The impact of cancer on spouses' labor earnings†
A Population-Based Study
Presented at the Fourth Biennial Cancer Survivorship Research Conference entitled “Cancer Survivorship Research: Mapping the New Challenges,” Atlanta, Georgia, June 18-20, 2008.
Cancer affects patients' incomes, but to the authors' knowledge few studies to date have examined how the income of the patients' spouses may be influenced. In this population-based study from Norway, the effects of cancer on both partners' earnings are analyzed.
The difference between labor earnings the year before the cancer diagnosis and that 2, 5, or 8 years later was compared with the difference in earnings over a corresponding period for similar persons without cancer, applying linear regression models to national registry data. Approximately 1.1 million married persons ages 35 to 59 years were included, among them 17,250 persons diagnosed with cancer during 1991 through 1999.
Two and 5 years after a cancer diagnosis, married men experienced lower earnings than they would have absent the illness. Cancer in wives, however, did not affect men's earnings. Women's earnings were adversely influenced to the same extent by their own as by their spouses' cancer. Brain, lung, and colorectal cancer in male spouses produced the most adverse effects on women's earnings. All effects were most pronounced for women no longer married.
Women's earnings are lower after both their own and their spouses' cancer illness, and divorced and widowed women experience the most pronounced reduction after spousal cancer. Men's earnings are lower only if they are diagnosed themselves. This may reflect traditional sex roles, with men as main breadwinners and women as caregivers. For family households, cancer in men may result in greater financial difficulties than cancer among women, although the effect will depend on breadwinner roles before diagnosis. Cancer 2009;115(18 suppl):4350–61. © 2009 American Cancer Society.
A growing population of people living with a history of cancer has resulted in more attention being paid to the long-term health and well-being of cancer survivors and their families.1 The effect of cancer on productivity and work ability is of particular concern. An altered affiliation to working life among cancer survivors may threaten the economic well-being of entire families, and negatively affect people's identity, life satisfaction, and social relationships.1-4 Economic problems will also be aggravated if the spouse finds it difficult to maintain the earlier work activity.5
People's annual labor earnings are determined by the number of hours worked and the wage, the latter reflecting several factors, including productivity. Important determinants of people's working hours and productivity include age, health status, educational level, parental status, and caregiving responsibilities at home.6 Labor earnings may thus be affected by whether people encounter serious illnesses such as cancer, and perhaps also suffer from adverse treatment related side effects.6-9 In addition, healthy spouses may need to spend time on caregiving tasks and thus reduce their own working hours or take on less challenging and less economically rewarding jobs. However, it is also possible that spouses work hard to earn higher incomes, to compensate for losses in partners' earnings. This may depend on couples' organization of their households before disease onset, as well as men's and women's roles in society more generally.10 Cancer in either spouse could thus be hypothesized to affect both patients' and partner's earnings, and thus have a marked impact on the household's overall economic situation.
To our knowledge, only a few studies to date have addressed cancer's impact on the earnings of married couples, and the focus has always been on breast cancer and thus on women. These studies suggest that breast cancer does affect earnings.2, 11-15 The overall effects on households have, however, been reported to vary considerably.11-15 Norwegian women earn on average 30% to 40% less than men, primarily because they work part-time and hold lower level positions, which pay less.16 This is also the case in the other Nordic countries, Canada, the United States, and the United Kingdom.10, 17 Knowledge of the impact on household earnings of other cancer forms, and perhaps in particular among men, is thus warranted.
In this study, we explore how the earnings of married women and men in Norway are affected by their own or their partner's cancer experience. We were particularly interested in investigating whether there are sex differences in the response to own and spouse's cancer even in the supposedly sex-egalitarian Nordic setting. The effects of cancer are also likely to be time dependent. There may be marked changes in earnings during acute illness phases, followed by longer periods during which the influence of illness and treatment may be more modest. At the same time, couples may handle pronounced declines in earnings over a short time period, but be adversely affected by smaller, but persistently lasting effects. This study assesses both short- and long-term effects by analyzing earnings 2, 5, and 8 years after diagnosis. In addition, several types of cancer are considered, defined by site and stage at diagnosis, which may have different effects on subsequent earnings. The potentially modifying effect of the presence of children, educational level, and earnings before illness onset are also addressed, as these factors may influence both the level of earnings that are needed to avoid economic hardship for households and couples' flexibility in organizing their private and working lives after cancer in a family member.
MATERIALS AND METHODS
Data from 4 sources were linked by means of the personal identification number assigned to everyone who has lived in Norway after 1960. The Norwegian Directorate of Taxes provided information on yearly gross labor earnings for all citizens. These include sickness benefits, which are limited to 1 year, but not compensations beyond this, such as disability pensions or rehabilitation benefits. Information regarding dates of birth, dates of death, immigration, and emigration, dates of changes in marital status, identification numbers of spouses, and dates of birth of children was retrieved from the Norwegian Population Register. Educational levels were extracted from Statistics Norway's population censuses of 1960, 1970, 1980, 1990, and 2001. Cancer data came from the Cancer Registry of Norway, which has registered all cancer cases nationwide since 1953. Mandatory reports from clinicians and pathologists, as well as death certificates, ensure completeness and data of high quality on cancer form, basis for the diagnosis, and stage at time of diagnosis.18 Unfortunately, to our knowledge, no reliable registry data on treatment received, recurrences, or other adverse events are currently available.
All Norwegian men and women married at the initiation of follow-up (1990-1998) and alive and in the group ages 35 to 59 years at the end of follow-up (1999-2001) were included, for a total of 17 250 persons with cancer and 1.05 million persons without cancer. Labor earnings from 1990 through 2001 and all cancers diagnosed between 1991 and 1999 were considered. A detailed description of the study design and the number of observations included is shown in Table 1.
Table 1. A Visualization of the Design Employed
|1991||1992||2000||8|| || |
|1992||1993||2001|| || || |
| || || || || || |
|1994||1995||2000||5|| || |
|1995||1996||2001|| || || |
| || || || || || |
|1997||1998||2000||2|| || |
|1998||1999||2001|| || || |
Cancer may affect people's subsequent earnings for several reasons, including limitations in physical and psychological health and changes in values and priorities.2, 8 In addition, there are several factors that may influence cancer incidence, survival, and earnings.19 In a statistical model for earnings, these should ideally be included as control variables, along with an indicator of the cancer illness. Some of these potential confounders are easily measurable and available in the Norwegian registry data. This is, for instance, the case for age and educational level. The latter is positively linked to income, the chance of surviving cancer, and the incidence of breast and prostate cancer and some other malignancies, whereas it is negatively linked to the incidence of, for example, cervical and lung cancer.6, 19, 20 However, other potential confounders are unobserved, and to account for those of them that are time invariant (ie, constant individual-level characteristics potentially affecting earnings, cancer incidence, and survival), a difference-in-differences approach was used.
As an example, consider a married person with no history of cancer in 1995 and alive at the end of 2001. Using a difference-in-difference approach means that one estimates a model for the difference in his earnings between 1995 and 2001 and includes an indicator of whether he had cancer in 1996, later in the period, or never. One might also include the age difference, but it is likely to be unrelated to the chance of having cancer. The difference in education is most likely both unrelated to cancer and very small for the age range considered. Including age and education in 1995 may be more important, because they may affect the subsequent change in income as well as the cancer risk. Even with registry data, however, there are many people at relatively low ages relevant in such an analysis who develop cancer. We therefore constructed a larger data set that also included married men with no history of cancer in 1994 who were still alive in 2000, as well as married men with no history of cancer in 1993 who were still alive in 1999. The dependent variable was the difference in gross labor earnings between 2000 and 1994 or between 1999 and 1993, respectively. Similarly, the disease variable was whether the person had cancer in 1995 or later, or in 1994 or later. The labor earnings were not inflation adjusted, as fairly short time periods during which the inflation has been quite low were considered. For instance, $1000 in 1991 is equivalent to $1255 in 2001, whereas $1000 in 1994 is equivalent to $1183 in 2001, and $1000 in 1997 is equivalent to $1111 in 2001.21 Adjustments for inflation led to a generally smaller increase in earnings over the 9-year period considered (as evident by a smaller intercept), but left the cancer estimates practically unchanged.
Mathematically, the estimated models were of the type
in which Yi,t,t+6 is the difference in earnings between t (t is 1993-1995) and t+6 for person i as described above, Di,t is a vector of disease indicators further explained below, Xi,t is a vector of control variables, also defined below, and ϵi,t is an error term. α0, α1 and α2 are the corresponding coefficients. The variables are specified as categorical.
Similar analyses were also performed with observation windows of 2 years (differences in earnings between 1998 and 2001, between 1997 and 2000, or between 1996 and 1998, ie, Yi,t,t+3) and 8 years (differences in earnings between 1992 and 2001, between 1991 and 2000, or between 1990 and 1999, ie, Yi,t,t+9) (Table 1).
It is not quite reasonable to consider the observations for the 3 consecutive periods (for instance, t = 1993, 1994, 1995) as independent, because 1 person can contribute 1, 2, or 3 such periods. In line with common practice in multilevel modeling, one might consider adding a person-specific error term, which is known to increases the standard errors of the effects of the person-level variables but leave the point estimates unchanged.22 A few such models were estimated in PROC MIXED (SAS Institute, Inc., Cary, NC), but as virtually similar estimates of standard errors were obtained, the chosen approach was considered adequate.
For each of the 2-, 5-, and 8-year observation windows, 4 different sets of models were estimated. One was for men and included a D variable with the following 2 categories: the human was diagnosed with cancer in the relevant year (eg, 2, 5, or 8 years before 1999, 2000, or 2001) or never. A second set of models also included the men exclusively, but the D variable had these 2 categories: spouse was diagnosed with cancer the relevant year or never. These models were estimated on 3 different samples, 1 that included only the men who were still married; 1 that included those who had divorced, separated, or become widowers; and 1 that included all men irrespective of relationship status at the endpoint. Altogether, approximately 85% of the married men and women remained married throughout the study periods. The third and fourth sets of models were similar, except that they were estimated for women, with their earnings differences as the dependent variable. Cross-tabulations of spouses' cancer diagnoses demonstrated that there was hardly any concordance of cancer between spouses in the age groups considered here, in line with the findings of others.23 Limiting the analyses to only include healthy spouses when considering the effect of one's own cancer or vice versa left the cancer estimates unchanged, and such limitations were therefore not imposed. The least squares linear regression procedure PROC REG in the SAS statistical software package (version 9.1; SAS Institute, Inc., Cary, NC) was used for the estimations. The statistical significance level was set at 5%.
In some models, we introduced distinctions between the most common cancer forms in the age groups considered here, 8 for men and 9 for women. These were colorectal cancer, renal and bladder cancer, skin cancer, brain cancer, lung cancer, cervical cancer, other gynecologic cancers, breast cancer, testicular cancer, prostate cancer, and lymphomas. The stage at diagnosis was also considered in some models. The possibility of conditional effects of presence of children aged <18 years in households was addressed, along with possible effects of having attained a high versus low education, and having high versus low earnings at the start point. One might suspect that the presence of children could also be a confounding variable, because it may affect the likelihood of experiencing an income increase as well as the risk of getting cancer. However, adding number of children as a main effect variable did not change the key estimates.
Descriptive statistics (mean, median, and upper and lower quartiles) of the differences in earnings between the start and endpoint for persons without cancer, with cancer, and with spouses with cancer are shown in Table 2. The median earnings in the respective groups at the start and endpoint are also listed. From this table, it is evident that there are large variations. Some people experience large increases in earnings over the periods considered, whereas others experience a decrease, both among those with cancer and those without. However, on average, men or women with a history of cancer have less positive developments in their earnings. Spouses' cancer history also leaves an imprint on earnings. These differences are confirmed by the model estimates in Tables 3 through 6. Effects of spouses' cancer 5 years earlier are shown in Table 3, along with the effects of the covariates included in all models. Differences in earnings increase with increasing educational levels and decrease with increasing age. Similar patterns were observed for differences at 2 and 8 years (data not shown).
Table 2. Measures of the Distributions of the Earning Differences, by Observation Period and Cancer Status
| Mean (median)||2518 (5670)||6912 (6500)||9749 (8020)||−382 (2000)||2643 (3380)||5731 (5420)|
| Upper quartile||12,700||14,070||16,280||7580||8180||11,360|
| Lower quartile||−3340||0||1500||−5740||−580||0|
| Mean (median)||9833 (12,300)||13,419 (12,630)||17,827 (14,860)||5084 (6700)||6059 (6500)||10,576 (10,500)|
| Upper quartile||22,320||22,860||26,560||14,920||14,340||19,180|
| Lower quartile||0||3380||6380||−1960||0||540|
| Mean (median)||17,372 (16,000)||18,035 (16,800)||23,391 (19,220)||9782 (10,120)||9852 (10,300)||14,528 (14,800)|
| Upper quartile||29,700||29,060||32,840||21,460||20,460||25,100|
| Lower quartile||1180||4440||9400||−480||0||2380|
Table 3. Effects of Age, Educational Level, and Spousal Cancer on Earning Differences Over a 5-Year Period
| No cancer||1,228,039||0||Reference||1,313,094||0||Reference|
| Spouse cancer 5 y earlier||2818||−1856||.28||2675||−2011||<.001|
| Spouse cancer any other time§||42,222||−1678||<.001||40,537||−876||<.001|
|Age group, y|||
| Low or unknown||185,734||0||Reference||231,016||0||Reference|
| Elementary school||346,076||2527||<.001||555,079||2722||<.001|
| High school||327,990||6383||<.001||196,043||4769||<.001|
| ≤2 y of college||54,125||9585||<.001||32,333||6589||<.001|
| Bachelor degree||236,842||13,135||<.001||302,795||7473||<.001|
| ≥Master's degree||122,312||23,623||<.001||39,040||17,546||<.001|
Table 4. Estimates of Effects of Own and Spouse's Cancer on Earning Differences 2, 5, and 8 Years Earlier*
| 2 y earlier (1997-1999)||8123||−5542||1186||.03||−1252||3592||.38|
| 5 y earlier (1994-1996)||16,259||−6265||755||.05||−1856||2818||.28|
| 8 y earlier (1991-1993)||21,266||−4589||571||.18||−2777||2285||.11|
| 2 y earlier (1997-1999)||4302||−5007||2299||<.001||−1369||3820||<.001|
| 5 y earlier (1994-1996)||7531||−3966||1743||<.001||−2011||2675||<.001|
| 8 y earlier (1991-1993)||10,085||−2973||1365||<.001||−1836||2060||<.001|
Table 5. Effects of Spousal Cancer Type and Stage on Earning Differences*
| No cancer||1,254,762||0||Reference||1,228,039||0||Reference||1,192,126||0||Reference|
| Local cancer||1867||−874||.67||1643||−922||.70||1439||−2947||.21|
| Regional cancer||936||−618||.83||653||−3862||.28||498||−1687||.65|
| Metastatic cancer||330||−4720||.33||272||−4852||.38||154||−1267||.85|
| Stage unknown||220||1081||.86||53||10,010||.43||42||2038||.87|
| Blood, lymph,|| brain cancer||239||−3964||.48||197||2171||.74||151||−5607||.41|
| No cancer||1,308,939||0||Reference||1,313,094||0||Reference||1,310,851||0||Reference|
| Local cancer||1418||−924||.03||1249||−1301||.02||1036||−1266||.05|
| Regional cancer||832||−1875||<.001||610||−2918||<.001||434||−758||.42|
| Metastatic cancer||443||−3513||<.001||303||−4655||<.001||225||−3985||<.01|
| Stage unknown||600||240||.71||94||−618||.75||21||−3731||.38|
| Blood, lymph,|| brain cancer||527||−1606||.02||419||−974||.29||344||−2380||.02|
Table 6. Effects of Own and Spouse's Cancer According to Whether They Remained Married Through the Observation Period*
|Men, not married†|
| 2 y (1997-1999)||5618||−20,220||47||<.01||−4007||506||.07|
| 5 y (1994-1996)||13,511||−2406||60||.90||160||751||.98|
| 8 y (1991-1992)||17,519||−5328||60||.33||418||656||.80|
| 2 y (1997-1999)||8340||−4938||1139||.05||−512||3086||.74|
| 5 y (1994-1996)||16,768||−6647||695||.03||−1675||2067||.36|
| 8 y (1991-1992)||22,210||−4540||511||.24||−2933||1629||.17|
|Women, not married|
| 2 y (1997-1999)||5316||−2358||86||.22||−4808||1001||<.001|
| 5 y (1994-1996)||9461||−4184||150||.03||−5380||1138||<.001|
| 8 y (1991-1992)||11,920||−2175||179||.17||−4542||999||<.001|
| 2 y (1997-1999)||4201||−5257||2213||<.001||−510||2819||.08|
| 5 y (1994-1996)||7219||−3952||1593||<.001||−462||1537||.32|
| 8 y (1991-1992)||9641||−3107||1186||<.001||−505||1061||.40|
Overall effects of cancer for all 3 time periods are shown in Table 4, and for comparison we have included the estimated increase in earnings (intercept, α0) for cancer-free men and women in the reference categories, that is, those ages 35 to 39 years and with a low education. Women with cancer, or women with husbands with cancer, have a significantly less positive development in their subsequent earnings over a 2-year, 5-year, and 8-year period than other women. Although the point estimates are lower the longer the observation window, this does not necessarily suggest that there is an early income disadvantage that the women partly compensate for later. Different women are included in the 3 analyses, and those who were included in the 8-year analysis had, for example, survived at least 8 years and may well have had a smaller income disadvantage during the first 2 years than those included in the 2-year analysis, some of whom will have died within the next 6 years. An impact of own cancer is seen also among men, although it is not statistically significant for the 8-year period. Cancer in married men also affects their wives' earnings, thus reducing the household earnings substantially.
Overall, the effects at the household level are greatest at 5 years. From Table 4, it is evident that the household will experience a statistically significant loss in earnings of $8276 ($6265 + $2011) if the husband is affected by cancer, and a loss of $3966 ($3966 + $0) if the wife is affected. The differences are much smaller after 2 years ($6911 for cancer in men and $5007 for cancer in women). After 8 years, however, the household losses are greater for cancer in women compared with men ($2973 vs $1836). For the specific cancer forms considered, declines were particularly pronounced for male brain, prostate, lung, and colorectal cancer. Whereas the effect of brain ($3041, P = .02) and prostate cancer ($3140, P < .01) was observed for the first 5 years only, the effect of lung and colorectal cancer remained fairly stable over the 2-year to 8-year time periods considered here ($3419, $3088, and $3942 [P < .01] vs $2191, $3189, and $1841 [P < .05]). Long-term effects were also noted for a husband's renal, bladder, and skin cancer. Cancer in married women, conversely, did not appear to impact on husband's earnings overall, and no statistically significant estimates could be obtained for any of the specific cancer forms.
Effects of stage at diagnosis for spouses' cancer are shown in Table 5. As persons must be alive at the endpoint to be included in the analyses, stage at diagnosis is a less relevant variable in the analysis of long-term effects of own cancer. Individuals' earnings have, however, been shown to decline with more advanced stage in a previous study using similar data.16 Statistically, husbands' earnings remain similarly unaffected regardless of cancer stage and irrespective of time from diagnosis. Although some point estimates suggest an effect of wives' metastatic cancer, statistical significance was not reached. Wives' earnings, conversely, are strongly influenced by their husbands' cancer stage. A more adverse effect on wives' earnings was noted consistently across all time periods for metastatic cancer.
Results for those who remained married and those who separated, divorced, or lost their spouse are shown in Table 6. Most importantly, among women who are no longer married, spousal cancer was found to have a significantly negative effects on their earnings, in the short as well as longer term. Among these women, there is actually a larger loss associated with husbands' illness (from which they have most likely died) than their own illness. This may in part be what has been picked up by the effect of advanced cancer in Table 5. Women who remain married, conversely, do not have their earnings adversely affected by their husbands' cancer illness. These women experience, however, a particularly sharp effect of their own cancer. This tendency is similar also for married men at 2 and 5 years. Men who become alone see a significant effect only of own illness, and only short-term. A similar pattern was observed for men who remain married.
Approximately 67% of the Norwegian population ages 35 to 59 years are married, and approximately 65% have children aged <18 years.21 The presence of young children could be expected to modify the effect of cancer in both men and women, and in additional models an interaction term between one's own or spouse's cancer; therefore, the presence of children aged <18 years for the time periods considered was included. These interactions were not, however, found to be statistically significant (data not shown).
Employment and earnings may be affected more severely for persons in lower social strata,16, 24 and the effects of cancer for people with high versus those with low earnings at the start point (defined as above or below $30,000/year) were therefore estimated. By and large, there were no differences between the groups (data not shown). This means, however, that earnings after cancer will be relatively lower in the low-income group than among those who are more economically advantaged at the outset, which means that the former group is more likely to experience economic hardship. Whether a threshold level exists is not clear. Much information is ignored by dichotomizing income level, and the inclusion of an interaction term between cancer diagnosis and initial earnings as a continuous variable gave slightly different results for women; the higher their earnings at the outset, the more negative the effect of both own and spouse's cancer (Pinteraction<.01 for all estimates). No significant interaction terms were obtained for similar analyses of the conditional effect of education, and analyses stratified on low versus high educational level yielded comparable negative cancer effect estimates. This was the case for both sexes and for all time periods (data not shown).
To the best of our knowledge, only a few earlier studies have examined the effects of spouses' illness specifically, and only breast cancer has been considered, thereby limiting the findings to male spouses only.2, 13, 14 Compared with cancer-free people, we found that a married woman's earnings are lower after both her own and her spouse's illness, whereas a married man's earnings are lower only if he becomes ill himself. In light of men's generally higher earnings, it is not surprising that larger absolute effects are observed for men than for women. There is, however, more statistical uncertainty regarding the effects for men, in part because of a nearly two-fold prevalence of cancer among women in the age groups considered here.18 Furthermore, we found that household earnings are more negatively affected by the husband's cancer than the wife's cancer, and that adverse effects are present 2, 5, and 8 years postdiagnosis.
Sex Roles and Paid Versus Unpaid Work
Whereas review publications of individuals' employment rates after cancer have shown relatively similar percentage declines for men and women, no summaries of cancer-associated declines in earnings could be identified.25-27 We have previously found similar percentage declines in earnings for Norwegian men and women after cancer,16 in line with those observed after breast cancer.24, 28 Earnings are, however, in general lower among women than men in most developed countries,10, 16 and fairly similar percentage effects may have a different impact on the total economic situation for households.
Our findings may reflect that greater economic household responsibilities for men may make them less likely to reduce hours or take on less demanding positions after their own or their spouse's illness, although studies show that some do reduce their workload after cancer illness in family members.12 To some extent, men may also have less caregiving experience, and thus believe themselves either less competent or less obliged to divert time away from work to care for an ill partner.29, 30 Married women, conversely, are the primary breadwinners in <10% of the households considered in this study.21 They contribute less on average in households' combined earnings, partly because their burden of unpaid tasks in the household, such as for instance caregiving tasks related to childrearing.17 This could be hypothesized to translate into a higher inclination to provide also care for an unhealthy spouse.5, 31
For some, not having to work may be considered a privilege after serious illness and perhaps disability. Women with cancer who have male providers may be able to afford to work less, as their contributions in the household economy may be of less significance, whereas women with cancer who become alone may need to remain in the work force to provide for themselves. The different effect estimates for married versus nonmarried women may reflect this. A US study has shown that married women with breast cancer and health insurance tied to their husbands are more likely to reduce their workload than otherwise similar women with insurance linked to their employer,32 perhaps suggesting that given a choice, some women may prefer to reduce their work load after cancer. For others, the opportunity to work may be important for their wellness, both psychologically and financially.33, 34 Other data and designs are needed to clarify the background for the observed differences in the current study, and international comparisons in this regard are urgently needed but currently lacking.
Women who remain married may have healthier husbands with good prognoses compared with women who lose their husbands to cancer, as suggested by the increasing effect with advanced cancer stage. This may suggest less of a caregiving burden on these women, and is further supported by the finding that the effect for women who do not remain married is most pronounced in the short term. Longer-term effects do remain, but are not as striking. Interestingly, in absolute values, the earnings of women who become alone decline more when their husbands are affected by cancer than when they themselves are affected.
Breast cancer accounted for approximately half of the cancers diagnosed in spouses in this study, and thus impacts strongly on the overall estimates presented. Breast cancer affects women's earnings, but not the earnings of their spouses. Breast cancer has been shown to influence functional status, work capability, and therefore also labor earnings to a lesser extent than some other types of cancer (eg, colorectal, lung, and brain cancer).16 The cancer forms that most clearly affect wives' earnings are colorectal, lung, renal, bladder, and brain cancer. Renal, bladder, and colorectal cancer are relatively common and may be quite debilitating, but at the same time have reasonably good prognoses,18 perhaps resulting in a large care burden over time. Brain and lung cancer have poorer prognosis, and this may in part explain the short-term effect seen for female spouses who have husbands with brain cancer. The persistent effects of lung cancer are harder to explain. It appears, however, that many women remain outside the workforce after cancer in a spouse, and tend to remain on the outside also after they become alone and thus likely have reduced caregiving burdens and presumably could return to work. It remains unclear whether this reflects a choice or difficulties related to work re-entry, but it may nevertheless have negative implications for their social lives as well as economic situations. As women often take on subordinate positions, they may have fewer possibilities for adjustments in daily work schedules, which may be necessary after own or spouse's illness. At the same time, however, part-time work is more common and accepted for women, and this could contribute to an increased flexibility during or after illness in oneself or one's spouse. More knowledge about the situation experienced by these women is, however, clearly needed.
The sex differences observed in this study correspond well with contemporary societies' male breadwinner practices. Although this tendency has weakened over time, it persists to exist for married persons in particular, as demonstrated by the differences between persons who remain married and become alone in this study.
Modifying Effects of Children, Prior Earnings, or Educational Level
The presence of children in households could be expected to increase the care burden and thus limit possibilities of maintaining working positions or hours after cancer in either spouse. In line with the male breadwinner model, however, the presence of children could be expected to impact less on men's earnings when they or their spouses get cancer, as having a larger family to provide for could decrease the likelihood of opting to reduce working hours. Conversely, some studies show that many men (and women) become more family orientated after encountering cancer and thus focus less on work and more on spending “quality time” with family members.35, 36 The effect of children was minor, however, and no sex differences were observed.
Cancer's impact on couples' earnings could be expected to relate to the absolute size of their earnings. On the one hand, a given absolute loss of income will be less important for the welfare of persons living in high-income families than for those living in low-income families. The former may thus be inclined to choose a level of work activity after cancer that leads to a larger income loss. Conversely, wealthy persons may have jobs in which possibilities exist to avoid great income losses even if the hourly work intensity is reduced. We observed sex differences in this respect. It was especially the higher earning wives who experienced pronounced declines in earnings, perhaps indicating that societal norms regarding traditional female caregiver roles still apply in the case of male illness. The educational level has also been shown to modify the effect of cancer on individuals' earnings in previous studies.4, 9, 16, 37 This has not been previously examined for spouses of cancer survivors, but no differences between spouses with high or low education were observed. This may indicate that irrespective of educational level, many spouses undertake major caregiving tasks, as shown also in other studies of the provision of formal and informal care after cancer.12, 14, 38
National, high-quality registry data on the entire married population were used. Selection and information bias is thus minimal, which is an important asset in studying cancer and earnings.26, 39, 40 A disadvantage with the use of registry data is, however, that the causal mechanisms that cancer operates through in affecting earnings, such as wages, working hours, or changing perceptions of the value of work, cannot be identified, and neither can the perceived impact of the lower earnings on the persons' well-being more generally.
The difference-in-differences approach used allowed for control for unobserved constant characteristics that may affect both a person's earnings and his or her chance of developing cancer, or of having a spouse who develops cancer. This is definitely an improvement compared with conventional regression analyses of earnings. However, sources of confounding likely remain, as there may be earnings determinants that change over time and that are linked with the chance of developing cancer. As the chosen approach only allows a control for factors that are time invariant, one should still be cautious in concluding that the estimated effects of cancer on later earnings are causal.
Implications of Results
As public healthcare is available and provided to all Norwegian citizens free of charge, the direct costs associated with becoming ill with cancer, that is, diagnostic workup and treatment, are minimal. The welfare state, common to many Western societies, will to a certain extent compensate for increased expenses as well as illness-related declines in earnings. However, we have seen in this investigation that spouses' incomes are also affected by malignancies (although significant only in the case of men's illness). This is not reflected in public policies or compensatory economic mechanisms, which primarily are directed toward and calculated for the individual with a health problem, and do not extend to the caregiver or family as a whole. The compensatory measures are, unfortunately, not available in our data, and the cancer-associated declines in earnings detected here relate only to changes in labor market activities. The impact on couples' purchasing power overall is most likely smaller than indicated by our results. Furthermore, research is needed to assess total economic effects at the household level both in Norway and in other countries.
The sex equality concept has evolved quite substantially in Norway over the last 3 decades compared with many other countries, and documented household differences in earnings after cancer related to sex are likely to be relevant also for other countries. Norwegian men and women hold to a large degree similar positions and receive equal pay for equal performance in the workplace, but differences on group levels nonetheless continue to exist. As in most other developed countries, men remain the primary breadwinners in particular within married cohorts, with implications for households in which 1 family member is affected by cancer.21
In married couples, women's earnings are reduced after both their own and their spouses' cancer illness, whereas men's earnings are reduced only if they themselves are affected, and only during the first 5 years. For women who become alone, there is actually a larger loss associated with their husbands' cancer than their own illness. Households will thus experience more adverse economic situations when the man is affected by cancer, possibly indicating a significant caregiving burden placed on women in these situations, although the effect will depend on breadwinner roles before illness onset. Because most public and private economic compensatory measures are based on an assumption that men and women have fairly similar roles in society, our findings may be of concern and present practical as well as ideological challenges. Potential implications for children and other household members warrant further study.
Conflict of Interest Disclosures
Cosponsored by the National Cancer Institute's Office of Cancer Survivorship, the Office of Cancer Survivorship of the Centers for Disease Control and Prevention, and the American Cancer Society's Behavioral Research Center.
The study was supported by a grant from the Norwegian Cancer Society.