Post-treatment work patterns amongst survivors of lymphoma treated with high-dose chemotherapy with autologous stem-cell transplantation

The data source is a national multicentre cross-sectional follow-up study where all survivors treated with HDT-ASCT for lymphoma in Norway between 1995 and 2008, aged ≥18 years at HDT-ASCT, alive at survey, residing in Norway and currently not undergoing systemic therapy for active malignancy were eligible and invited to participate (n = 355, Fig. 1). The survey, performed in 2012–2013, consisted of a detailed self-reported questionnaire with a set of well-established patient reported outcome measures [3, 6], a comprehensive out-patient clinical examination [7], as well as data retrieved from patients’ charts and the clinical quality register for lymphoma at Oslo University Hospital (OUH). For the present study, we include only survivors treated with BEAM (carmustine, etoposide, cytarabine and melphalan), which has been the standard high-dose regimen in Norway since 1995, excluding six survivors treated with total body irradiation (TBI). In Norway, all citizens are entitled to old-age pension from 67 years of age. Respondents who reported receiving old-age pension (n = 40) or being students (n = 3) at survey are excluded in the present study since the focus here is on work patterns before and after treatment, and these two groups are not considered part of the formal labour force. In general, the survey had a low percentage of missing data.

Survivors are categorized according to primary lymphoma entity: Hodgkin lymphoma (HL), indolent non-Hodgkin lymphoma (NHL) and aggressive NHL, number of regimens prior to HDT-ASCT, and whether they experienced relapse after HDT-ASCT or not [3]. Body Mass Index (BMI) is from the clinical examination (kg/m2). We replaced 36 missing values using statistical measures based on self-reported BMI.

Patients were asked to retrospectively report their employment status when first diagnosed and before HDT-ASCT, as well as their current work situation (at survey), according to eleven categories. We use this information to construct three categories for employment status; fulltime workers (having a fulltime job, being self-employed, or on sick leave), part-time workers (part-time job), or not employed (unemployment insurance, disability insurance, temporary disability insurance, or homemaker).

Patients rated their work ability on a scale of 1 to 10 (where 10 is best), both as they perceived it when answering the survey, and how they remembered it before onset of illness (i.e. when first diagnosed). They were also asked to report the number of weekly work hours at survey and when diagnosed. Their current employment status, current work hours and current work ability are the main outcomes in our analyses.

Based on their current work situation (at the time of survey), we construct a three-part categorical variable for employment; full-time workers, part-time workers and not employed (Table 1), and a binary variable for being employed at follow-up (not distinguishing between full and part time) or not.

For these questions, we have used a modified version of the Work Ability Index (WAI) [8], and questions developed by our group in previous studies [3, 9].

We use the three questions that pin their employment status to three points in time (when diagnosed, before HDT-ASCT, and at survey) to create the tree shown in Fig. 1 to describe withdrawal from work life. We postulate that once you move from a blue box (inclusion) to a red box (exclusion) it is harder to return to a blue one. We refer to this as “withdrawal” from work life, conditioned on inclusion at the time of diagnosis, thus only defined for the left-hand side of the tree. Non-withdrawals are the respondents on the left branch of the left-hand side of the tree (three positives, n = 125).

We exploit the variation in time from when a patient received HDT-ASCT, until he or she completed the questionnaire. We construct three intervals: 3–7 years, 8–12 years and 13 years or more, of relatively equal size. We thus simulate panel data, creating ‘pseudo panels’ for current work hours and work ability, supplemented with work hours and work ability from when they were first diagnosed.

The term ‘pseudo panel’ refers to the fact that we do not have the opportunity to follow the same patients over time. Instead, we group respondents according to how long it has been from the time they received treatment (HDT-ASCT) until they completed the questionnaire.

We calculate ratios for work hours as work hours in the withdrawal group divided by work hours amongst those who stayed employed from diagnosis to survey, repeating the calculation for work ability.

We estimate an economic loss from the difference in work hours between the withdrawal group and the non-withdrawals using wage statistics from Statistics Norway (https://​www.​ssb.​no/​en/​statbank/​table/​08057). We estimate an expected mean wage for the withdrawal group using monthly earnings (NOK) for 2012, covering all employees, matching them by gender and level of education.

We define marriage as being in a paired relationship, and higher education as more than 12 years of education. We construct two binary variables for somatic illnesses; one for having had one or more of three heart diseases (myocardial infarction, angina pectoris or heart failure), and another for second cancers (new cancer diagnosis, other than lymphoma). Chronic fatigue is assessed according to the Fatigue Questionnaire [10], containing 11 items concerning physical (7 items) and mental (4 items) fatigue during the last month. Two additional items cover duration and extent of fatigue. Responses are dichotomised (0 and 1 scored as 0, and 2 and 3 scored as 1), with CF defined as sum score of ≥4 of the dichotomised responses with duration of ≥6 months. Anxiety is derived from the Hospital Anxiety and Depression Scale (HADS), consisting of an anxiety and a depression subscale with seven items each [11]. Each item is scored from 0 (not present) to 3 (highly present), and anxiety caseness is defined as a sum score of ≥8 on the anxiety subscale.

We run a multinomial logistic regression, allowing for comparison between more than two groups (Table 2). The dependent variable was the categorical three-part variable divided into not working (base outcome), part time (work) and full time (work) at the time of survey. We include six covariates, relaxing the rule of thumb of 10 events per variable in logistic regression [12]. Employment before treatment is our main variable of interest. Gender and age at survey are necessary individual characteristics. We choose second cancers, chronic fatigue, and anxiety based on their significance in the regression models (Table 2) and their clinical relevance. Other covariates were tested, but they had limited explanatory power, and were not included in the proceeding analyses.

We run five regression models with the binary variable for employment (i.e. employed or not) at follow-up as the dependent variable, conditioned on being employed at diagnosis (Table 3). In model 1, employment before treatment is the sole covariate. We expand the model stepwise, adding new covariates in each step. In model 2, we include the sociodemographic variables gender, relationship status, education and age at survey. In model 3, we add the somatic health variables BMI, heart disease, second cancer and relapse of lymphoma. Next, we add the mental health variables chronic fatigue and anxiety (model 4). In model 5, we add the number of treatment lines before HDT-ASCT and lymphoma type.

We use a standard t-test to test whether the difference in means between withdrawals’ and non-withdrawals’ work hours and work ability at more than 13 years is different from the difference in means at onset. We restrict the test to the patients observed at diagnosis and 13 years or more after treatment. We test the null hypothesis, that there is no difference in means.

We exclude respondents with missing observations from the analyses when the relevant variable enters the equation. Despite a somewhat higher incidence of missing values for work hours and work ability, we choose not to use statistical measures to replace them, considering the risk of manipulating the results too great. We do not have additional information (as we did for BMI), which could be used to estimate missing values, and would therefore have to rely on imputations based on other respondents’ values.

The South-East Regional Committee for Medical and Health Research Ethics (REC South East) approved the study, and all participants gave written informed consent.

There are no differences between the participants and non-participants with regard to age at diagnosis, HDT-ASCT or survey, nor gender, observation time or lymphoma entity.

In total, 274 survivors completed the questionnaire (77% of eligible survivors), 49 were excluded, leaving 225 respondents for the final analyses (Fig. 1).

In total, 39% of the participants are female, 74% are in a paired relationship, and 47% have higher education. Their mean age was 40, 43 and 52 years at diagnosis, HDT-ASCT, and survey, respectively (Table 1).

Labour market characteristics show that 85% of the participants were employed when diagnosed. They worked on average 34 h per week. On a scale of 1 to 10, their work ability averaged 8.5. At the time of the survey, median 10.8 years later, 69% were employed, they worked 21 h per week, and their work ability was 6.1 (Table 1). The 99% confidence intervals show no overlap for the respective means.

Results from the multinomial logistic regression model show that being employed before HDT-ASCT increases the probability of employment at survey, part or full time, compared to not being employed (base outcome) (Table 2). The other covariates; female gender, higher age, diagnosed with second cancers, chronic fatigue and anxiety; reduce the probability of working full time at survey (coefficients < 1), while anxiety reduces the probability of working part time.

The main result from the five regression models (Table 3) is that employment before HDT-ASCT is significantly and positively correlated with employment at survey throughout all five models. The constant in model 1 predicts a 33% probability of being employed at survey if not employed before HDT-ASCT, whereas being employed before HDT-ASCT adds 44.8 percentage points to this probability. In models two to five, employment before HDT-ASCT adds 44.5, 43.6, 38.4, and 37.3 percentage points, respectively.

Three covariates; second cancers, anxiety, and being in a paired relationship; are associated with a lower probability of employment at follow-up. Results from the most extensive model (5) suggest that getting a second cancer diagnosis reduces the probability of employment at survey by 47.4 percentage points. Anxiety is associated with a reduction in this probability of 19.2 percentage points, while being in a paired relationship is associated with a reduction of 14.3 percentage points. The remaining health variables have no significant effect on the probability of future employment.

In Fig. 2 (the top two panels), we show trajectories for the pseudo panels for work hours and work ability for non-withdrawals and withdrawals; at diagnosis, their average reported weekly work hours were 38.2 and 34.5, respectively, similar to a regular workweek (37.5 h). Non-withdrawals rated their work ability at 9.2, while withdrawals rated theirs at 7.6, on average. After HDT-ASCT, average weekly work hours in the withdrawal group drop to 7.1 amongst those who answered the survey 3–7 years later. In the group who responded 8–12 years later, it drops to 1.1, while it is 0 for those responding ≥13 years after HDT-ASCT. The drop in work hours for non-withdrawals is smaller, and stabilises at 30.5. Withdrawals’ reported work ability at survey is 2.9 for those who respond 3–7 years after HDT-ASCT, and 3 for those who respond 8–12 years later. However, for those who respond more than 12 years later, it is 5.1. Work ability for non-withdrawals drops steadily to 7.5, 7.1 and 7. The trajectory for work ability in the withdrawal group suggests a rebound effect over time.

We amplify the differences in trajectories for withdrawals and non-withdrawals in the last panel in Fig. 2. At diagnosis, work hours and work ability in the withdrawal group were around 90 and 80% of non-withdrawals’, respectively, but, while withdrawals’ work hours drop to zero, work ability rebounds to a level similar to where it started 13 years or more after treatment (Fig. 2). The difference in average weekly work hours between withdrawals and non-withdrawals is − 5.4 h at diagnosis, compared to − 29.4 h 13 years or more after treatment (Pr(|T| > |t|) = 0.0001). However, there is no statistically significant difference in average work ability (2.6 points lower in the withdrawal group at diagnosis, and 4 points lower after 13 years (Pr(|T| > |t|) = 0.265)).

The withdrawal group works on average 26.6 h less per week than non-withdrawals at survey. Using wage statistics for 2012 (year of survey start) from Statistics Norway, by gender and level of education, we estimate an expected mean hourly wage of NOK 265.40. This gives us a yearly loss per person of NOK 331,555 (EUR 34,791), equivalent to 70% of the average yearly wage in Norway in 2012 of NOK 470,900 (Statistics Norway).

Compared to other European countries, Norway has a relatively high employment rate. On the other hand, a relatively high share of the working-age population receives health-related benefits [17]. Work force participation depends on individual characteristics, such as education and work ability, but also on labour market regulations and the income insurance system.

Strengths of this study include the completeness and representativeness of the study population, with all lymphoma survivors after HDT-ASCT in Norway being accounted for, together with the high response rate of 77%. Furthermore, participants and non-participants are highly comparable, strengthening the generalisability of our results.

Our study is limited by the cross-sectional design preventing any conclusion regarding causality to be made. All work-related data were collected by questionnaire at one time point, with a risk for recall bias, and not controlled by interviews and/or data from The Norwegian Work and Welfare Administration.