Long-term effects of cancer on earnings of childhood, adolescent and young adult cancer survivors – a population-based study from British Columbia, Canada

The population of childhood, adolescent, and young adult cancer survivors has rapidly increased over the last three decades. The percentage of cancer survivors diagnosed between the ages of 0–19, who have lived for 5 years or more post-diagnosis has increased from 56% during the 1970s to 83% for cases diagnosed in the 2000s [1]. This is due to substantial progress in a multimodal of treatment regimens and improvements in supportive care [2]. However when looking at long-term outcomes of the majority of childhood cancer survivors whom are now adults even with the improved survival, they are living with late adverse treatment effects which are shown to compromise their future physical, cognitive and psychosocial health, but also their employment status [3‐5].

The impact of cancer on income level and employment status has shown varying results. Previous studies have reported both negative [6‐13] and neutral [14, 15] consequences of cancer on employment status. Cancer and its treatment effects have been shown to limit cancer survivors’ ability to work, compared to their labor force participation and productivity prior to diagnosis [16].

The most recent studies on labor economics emphasize the impact of employment status on health [17]. Most of the studies were conducted in the US, where health insurance is highly associated with employment status. Survivors, who are still able to work, are forced to stay active in the labor force because the implications of losing their jobs, and ultimately their health insurance plans, may be significant [18;19]. In the US, individuals who lose their employer-provided health insurance due to illness have limited alternative options to obtain insurance. In these circumstances, despite cancer-related morbidities such as physical and emotional pain, the survivor may be forced to stay employed in order to keep his or her health insurance [18, 19].

Most of the previous research on economic outcomes is focused on adult cancer survivors [16, 20], with pre-diagnosis employment. However, little is known about the employment experience and economic well-being of childhood, adolescent, and young adult cancer survivors. These individuals are at a considerably higher risk of substantial morbidity due to exposure to chronic toxicity of the treatment at a young age, prior to entering the labour force [3].

Much existing research on income and employment among cancer survivors is affected by five main methodological weaknesses. First, self-reported data are used to measure the income level of the survivors. Self-reported employment status and income level tends to be systematically mis-report, since many people are uncomfortable about disclosing their true income [21‐23]. Responses may also be subject to recall bias. Second, some of the previous studies are cross-sectional in design; thus, changes in the determinants of income and employment status of cancer survivors cannot be evaluated over time [20;24]. Third, previous studies have used current or short-term household income (i.e., 6–12 months) as opposed to longitudinal income over a longer period of time [20;24]. Fourth, most studies have focused on short-term outcomes closer to the treatment phase of cancer. Finally, only a limited number of studies included a control group from the general population; as such, relative effects were not taken into account [20, 24].

The aim of this study is to examine the relationship between cancer survival and income for childhood, adolescent, and young adult cancer survivors. In this study we analyzed a unique, large, population-based set of longitudinal registries, clinical, and administrative datasets.

Our study used a number of econometric regression models to estimate the impact of cancer and its treatment on the income of survivors over time. Given the richness of the data with respect to the clinical characteristics, we explored the impact of cancer type, treatment type, and disease severity on the economic well-being of cancer survivors. We then investigated whether survivors experienced lower income than the general population based on demographic (i.e., sex, age, and marital status) and area-based (i.e., urban-rural classification and health authority) variables.

We explored whether an income gradient exists based on the level of education of survivors for the sub-sample of survivors we were able to link to the education database. One potential concern with our estimates for education was the direction of causality between level of education and income; it is possible that labour supply participation is affected by the level of education of the survivor as well as his or her health status. To examine this relationship, we estimated two different models, one taking education level at the time of diagnosis, which is unlikely to depend on being diagnosed with cancer, and the other model using education level at the end of the follow-up period.

We constructed four different models with the same sets of regressors. The first model (pooled linear regression (OLS)) included age, sex, age at diagnosis, treatment type, relapse, cancer site, urban-rural classification and health authority. The second model (OLS with robust standard errors) used the same regressors and calculated robust standard errors that control for heteroscedasticity.

The third and fourth models are random effects models with and without adjustment for cluster-effect. Fixed effects (FE) and random effects (RE) models are statistical analysis used for longitudinal data. The FE model assumes that time-invariant characteristics of the individuals are perfectly collinear with the person [29]. In FE model variables that do not change over time (i.e. sex, cancer site, survivor or not, level of education, urban-rural classification, and health authority) are absorbed by the intercept. The RE model assumes that the error term is not correlated with the predictors which generates an estimate for time-invariant variables. In our study we used the RE model because the key explanatory variable is constant over time (i.e., cancer survivor or not) [29].

In the model we used, a series of multivariate panel data regression models determined the relationship between survival and income. A random effects model was chosen because a fixed effects model does not produce an estimate for constant exogenous variables (i.e., sex, cancer site, survivor or not, level of education, urban-rural classification, and health authority).

The fact that we have a large individual dimension (N) relative to the time dimension (T), the random effects model was considered to be an attractive model since individual effects can be viewed as random.

We also ran the Breusch and Pagan Lagrangian multiplier (B-P/LM) test for random effects. The B-P/LM assesses whether the estimated variance of the residuals are dependent on the values of the independent variables [38]. We also calculated cluster robust standard errors which provide consistent parameter estimates in the presence of heteroscedasticity [29]. We applied the Huber/White/Sandwich VCE estimator as discussed in Arellano and Wooldridge [29, 39].

In all our models, we adjusted for age, sex, cancer type (i.e., bone, central nervous system (CNS), carcinoma, germ cell, leukemia, lymphoma, soft tissue sarcoma, and other), treatment type (i.e., chemotherapy, radiotherapy, and surgery), marital status (i.e., married, widowed, divorced, separated, and single), health authority (i.e., Vancouver Costal, Interior, Fraser, Island, and Northern), and urban-rural classification from Statistics Canada (i.e., Census Metropolitan Areas (CMAs) and larger Census Agglomerations (CAs), large communities, and rural and small towns). The analyses used Stata version 13.1 [40].

This study was funded by the Canadian Cancer Society Research Institute to utilize available data from the CAYACS program. Ethics approval for the study was obtained from the BCCA/University of British Columbia Research Ethics Board; and approvals to access, utilize and link the data were obtained from BCCA, BC Ministries of Education and Advanced Education, and Statistics Canada.

The demographic and clinical characteristics of our survivor cohort are described in Table 1. The initial cohort had 4390 cancer survivor records. Records with incomplete birth date information (n = 352) were discarded because they could not be linked to data in income data files, and 11 were discarded because they were linked to only one-year of tax information (Fig. 1). Of the linked cases, 2024 (51%) were males. This can be explained partly by the fact that women may be considered a loss to follow-up because they may marry into a new family name. Of the unmatched cases, 144 (41%) were males. We excluded 83 survivor records whose age was below 15 years at the time of their first tax return. Our cohort included 106,777 person-years. Table 2 compares general characteristics of the cohort and the control group. Both have similar distribution in terms of sex, being married, health authority, and rural-urban classification.

We achieved a linkage rate of 92%. The majority of the survivors (66%) were diagnosed from age 15 to 24 years, which is 70,557 person-years. We followed survivors for 5 to 29 years with an average of 16 years. Mean age (SD) at diagnosis was 14.4 (3.7) years. Mean age (SD) at diagnosis was lower for leukemia survivors (9.8 (6.8) years) compared to all other sites (greater than 12 years). Twenty-two percent of survivors had a diagnosis of lymphoma and 20% of survivors were treated with surgery only. We generated binary categories for treatment types and 10% received only chemotherapy and 4% radiotherapy only. Most survivors attained K-12 educational level (76%). Mean log eqivalized income was lowest for CNS 10.3 (1.0) and highest for the germ cell and lymphoma 10.6 (1.0) Table 3.

This population-based, long follow-up study showed that childhood, adolescent, young adult cancer survivors earned significantly less than the general population of BC. Survivors of CNS tumors had a significantly lower income than lymphoma survivors. The study’s findings demonstrate that demographic and cancer characteristics significantly predict survivors’ income. As similar to previous studies utilizing self-reported income and out-of-pocket expenses [24], lower incomes were reported among cancer survivors when compared to the general population.

Our study demonstrated both sex and education levels of the cancer survivors impacted on their income. Specifically, a lower income was reported for older female cancer survivors; this finding has been reported in other studies [20, 41, 42]. This adds further support to the hypothesis that the effect of health on income is an important determinant which is more pronounced for older females. Older females remain among the most economically vulnerable groups. It is of a major concern whether older female Canadians will outlive their financial resources.

In addition, cancer survivors with higher education levels earned significantly more than those with less educational qualification. This is similar to what previous studies have found in cancer survivors [43] and in other chronic diseases like diabetes [44]. Survivors who received radiation therapy have significantly lower income compared to those received a mix of two or more treatments. Radiation therapy that directly impact nervous system may result in sensory, motor, and cognitive damage that may have adverse consequences on educational attainment and future employment. In this study a lower income was reported for CNS survivors; this is in line with results of previous studies [22, 24, 45, 46]. Neurocognitive impairment and negative emotional and psychosocial late effects, poor educational outcomes, and higher odds of unemployment after treatment for CNS tumors is common in survivors [5, 22, 24, 47‐54]. The lower income observed in this study is likely to be attributed to the multiple, severe long-term health problems survivors of CNS generally experience.

The strength of our study is the long follow-up of survivors and the abstraction of treatment data from medical records. Self-reported income is used in most previous studies, which is prone to bias. This study utilized tax-reported income, which eliminates the bias of self-report.

A potential limitation associated with the use of income from Statistics Canada is the lack of information regarding job-type and hours worked; however, differences in full- vs. part-time employment would be reflected in insurance benefits that are included in annual income. Future studies should examine employment status as an outcome of interest and examine its association with income of survivors overtime.

We have matched the case and the control by sex and birth date. Urban rural classification and marital status were obtained after our data is linked to income data from Statistics Canada. We did not have prior knowledge of these variables. The distribution of cases versus control by urban rural classification is different. Survivors seem to be more likely to live in small areas and the proportion of ‘single’ in the control group is lower than the survivor group. It is important in future studies to include marital status and urban-rural classification in matching survivors to the control group.

Another limitation of the study is absence of information related to the level of education and ethnicity of subjects. We assessed the impact of education for a sub-sample of cancer survivors but not for the control group that was created from T1FF. A further potential limitation with the data from Statistics Canada, or any administrative data source, is a lack of control over data selection, quality, and collection. The prevalence of random or systematic data entry errors can be difficult to estimate; however, systematic, differential errors are unlikely, and random errors should be evenly distributed between the study groups. Additionally, subject attrition from datasets due to out-of-country migration can be problematic. However, follow-up by using the T1FF will include data from across Canada; national emigration in the relevant age groups has been low.

Finally future studies on income of cancer survivors should look at the dynamic relationship between income in adolescent versus young adult, and adult age. Dynamic panel models or trajectory analysis could provide some light into this association.

Using unique linkage of the cancer cohort with the Longitudinal Administrative Data from the Statistics Canada, we estimate long-term economic impact of cancer on labor market outcomes of cancer survivors by comparing their outcomes to those of the comparison group consisting of people from the general population. Young cancer survivors earned significantly less than the general population. In addition, survivors of central nervous system tumors have significantly lower incomes than lymphoma survivors. Survivors who received radiation therapy have significantly lower income..

In light of these findings, it is vital that the Ministry of Health, cancer care institutions, labor force organizations, disability programs, and industry develop policies to address the economic status of survivors of childhood, adolescent, and young adult cancers. Particular attention should be given to those previously treated for CNS tumors and to those who received radiation therapy Economic inequality can only be associated with hardship for survivors. The development of programs that provide life-long support and rehabilitation is critical.