relevant articles were reviewed by 2 investigators (A.dB. and J.V.). Disagreements were resolved with a third reviewer (F.vD.) through consensus.
Data extraction and quality assessment were performed independently by 3 reviewers (A.dB., J.V., and T.T.) with 2 reviewers per article. Unemployment data were extracted using several methods. Students, homemakers, and individuals retired for reason of age were excluded (if possible) so that only patients eligible for work were included. If applicable, patients who were men and women, working full time and part time, unemployed, and receiving a disability pension were combined. If reported, the number of patients not working because of disability was recorded. Next, if necessary, percentages of unemployed and employed individuals were converted into frequencies. Finally, the number of individuals unemployed and the total number of participants eligible for work in the patient and control groups were entered into the data extraction form. Authors were contacted in case of uncertainty about the data.
The diagnoses were sorted into diagnostic groups. If at least 50% of the patients in a study had a specific diagnosis, the study was included in a diagnostic group for that specific cancer—if not, it was categorized under mixed diagnoses.
The following information was independently extracted by 2 authors (A.dB., J.V. or T.T.) and entered in the data extraction form: country of origin, diagnosis, average age at diagnosis, sex, follow-up time since diagnosis, characteristics of the control group, and source of data (registry-based or not). Disagreements were resolved by discussion.
Methodological quality was assessed using the Methodological Index for Non-randomized Studies (MINORS).14 Quality assessment was conducted independently and in duplicate by 2 investigators for each article (A.dB., T.T. or J.V.). Studies were scored on 12 items: aim of the study, inclusion of consecutive patients and participation rate, prospective data collection, inclusion of employment measure, unbiased assessment of study end points, appropriateness of follow-up time after diagnosis, inclusion of loss to follow-up, prospective calculation of the study size, comparable control group, contemporary control groups, baseline equivalence of groups on several factors, and adequate statistical analysis. Studies received 0 to 2 points for each of these 12 components. The total score ranged from 0 to 24 points. Low-quality and high-quality studies were defined as earning fewer than 16 points and 16 points or greater, respectively, on the MINORS test. An adjusted cutoff point was calculated for a sensitivity analysis. The adjusted score summarized the 4 items most important to our review: prospective data collection, loss to follow-up, comparable control group, and basic equivalence of groups on several factors. The adjusted score ranged from 0 to 8 with a cutoff score for high quality defined as 5 or greater.
The criteria were tested on a separate set of articles to ensure agreement between assessors.
The results of all controlled studies were plotted as relative risks (RR) and then pooled using the inverse variance method. To avoid unnecessary heterogeneity, we formed homogeneous groups of studies according to cancer diagnosis and used these as subcategories. A subgroup meta-analysis was performed for the outcome of unemployment due to disability.
The RR was used to summarize the dichotomous unemployment data. The random effects model, described by DerSimonian and Laird,15 was selected over the fixed effects model because it incorporates within- and between- study variability, which is applicable to this meta-analysis that involves observational studies with inherently more variability than randomized trials. Statistical heterogeneity was assessed with χ2 tests and quantified with the I2 statistic, which describes the percentage of total variation across studies that is attributable to heterogeneity rather than chance.16 I2 values of 25%, 50%, and 75% have been suggested as indicators of low, moderate and high heterogeneity, respectively.16
Quiz Ref IDTo explain heterogeneity between studies and to examine the influence of prognostic factors on unemployment, we performed a Bayesian meta-regression. The following factors were studied: average age of study group participants (≤50 vs >50 years), diagnosis (grouped into 5 categories: testicular, breast, prostate, blood, and other and mixed), geographic area related to social security systems and health care insurance policies (Europe, United States, or other countries), and background unemployment rate (control group unemployment rate at follow-up minus mean unemployment rate from all studies at follow-up). Sex was not included in the analysis because of the high correlation between certain diagnoses and male or female sex. An additional analysis was performed for the studies from the United States vs all other countries.
We analyzed the influence of these variables first in a univariate analysis and next in a multivariate analysis using a Bayesian meta-regression model. The Bayesian analysis yields posterior distributions for the meta-RRs for which the median values are used. These meta-RRs indicate the change in the study RR for studies with the characteristic of interest compared with studies without this characteristic. A Bayesian analogue of a confidence interval is called a credibility interval (CrI), which is a posterior probability interval and also includes knowledge of a prior distribution in addition to the data.17- 19 Noninformative normal priors were used for log RRs and noninformative γ priors for the corresponding precisions.
A sensitivity analysis was performed to account for methodological quality differences. The multivariate analysis using a Bayesian meta-regression model was repeated for the high-quality studies only with the 2 cutoff points for quality. Statistical meta-analyses were performed using Review Manager software (RevMan version 4.2.8, the Cochrane Collaboration, Oxford, England). WinBugs version 1.3 (MRC Biostatistics Unit, Cambridge, England) was used in all Bayesian meta-regression models.20,21 We used the Egger test to assess the possibility of publication bias.22 The power of the random-effects meta-analysis was assessed with the Hedges and Pigott procedure.23
Unless stated otherwise, α was set at .05 and all tests were 2-sided.
A total of 1766 abstracts were retrieved from the electronic databases (Figure 1). Of these, 33 original articles provided data on the employment status of cancer survivors compared with healthy control participants.1,24- 55 Duplicate publication was identified in 7 articles and for these we used data from only the first published article, which resulted in 26 original articles.1,24- 48 Data on 2 to 8 cancer types were reported in each of 3 articles25,26,36 and the separate results for each cancer diagnosis are presented as individual studies. Thus, this meta-analysis included 26 articles reporting results from 36 studies of different cancer diagnoses.
Table 1 and Table 2 summarize characteristics of the 26 included articles. There were no articles from 1966 to 1993, 4 articles from 1994-2001, and 22 articles from 2002-2008. Quiz Ref IDThere were 10 studies of breast cancer, 7 of blood cancers (eg, leukemia and [non]-Hodgkin disease), 3 on testicular cancer, 3 on prostate cancer, 2 on cancers of the female reproductive system (cervical and ovarian cancer), 2 on gastrointestinal cancers, 1 each on melanoma, nervous system tumors, thyroid cancer, nasopharyn geal tumors, sarcoma, and 4 on mixed diagnoses. Fourteen included articles1,25,27,29- 35,38,41,43,46 are from the United States; 2 of each are from Canada,37,45 the Netherlands,40,48 and France42,47; and 1 of each is from Sweden,39 Finland,36 Korea,24 Norway,26 Taiwan,44 and Australia.28 The average age at diagnosis ranged between 40 and 56 years. Mean follow-up time since diagnosis varied from 9 months to 15 years. Excluding articles on breast cancer and cancers of the male and female reproductive system, the percentage of female patients ranged from 26% to 69%. Patients in 12 articles were identified through a hospital-based registry,24- 27,30,34,40,43- 46,48 and patients in 13 articles were identified from a national or regional registry.1,28,29,32,33,35- 39,42,47 In 1 article, the patients were followed-up in a prospective cohort.31 In 12 articles,1,24,28,30,34,35,39,42- 46 students, homemakers, and retired individuals were identified and excluded from analyses.
Study methodological quality is shown in Table 3. The MINORS quality score ranged from 10 to 21 points. Only 10 (38%) of the articles included employment outcomes as part of the main study aim, while it was a secondary outcome in the other studies. A comparable control group with no history of cancer and similar unemployment risk was present in 7 (27%) of the articles. Seventeen (65%) of the articles included a control group with baseline equivalence on age and sex.
The overall meta-analysis of the 36 studies included 177 969 participants, composed of 20 366 cancer survivors and 157 603 healthy control participants. Cancer survivors were more likely to be unemployed than healthy control participants (33.8% vs 15.2%; pooled RR, 1.37 [95% CI, 1.21-1.55]; Figure 2). Additional meta-analysis by diagnosis showed an increased risk of unemployment for survivors of breast cancer (35.6% vs 31.7%; pooled RR, 1.28 [95% CI, 1.11-1.49] Figure 2), gastrointestinal cancers (48.8% vs 33.4%; pooled RR, 1.44 [95% CI, 1.02-2.05]), and cancers of the female reproductive organs (49.1% vs 38.3%; pooled RR, 1.28 [95% CI, 1.17-1.40]). The highest risk for unemployment was identified among survivors of nervous system cancer (RR, 1.78 [95% CI, 1.58-1.99]) and nasopharyngeal cancer (RR, 2.47 [95% CI, 1.67-3.66]), but these involved single studies only. Higher risks of unemployment compared with healthy control participants were not shown among survivors of blood cancer (30.6% vs 23.7%; pooled RR, 1.41 [95% CI, 0.95-2.09]), prostate cancer (39.4% vs 27.1%; pooled RR, 1.11 [1.00-1.25]), and testicular cancer (18.5% vs 18.1%; pooled RR, 0.94 [95% CI, 0.74-1.20]).
Seven studies reported1,27,30,38- 40,42 unemployment because of disability. Subgroup meta-analysis showed a higher risk for unemployment because of disability for cancer patients compared with control participants (RR, 2.84 [95% CI, 1.91-4.20]).
Table 4 summarizes results of the univariate and multivariate Bayesian meta-regression analyses. The crude univariate meta-RR for the United States compared with Europe shows a higher unemployment risk for cancer survivors in the United States (meta-RR, 1.48 [95% CrI, 1.15-1.95]). The crude univariate meta-RR for cancer survivors in the US studies compared with all other studies in Europe, Canada, Australia, and Asia was 1.36 (95% CrI, 1.05-1.79). The crude meta-RRs of unemployment risk were higher for all cancer diagnoses compared with testicular cancer (meta-RRs, 1.21-1.58), but findings were not statistically significant. No differences in unadjusted unemployment risk or unemployment risk adjusted for country, age, diagnosis, and background unemployment rate were identified for older patients (>50 years of age) compared with younger ones (meta-RR, 0.99 [95% CrI, 0.70-1.40] and meta-RR, 1.08 [95% CrI, 0.70-1.69], respectively). Studies with a low background overall unemployment rate showed lower risks for unemployment among cancer survivors compared with studies with higher background unemployment rates (meta-RR, 0.24 [95% CrI, 0.11-0.54]).
In the multivariate analysis, the unemployment risk for cancer survivors in US studies was not different compared with European studies after adjustment for diagnosis, age, and background unemployment risk (meta RR, 1.24 [0.85-1.83]). After adjustment for country, diagnosis, and age, the adjusted meta RR for the effect of a low background unemployment rate was not statistically significant at 0.38 (95% CI, 0.11-1.27).
When 11 low-quality studies identified with the MINORS score were omitted, the overall RR for unemployment among cancer survivors compared with healthy control participants decreased to 1.22 (95% CI, 1.11-1.34). In the sensitivity analysis, which included the 25 high-quality studies only, none of the adjusted meta-RRs were statistically significant (Table 4). Meta-RRs, similar to the adjusted meta-RRs of the high-quality studies in Table 4, were found in a second sensitivity analysis with 26 high-quality studies identified with the adjusted MINORs score.
Results of the Egger test revealed there was no publication bias (P = .89). Results of the Hedges and Pigott procedure for power analysis showed that the power of our random-effects meta-analysis ranges from 0.77 to 0.98 for detecting a mean effect size above zero, with a mean effect size of 0.31 (based on the relative risk of 1.37) of the 36 studies.
Quiz Ref IDResults of this meta-analysis show that cancer survivors are 1.37 times more likely to be unemployed than healthy control participants. Increased risks for unemployment were identified for survivors of breast cancer, gastrointestinal cancers, and cancer of the female reproductive organs. Survivors of blood cancer, prostate cancers, and testicular cancer did not have a higher unemployment risk. Survivors in the United States were at 1.5 times higher risk of unemployment than survivors in Europe. After adjustment for diagnosis, age, and background unemployment risk or after omitting the low-quality studies, this association disappeared. No differences in unemployment risk related to age were found, but in studies with a relatively low background unemployment rate, the risk for unemployment for cancer patients was lower compared with healthy control participants than in studies performed in countries with a relatively high background unemployment rate.
This is the first study to systematically summarize the risk of unemployment for adults who survived cancer. A strength of our study was that we were able to extract distinct unemployment figures from studies meeting our inclusion criteria. In contrast to earlier studies, we summarized results in one overall estimate of unemployment risk. In addition, we assessed associations of prognostic factors such as diagnosis and country of origin with unemployment risk in a meta-regression analysis. Finally, we used a valid and reliable index for assessing methodological quality.
A limitation of this meta-analysis is that the quality of included studies varied depending on study design and objectives. In some studies, unemployment rates were secondary outcomes. Other studies aimed to find the best estimate of the unemployment risk in cancer survivors and included an age- and sex-matched control group. Furthermore, heterogeneity in the meta-analyses remained after dividing studies according to cancer diagnoses and country. Even though the meta-regression pointed at a strong influence of country of origin and diagnosis, the association of country of origin with unemployment was not significant after adjusting for covariates, and the association of cancer type with unemployment was not significant after low-quality studies were omitted from analyses. Age did not have a clear association with unemployment risk. We assume that the remaining heterogeneity is caused by factors other than those characteristics that we could measure with the information given in the studies. For instance, access to health insurance associated with employment may vary between countries, attention to unemployment issues may vary between treatment centers and hospitals, and occupational health services may not be available in all countries.
Even though the risk of unemployment is higher in cancer survivors, the unemployment rate among survivors across all studies was 34%. Thus, most cancer survivors remain employed after their cancer diagnosis, but there is opportunity for improving unemployment rates among cancer survivors.
Several possible mechanisms may explain the higher unemployment rates among cancer survivors. First, cancer survivors may be less available for the labor market. To address this, we excluded all homemakers, students, and retired individuals from our analyses when possible, in an attempt to include only those who were available for and seeking employment. Second, health is an important determinant of labor participation even though this relationship might be complex.56 Eighty-six percent of cancer patients (97/113) vs 76% of unemployed control participants (100/131) stated that the decision to stop working was their own in the study of Maunsell et al.37 In the same study, 47% (53/113) of the unemployed breast cancer survivors stated that they were unemployed because of health reasons compared with 18% (24/131) of the unemployed control participants. In 5 other studies,1,24,32,33,42 reasons for unemployment were provided. More often for patients than for control participants, reasons for unemployment were physical limitations, cancer-related symptoms, or both. Furthermore, voluntary unemployment is not likely unless patients have other resources for income,57 which is not the case for most cancer survivors. Finally, in the 7 studies1,27,30,38- 40,42 in our review that measured unemployment and disability, the relative risk of receiving a disability benefit or otherwise being disabled for work was almost 3 times higher for survivors compared with control participants. Therefore, the mechanism behind the higher unemployment rate among cancer survivors is likely to be a higher disability rate.
Except for blood cancers, the results of our meta-analysis are similar to those of earlier cohort studies on predictors of return to work,58- 62 in which patients with breast cancer58,59 or tumors of the head and neck59,60 showed more difficulty and men with testicular cancer59- 62 less difficulty returning to work than patients with other cancer diagnoses.
An interesting finding was a higher unadjusted risk for unemployment among cancer survivors in the United States compared with Europe. The effects of health on labor force participation are likely to be socially determined because they are the result of the interaction of the labor market, job search behavior, economic incentives, and health insurance.56 Almost all of these factors differ between the United States and Europe and could explain the difference. However, after adjustment for diagnosis, age, and background unemployment risk, or after exclusion of low-quality studies, the higher unemployment risk for cancer survivors in the United States vs Europe disappeared. Therefore, the difference can also be due to other determinants or confounding by low-quality studies. Half of the studies from the United States (8 of 15) and 1 from Europe (out of 16) were of low quality. Most of the excluded American studies had higher risk estimates and this might have caused the reduction of the adjusted risk in the sensitivity analysis.
More and better research on the work impact of cancer is necessary to inform the decisions of cancer survivors and the clinicians who provide their treatment. Better estimates of the risk of unemployment from high-quality studies are needed to enhance the identification of prognostic factors and vulnerable subgroups for unemployment among cancer survivors. Future research should involve large patient samples, include matched control groups, focus on different cancer diagnoses, compare differences between countries, and analyze a number of prognostic variables including socioeconomic and cultural factors on unemployment rates.
Apart from the effects on employment, there are probably long-term effects of cancer on work ability,63 work capacity,64 and wage losses64,65 for a large group of survivors. Quiz Ref IDEmployment outcomes can be improved with innovations in treatment and with clinical and supportive services aimed at better management of symptoms, rehabilitation, and accommodation for disabilities. Moreover, workplace interventions are needed that are aimed at realizing workplace accommodations and paid sick leave during treatment. The development and evaluation of such interventions is urgently needed because they could mitigate the economic impact of surviving cancer and improve the quality of life for survivors.
In conclusion, cancer survivors are at an increased risk of unemployment, especially survivors of breast cancer, gastrointestinal cancers, and cancers of the female reproductive organs and cancer patients living in countries or times with relatively high unemployment rates. Development of interventions involving clinicians and other professionals to enhance participation in the work life of cancer survivors is needed.
Corresponding Author: Angela G. E. M. de Boer, PhD, Coronel Institute of Occupational Health, Academic Medical Center, PO Box 22700, 1100 DE Amsterdam, the Netherlands (firstname.lastname@example.org).
Author Contributions: Dr de Boer had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: de Boer, van Dijk, Verbeek.
Acquisition of data: de Boer, Taskila, Verbeek.
Analysis and interpretation of data: de Boer, Taskila, Ojajärvi, van Dijk, Verbeek.
Drafting of the manuscript: de Boer, Taskila, Verbeek.
Critical revision of the manuscript: de Boer, Ojajärvi, van Dijk, Verbeek.
Statistical analysis: de Boer, Ojajarvi, Verbeek.
Administrative, technical, material support: van Dijk.
Study supervision: de Boer, van Dijk, Verbeek.
Data extraction and quality assessment: de Boer, Taskila, Verbeek.
Financial Disclosures: Dr Taskila reports that her contribution was funded by the Finnish Work Environment Fund. The other authors report no financial disclosures.
Role of the Sponsor: The Finnish Work Environment Fund is a nonprofit governmental funding organization. The Finnish Work Environment Fund had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.