| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the Finnish Institute of Occupational Health, Turku, Finland (J.V., A.L., J.P.); Finnish Institute of Occupational Health, Department of Psychology, Helsinki, Finland (M.K., A.V.); University of Helsinki, Department of Psychology, Finland (M.K.); University of Turku, Department of Biostatistics, Turku, Finland (H.H.); Research and Development Centre for Health and Welfare, Helsinki, Finland (M.E.).
Address correspondence and reprint requests to Jussi Vahtera, Finnish Institute of Occupational Health, Hämeenkatu 10, FIN-20500 Turku, Finland. E-mail: jussi.vahtera{at}ttl.fi
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Method: We examined the association between a serious event (death or severe illness of spouse, death or severe illness of other family member, death of close relative) and health among 6,095 male and 21,217 female public sector employees in Finland by repeated measures Poisson regression analysis with the generalized estimating equations method. The longitudinal data comprised self-reports of 3,556 events and their timing in 2000 or 2001, monthly sickness absences between 1997 and 2003, and psychiatric morbidity and suboptimal health 0 to 3 months, 4 to 6 months, or 7 to 12 months after the event. Adjustments were made for age, education, and marital status.
Results: Exposure to stressful events was associated with a greater increase in sickness absence and a longer recovery period among women than among men. For the women, death or illness in the family was also associated with self-reported health problems irrespective of the time lag between the event and the measurement of health, whereas for the men, this association was found only in the first months after the event.
Conclusion: Our findings suggest that women are more vulnerable than men in the aftermath of a death or illness in their extended family.
Key Words: sex • life event • sickness absence • health • vulnerability • recovery
Abbreviations: CI = confidence interval; GEE = generalized estimating equations.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Several factors may contribute to such mixed findings. First, sex differences may arise from a number of factors in the pathway from exposure to endpoint, such as the nature and frequency/occurrence of stress exposure, variation in coping capacity and biological response, and nature or type of health outcomes (14,15). Women have been found to be at a greater risk of depression and other affective disorders after stressful events that occur to important others in their social network than men (11,16–18). However, the direction of the sex difference may be totally different to that when myocardial infarction is an outcome (10,19). Unfortunately, previous literature on the health effects of life events has predominantly addressed such outcomes as emotional and psychiatric problems. Events affecting family members may also increase the risk of health problems, especially among women, because women have a greater responsibility in family-related roles than men do (20). Thus, being embedded in supportive social networks can be harmful for women due to the "cost of caring" (17) when negative events affect a loved one, whereas for men such costs could be smaller.
Second, the occurrence of stressful life events is, in part, dependent on the person. Individual differences that develop early may increase one's probability of ending up in an environment that is likely to produce stressful life events (21). In such cases, sex differences in the person-related factors that influence exposure to stressful events may confound comparisons. Sex-related personality differences can also affect the reporting of events and health problems because enduring personality traits predispose people to interpret different contexts and situations consistently. For example, trait anxiety could increase reports of stressful life events (particularly for those events that are not clearly defined) distress, self-rated nonoptimal health, and absenteeism (22).
Third, reversed causality is a potential source of spurious comparisons between men and women. Most life-event research is retrospective, and it is thus unable to reliably determine the time order between the event and the health risk. In the few prospective studies available, the assessment of preevent health has seldom been sensitive to minor health problems, which, however, may be sex-specific and predictive of subsequent, more serious health outcomes. Furthermore, studies with short follow-ups without repeated measurements of outcome are open to bias if periods of latency or recovery differ between men and women.
Our aim was to study sex differences in health after stressful life events in the family. In addition to postevent subjective health, we used sickness absence to measure changes in health during an extended period covering the time before and after a life event to eliminate many of the aforementioned methodologic problems. Routinely collected absence records cover a full range of different illnesses, including minor health problems, and they are useful in detecting changes in health on a monthly basis. The event was indicated by death or severe illness in the extended family, which is likely to occur for reasons that are random with respect to the exposed person and the health outcome. Focusing on this event minimizes the possibility of confounding by characteristics of the person.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Giving consent did not depend on the occurrence of an event (p = .21). The sample did not differ from the eligible population (i.e., all 12,453 of the male and 31,334 of the female full-time permanent or fixed-term public sector employees who had worked for over 11 months in the year of the survey) in terms of age or type of work contract (mean age 45 years and proportion of permanent employees 84% to 85% in both groups). The proportion of men (22%), manual employees (16%), and the mean days of sickness absence in 1997 to 2003 (13.3 per person-year) was slightly lower in the sample than in the eligible population (28% men, 23% manual employees, 15.2 days of sickness absence). The ethics committee of the Finnish Institute of Occupational Health approved the study.
Life Events
From a list of 16 life events (7), we selected the following categories: (1) death or severe illness of spouse (two items), (2) death or severe illness of another family member (two items), and (3) death of a close relative (one item). For each event, the participants were asked whether the event had occurred during the current year (yes/no) and, if yes, to give the date (month) of the occurrence.
Sickness Absence
We used the participants' personal identification numbers (a unique number assigned to each Finnish citizen) to link the data to the electronic records kept on work contracts and sickness absence by the employers (23). For the respondents who reported a stressful life event, we calculated the number of sick days during each month for the 36 months before the event to 30 months after the event. For those without stressful life events, we randomly selected a nonevent month and linked their sickness absence records to the data in the same manner as for those with an event.
To explore changes in sickness absence in relation to a stressful life event, we determined the mean level of monthly absence figures for the following six time periods: 13 to 36 months and 1 to 12 months before the event; the month the event occurred; and 1 to 6 months, 7 to 18 months, and 19 to 30 months after the event. The mean absence level in each time period was calculated from sick days per month of employment (i.e., days when the respondent was absent from work for reasons other than sickness were excluded from the denominator).
Other Indicators of Health
Postevent psychiatric morbidity was assessed using the 12-item version of the General Health Questionnaire, cases being those scoring 
4 (24). Postevent suboptimal health status was indicated by health ratings less than good on a 5-point scale (25). To explore postevent health in relation to a stressful life event, we determined the time lag between the occurrence of the event and the return of the questionnaire, classified into the time periods of 0 to 3 months (41% of all events among men and 39% among women), 4 to 6 months (28% among men and 30% among women), or 7 to 12 months after the event (31% among both sexes).
Background Variables
The demographic factors included in the analysis were sex, age, education (high school or not), and marital status (married or cohabiting; other). Trait anxiety was measured by the State-Trait Anxiety Inventory (26).
Statistical Analysis
The associations of sex with the background variables and stressful life events were analyzed with the 
2 test. Because sickness absence is a rare event and constitutes count data, the distribution of this discrete variable was modeled with a Poisson distribution in the analysis. For taking into account both the overdispersion of sickness absence between persons and the correlation between sickness absence within persons, we applied a repeated-measures Poisson regression analysis with generalized estimating equations (GEE) estimation (27).
To explore changes in sickness absence in relation to a stressful life event, we calculated the adjusted means of the monthly sickness absence figures and their 95% confidence intervals (95% CI) for six time periods (13–36 months and 1–12 months before the event; the month the event occurred; and 1–6 months, 7–18 months, and 19–30 months after the event) from models including the interaction term "time x event." We compared these means between and within the exposed and unexposed men and women, as well as the significance for the event and the time x event interaction.
We also studied the associations between life events and sickness absence, taking into account postevent psychiatric morbidity. Instead of an event, we used a combination variable with four values (no event, no morbidity; event, no morbidity; no event, morbidity; and event, morbidity) in the analyses.
Finally, we studied the associations between postevent psychiatric morbidity and suboptimal health, with particular attention to the time lag between the occurrence of the event and the return of the questionnaire. For each event, we formulated a variable with four values (no event [reference], health measured 0–3 months after the event, health measured 4–6 months after the event, health measured 7–12 months after the event). We used logistic regression models and expressed the results as odds ratios (95% CI). Sex differences in the associations between an event and the health outcome were tested using the time lag x sex cross-product terms. In these analyses, restricted to exposed persons only, the variable for the time lag was treated as a continuous variable.
All the analyses were adjusted for age, education, and marital status. We used the SAS 8.2 program package for the analyses.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
|
Altogether, 1,698,900 sickness absence days and 132,613 person-years at work were recorded for the participants. Figure 1 illustrates the mean levels (95% CI) of sickness absence adjusted for age, education, and marital status among the employees exposed to different life events and those with no such exposure at the following five points in time: 2 to 3 years before the event (i.e., at baseline), the month the event occurred, the first 6 months after the event, and the following two 12-month periods. The level of sickness absence at baseline between the exposed and unexposed male and female participants was the same in relation to all the events. The only exception was that for the women who experienced death or severe illness of a family member other than their spouse. They had a 1.25 times (95% CI, 1.13–1.39) higher absence rate than other women already 2 to 3 years before the event.
|
In relation to all the event categories, there were sex differences in changes in sickness absence in the month the event occurred. In absolute terms, the mean levels of monthly sickness absences among the exposed women, compared with those of the unexposed women, were 0.94 to 1.20 days higher; for the exposed men, this difference was only 0.28 to 0.49 days. Compared with the baseline level of sickness absence, the rate of absence in the month the event occurred was 2.34-fold (95% CI, 1.76–3.11) for the women and 1.62-fold (95% CI, 0.86–3.15) for the men after the illness or death of a spouse. The corresponding ratios for the women and men were 1.74 (95% CI, 1.47–2.06) and 1.84 (95% CI, 0.93–3.66) after the illness or death of another family member, 2.04 (95% CI, 1.79–2.33) and 1.26 (95% CI, 0.85–1.86) after the death of a close relative, and 1.93 (95% CI, 1.74–2.14) and 1.34 (95% CI, 0.98–1.84) after any of the aforementioned events, respectively.
Figure 1 also shows that, in relation to death or severe illness of a spouse, the absence rate of the exposed women was 1.28 times higher (95% CI, 1.01–1.62) than that of the unexposed women, even 19 to 30 months after the event, whereas for the exposed and unexposed men, the absence rates at that time were at the same level. A corresponding pattern, although less obvious, was found for recovery from the death or illness of another family member. For recovery from any event, no sex differences were found. For the women, there was a significant interaction between all the events and time. For the men, no such associations were observed. These tests statistically support the observed sex difference in sensitivity to stressful life events.
A further adjustment for trait anxiety, in addition to demographic characteristics, did not materially change the findings. In the month the event occurred, the mean level of sickness absence among the exposed women compared with that of the unexposed women was 1.01 days higher for death or illness of a spouse, 0.85 days higher for death or illness of another family member, and 0.95 days higher for death of a close relative. The corresponding differences among the men were only 0.27 to 0.28 days. The interaction between the event and time remained significant for the women (p = .001 after death or illness of a spouse, p = .003 after death or illness of another family member, and p < .001 after death of a close relative) but was nonsignificant for all of the event categories among the men (p > .20).
Table 2 presents the associations between the events and postevent psychiatric morbidity and suboptimal health 0 to 3 months, 4 to 6 months, or 7 to 12 months after the event. Among the women, a death or illness in the family was statistically significantly associated with health impairment, irrespective of the time lag between the event and the measurement of health. For psychological distress, the strongest effect of this event was found in the first months after the event, followed by a gradually decreasing effect by increasing time lag. For suboptimal health, such an effect attenuation during the 12-month period was largely lacking. For the men, a corresponding pattern was seen for psychological distress after the death or illness of a spouse but not for suboptimal health. Regarding the death or illness of another family member, the odds for poor health were increased only for the first months after the event, the difference being further supported by a significant interaction between sex and a decreasing prevalence of suboptimal health over time. The death of a close relative was associated with health only for the women and for no longer than 6 months.
|
Finally, we studied whether the increased vulnerability to stressful life events among the women could be explained by postevent psychiatric morbidity. The mean levels of monthly sickness absences based on the combination variable with four values (no event, no morbidity; event, no morbidity; no event, morbidity; and event, morbidity) are shown in Table 3 for the women and in Table 4 for the men. The absence level in the month of death or severe illness of the spouse was elevated only among the women with postevent psychiatric morbidity. Their absences were 2.21 (95% CI, 1.63–2.99) times higher than those with psychiatric morbidity but no event. Among the women with no morbidity the absence rate after the same life event was only 1.21-fold (95% CI, 0.68–2.14) the level of the no-event group. For other life events, psychiatric morbidity was not a marker of increased vulnerability to sickness absence. Among the women with no postevent morbidity, the absence level in the month of the death or illness of a family member was 1.69 (95% CI, 1.52–87) times higher and in the month of the death of a close relative, 1.84 (95% CI, 1.55–2.19) times higher than in the no morbidity-no event group; for those with poor postevent psychiatric morbidity, the corresponding absence levels were almost the same as those in the group with psychiatric morbidity but no event (1.60 [95% CI, 1.24–2.08] and 1.83 [95% CI, 1.2–2.3], respectively). These findings suggest that the association between a stressful event and an increase in sickness absence could not be solely explained by postevent mental health.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
We found that women reported the occurrence of a death or illness in the extended family more often than men. However, in relation to the death or illness of a spouse, no sex differences were found. Our data did not permit us to determine whether these differences stem from women having larger networks or being more sensitive to adversities that occur in their networks. Nevertheless, our results provide partial support for the hypothesis that sex differences in the rates of reported stressful events are greater in the distal social network than in the proximal one (17).
We found that exposure to stressful events was followed by a greater increase in the absence figures of women than in those of men. The period of recovery was also longer among the women in relation to the death or severe illness of a spouse and other family members but not to that of close relatives. The main body of these findings was replicated by other established measures of postevent health, namely, minor psychiatric morbidity and the overall rating of suboptimal health. This finding suggests that women are more vulnerable to the effects of stressful events in the family than men are. Our results agree with earlier findings on women's greater risk of depression after stressful events that occur to important others in their social network (11,17). Furthermore, our findings on employees with no psychiatric morbidity after a death or illness in the extended family add to the literature by showing that women's increased vulnerability is not necessarily restricted to depression and other mood disturbances.
Several factors may play a role in women's vulnerability. One possibility is that events affecting family members increase the risk of health problems more among women because women have a greater responsibility in family-related roles than men do (20). The fact that there was a strong association between the death of a close relative and sickness absence among the women and practically no association among the men is in accordance with the explanation for the "cost of caring" (17).
Our findings do not support the argument of Kendler et al. (13) that most life-event categories equally influence the risk of health problems among men and women. Several methodologic differences may account for the apparently conflicting findings. Kendler et al. (13) focused on major depression, a specific outcome that does not account for health deterioration from other illnesses affecting sickness absence figures. In addition, Kendler et al. (13) only examined depressive onsets in the month of event occurrence; the assessment of the timing of the outcome was retrospective; repetitive measurements of the outcome of interest were not carried through; and, finally, their sample, composed of Caucasian twins born in Virginia, represents a group from a different cultural area.
A major contribution of our study may be methodologic in nature. We measured single events, such as death or severe illness of a loved one, that are likely to occur for reasons that are random with respect to the behavior of the participants and the outcome. Our prospective study design allowed us to measure postevent changes in health intraindividually over a long period relative to 2 to 3 years before the event, a strategy that effectively controlled for bias resulting from stable third factors, such as differences in reporting style. By using reliable records of routinely collected sickness absence, we were able to eliminate the subjectivity problem, a potential source of bias in epidemiologic studies (23). Public-sector employees are paid full salary during sick leave. Regulations allow an employee to be absent from work to attend the funeral of a family member and to care for acutely ill children under 10 years of age for up to 3 days without loss of salary. Such absences are not recorded as sick leaves.
Earlier research on mental health has often employed a retrospective design. However, depression can cause some events, and people with a history of depression have reported more events than others (28). Moreover, depressed mood in experimental studies has been shown to be associated with reporting style and the liability of illness (29). Thus common causes of event exposure and the outcome in retrospective studies may have biased some earlier results by inflating associations between life events and mental illness. Our findings with less specific health outcomes largely eliminated such potential methodologic problems.
Five limitations are noteworthy. First, we used sickness absence records to measure health. Sickness absence can be considered a measure of health if the concept of health is understood in terms of social, physical, and mental functioning (30). Such independent archival data minimized the possibility of common-method bias, the risk of selective recall bias, and other problems characterizing research with self-reported data. Both for men and women, sickness absence has been found to be a more powerful predictor of all-cause mortality than established self-reported health measures and available objective measures of specific physical illnesses and medical conditions (31), as well as a strong predictor of specific causes of death, such as cardiovascular disease, cancer, alcohol-related causes, and suicide (23), and a risk marker for future disability retirement (32). Observational studies have shown an increase in the risk of health problems after stressful events, including acute infections (2,3), headache (33), mental disorders (4,6), and asthma (5). All of these problems are common causes of sickness absence (34). It is therefore possible that sex differences in the extent to which various illnesses are related to sickness absence could explain our findings. However, some of the sick leaves obviously represent voluntary absenteeism not related to physical or mental illness (35), and some employees work while ill and record no absences (36). Thus, a potential explanation for the sex differences found could also be that women tend to report in sick more often than men in order to provide social support for ill or bereaving family members or close relatives (37). Our recent results suggest, however, that providing support in intimate relationships decreases rather than increases the risk of sickness absence among employed women (38). Although the findings on sickness absence were replicated by minor psychiatric morbidity and the overall rating of suboptimal health, none of these measures are closely associated with more objective health indicators such as clinical diagnosis or pathophysiologic changes. However, comprehensive data on such indicators are rarely obtainable in large-scale epidemiologic studies.
Second, there is a possibility that the specific nature of the type of event considered exacerbated the interpretation of the results of the study in relation to its outcome measure, sickness absence. The field of events that was studied comprises the deaths and serious illnesses of family members. Women traditionally fulfill the carer role, particularly in these circumstances, in either looking after severely ill family members or providing support to bereaved family members. Although an increase in the carer role may produce health problems and thus lead to sick leave, the need to provide additional care that leads to time off work may also become cloaked as sick leave. However, we were able to replicate the findings concerning sickness absence by using the time lag between the event and the measurement of postevent suboptimal health and minor psychiatric morbidity. For this reason, the possibility that, for reasons not related to health, women are more likely than men to be absent from work after a death or an illness of a family member is unlikely.
Third, a key feature of the checklist approach is that all life events of a given type are treated equivalently. Thus, the influence of the context on the stressfulness of the event is ignored. Intensive interviewing techniques have been shown to be much more effective in recalling and accurately dating events and in reducing response errors due to individual differences than checklist measures are (13). Although the strength of the association between the event and the outcome would probably increase with the contextual approach, such a strategy is not applicable in large-scale epidemiologic studies such as ours. By demonstrating a sharp increase in sickness absence in the month the event occurred, our study is the first to demonstrate that checklists measure date events relatively accurately in large epidemiologic studies.
Fourth, we cannot rule out the possibility that our results were influenced by sex differences in the recall and reporting of stressful life events. For example, women may include their own parents and siblings in their family more often than men. If so, men would report the deaths or illnesses of closer family members than women. However, this was not the case in our study. Furthermore, the findings on the heightened vulnerability of women cannot solely be explained by sex differences in the recall and reporting of events, because there were no sex differences in the reporting of the death or illness of a spouse. Sex differences in trait anxiety, a factor potentially affecting the reporting of events, is not a plausible explanation because the mean level of trait anxiety was the same for the men and women, and controlling for trait anxiety did not alter the results on the effects of life events on health.
Fifth, the questionnaire and consent to access sickness records were completed on the same occasion. It is possible that those who chose not to participate had a different pattern of association between events and health than the participants did. The fact that giving consent did not depend on the occurrence of an event and that our sample represented the eligible population well speaks against this possibility as a major bias. Our cohort was 78% female and racially homogeneous (white employees), corresponding to Finnish municipal workers in general (39). However, future research with more diverse samples is needed to evaluate the generalizability of our findings.
In conclusion, we were able to document a variation in the immediate impact and the length of the risk period associated with a death or illness in the extended family among men and women. Our findings suggest that women are more vulnerable than men in the aftermath of death or illness in the family or the death of a close relative. The sex differences in the association between stressful life events and illness are not likely to have resulted from a confounded association.
|
NOTES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
This study was supported by the Academy of Finland (project 105195), the Finnish Work Environment Fund (project 103432), and the participating towns.
DOI:10.1097/01.psy.0000203238.71171.8d
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
