Women of reproductive age comprise a substantial proportion of the total workforce in Europe.1 Pregnant women at work are considered to be vulnerable to the quality of their work environment as regards both their own function and health as well as pregnancy outcome.2^–4 However, workplaces are not commonly adapted to the needs of pregnant women3 4 and this is reflected in the high frequencies of pregnancy-related sick leave reported in several studies.5^–12

Absence risks involve factors at the individual level (eg, health, physiological condition, personality, occupation), the group level (eg, work conditions and work organisation) and the macro level (eg, legislation and macro-economics).13 14 A persistent finding in studies concerning pregnant workers has been the variation in sick leave according to work conditions, in particular physical work load, work schedules and organisational factors.7^–12 Accordingly, article 5 of the current European Union legislation15 requires employers to assess health and safety risks to pregnant workers, and where necessary to adjust working conditions and/or working hours temporarily.

Only few studies have addressed job adjustment in pregnancy. The need for adjustment and its effect on sick leave were studied in a national sample of 2713 employees who gave birth in 1989 in Norway.11 The majority considered that there was a need for adjustment, and achieving adjustment was associated with reduced sick leave.11 Sick leave was more common in a French national sample of women who had been employed during pregnancy in 1981 if requests for job adjustment had been refused.10 To our knowledge no intervention studies have been published.

More than 80% of Norwegian women aged 25–39 years belonged to the work force in 2005.16 Work conditions and job adjustment are therefore important elements in the Norwegian Mother and Child Cohort Study (MoBa).17 The objective of the present study was to examine absence from work in mid-pregnancy among employees in MoBa. The aim was to examine absence from work as well as return to work and their determinants, in particular the effects of job adjustment and other job characteristics and work environment factors. Another objective was to study the role of personality factors on the relationship between work environment, job adjustment and work attendance.

METHODS

Design

This study is based on the Norwegian Mother and Child Cohort Study (MoBa) conducted by the Norwegian Institute of Public Health. In brief, MoBa started in 1999, with the aim of including 100 000 pregnant women by 2008. Most pregnant women in Norway are invited to participate, and the response rate so far is around 44%. Pregnant women are recruited to the study through a postal invitation in connection with a routine ultrasound examination offered to all pregnant women in Norway at 17–18 weeks of gestation (www.fhi.no/tema/morogbarn).

The current study is based on participants recruited between 1999 and 2005. Informed consent was obtained from each participant before the study, and the Regional Committee for Medical Research and the Norwegian Data Inspectorate approved the study.

Study participants received three self-administered questionnaires during pregnancy. The cohort was linked to the Medical Birth Registry of Norway (MBRN).18 Further details can be found in the study protocol.19

Study population

A total of 64 136 pregnancies in 1999–2005 were included.17 This constitutes 42.7% of the 150 309 women who were invited to participate.17 The present study included participants who were employed and answered key data in questionnaire 1 (filled out around week 17) and questionnaire 3 (filled out around week 30 in pregnancy). A total of 50 849 women completed the week 17 questionnaire. We excluded 22.5% who were not employed in week 17, the largest groups being students/apprentices (6.4%), housewives (6.2%) and the self-employed (3.7%). Among the remaining 39 429 who were employed in week 17, we excluded 8.3% (3267) who did not answer questionnaire 3 and 19.2% (7551) who lacked key information (job adjustment, work environment and absence from work). This left 28 611 study participants, comprising 72.6% of all those employed in week 17.

Study variables

Data were based on the participants’ self-report in questionnaires 1 and 3 (available in appendices 7 and 9 in the protocol),19 except for those who had multiple births whose data were accessed from the MBRN. Questions have been translated into English in the protocol,19 but the questionnaires were only available in Norwegian.

Study outcomes

A number of questions concerned work attendance and absence in pregnancy. A week 17 question: “are you absent from your work at the present time?” allowed the population to be separated into the 22 932 participants who were not absent (group 1) and the 5679 participants who were absent (group 2).

Detailed questions on the number and duration of sick leave spells in specified periods in pregnancy were used to compute the total number of sick leave days in weeks 13–16, weeks 17–20, weeks 21–24 and weeks 25–28 in pregnancy.

There were two main outcomes in the study. The outcome in group 1 was subsequent absence from work for more than 2 weeks, which was based on answering “yes” to the week 30 question: “have you been absent from your normal work for more than 2 weeks after week 13?”. The outcome for group 2 was attendance at work in week 30, based on answering “no” to the question: “are you absent from your work at the present time?”. The categorisations included all reasons for absence, but supplemental questions in weeks 17 and 30 indicate that approximately 90% of absences were sick leaves. We had no information that allowed separation between diagnoses and work-related sick leaves.

Independent variables

We collected data on job characteristics and work environment, which were considered the main study determinants. Furthermore, we considered some individual characteristics and background factors.

Job adjustment classification was based on the following questions in the week 30 questionnaire: “have your working conditions changed since you became pregnant, making it easier for you?” and: “if no, why have your working conditions remained unchanged?”, with the options “not necessary” and four alternatives. Responses were collapsed into three categories: “not needed”, “needed and obtained” and “needed, not obtained”.

Data on work environment were collected from the week 17 questionnaire. There were a large number of questions, so we chose three questions on organisational work environment (decision latitude, pace, monotony) and two questions on physical load (heaviness, turning and bending) that were most strongly associated with absence between weeks 17 and 30. These were based on the statements: “I am able to decide how my work is to be carried out”, “I have a hectic work pace”, “my work is very monotonous” and “I have physically heavy work” with four ordered response options. In addition, the question “do you have to turn and bend many times in the course of an hour?” had four ordered response categories.

Other job characteristics in week 17 were also recorded. These were employment category (public sector, private sector), weekly working hours (four categories) and work schedule (day only, evening only, shift work, night only, other).

We considered individual characteristics. These were health complaints as back pain and posterior pelvic pain in week 17, registered as yes/no dichotomies. We also included two personality scales: Rosenberg’s four-item self-esteem scale20 (week 17) and a five-item version of generalised self-efficacy21 (week 30). All items have four ordered response options. The questions can be accessed in the study protocol, appendix 7 (question 134) and appendix 9 (question 113), respectively.19 We constructed self-esteem and self-efficacy indices allocating each item values from 1 to 4. Thus, minimum values for self-esteem and self-efficacy were 4 and 5, respectively, and corresponding maximum values were 16 and 20.

Background factors considered as potential confounders are included in table 1. The distribution of those variables in the two study groups suggests that group 2 consisted of women of a somewhat lower socioeconomic position and included fewer nulliparous women and more multiple pregnancies. Examination of the 10 818 women who were employed but did not participate showed only small differences from the 28 611 study participants except for a much larger proportion of missing values and a somewhat higher proportion of pregnant women with higher university education (data not shown).

Analysis

Analyses were performed with Stata/SE 9.2 software. We computed the (approximate) 3-month risk of absence for more than 2 weeks in group 1 categories, and fractions of participants attending work in week 30 in group 2 categories (those who had been absent in week 17). The main study determinants were job adjustment, work environment factors and other job characteristics. Job adjustment was analysed in three categories with “adjustment needed, not obtained” as reference. Work environment factors were analysed in four categories with the no problem category as reference for group 1 and the worst problem category as reference for group 2. Job characteristics were included in categories, with public sector, 30.1–37.5 weekly working hours, and the “day only” schedule as references. Associations between determinants and outcomes were estimated as additive differences: absence risk difference for group 1 and differences in work attendance probabilities for group 2. Crude and adjusted point estimates of associations and corresponding 95% confidence intervals (CIs) were computed in binomial regression (the risk difference option in Stata’s BINREG procedure). Models including the potential confounders listed in table 1 lead to lack of convergence. However, the inclusion of those variables had only minimal influence on ratio associations between the main study determinants and study outcomes. Therefore, we decided to use adjusted models that included all job characteristics and work environment variables but not the background factors listed in table 1.

In order to investigate differences in job adjustment effects across subgroups, we analysed associations between job adjustment and absence risk (group 1) and return to work (group 2) in subsets according to educational level, work environment, job characteristics, self-reported health complaints and level of self-esteem and self-efficacy.

The pregnancy was the unit of observation, and some women participated with more than one pregnancy (27 758 women, 28 611 pregnancies). Including mother’s identity as a cluster variable in the analysis solved this problem by taking within-mother dependence on estimate variances into account.

Not only did we wish to estimate the strength of associations between job characteristics and work environment, and the study outcomes, but also the impact of those relationships on a population scale. We did this by computing population attributable fractions (PAFs). The PAF is a function of the strength of an association and the population prevalence of a risk (preventive) factor and can be interpreted as the proportional contribution to the population risk (prevention) attributable to the factor, given the whole population have the same probability of the study outcome as the reference category.22 Crude and adjusted PAF estimates (expressed in percentages) and 95% CI were computed in Stata’s AFLOGIT procedure, after including the factors we wanted to estimate as dummy variables in a Poisson regression model.

RESULTS

The 28 611 participants reported a total of 535 862 sick leave days between weeks 13 and 28 in pregnancy, which constituted 14.9% of all person-days for this period. Half the participants (50.7%) reported one or more days of sick leave, while 36.2% had more than 14 days of absence.

Figure 1 shows variations in sick leave proportions over pregnancy weeks. There was an increase from 17.0% (weeks 13–16) to 44.6% (weeks 25–28). Sick leave was dependent on job adjustment category, being markedly less frequent when adjustment was not considered necessary and highest when adjustment was needed but not obtained.

The total risk of more than 2 weeks of absence between weeks 17 and 30 was 0.308 (7072/22 932) in group 1. The probability of being present at work in week 30 among those who had been absent in week 17 (group 2) was 0.137 (780/5679). Table 1 shows the distribution of the outcomes for each group across categories of background factors. Throughout, absence in group 1 and return to work in group 2 were inversely distributed. In general, young age, low educational attainment, high parity, multiple pregnancies and tobacco smoking were associated with absence, whereas marital status or the couple’s native language had little influence.

Forty per cent of the study participants considered there was no need for job adjustment (table 2). The differences between study groups were distinct, however, and only one in five in group 2 saw no need for adjustment. The necessity for job adjustment was also strongly dependent on educational level, hectic pace at work and physical load at work. Approximately half of those who considered job adjustment to be necessary had obtained it. There seemed to be a tendency for those who needed adjustment most to be somewhat less likely to obtain it.

Absence risk in group 1 and associations with job characteristics and work environment are provided in table 3. Absence was dependent on job adjustment category, with a risk of 0.501 when adjustment was considered to be necessary but not obtained and only 0.151 when adjustment was considered unnecessary. Job adjustment (reference: “job adjustment needed, but not obtained”) was associated with a reduced absence risk of more than 10% (0.107) in the adjusted model. Associations between job adjustment obtained and absence risk in population subsets according to job characteristics and work environment, individual characteristics or background factors did not show much heterogeneity with some exceptions. The strongest job adjustment associations were found for employees with evening, night or shift work (absence risk reduction 0.176, 95% CI 0.147 to 0.204). A strong job adjustment effect was also found for 30 weekly hours of work or less (absence risk reduction 0.148, 95% CI 0.113 to 0.183). Job adjustment was more strongly associated with reduced absence among participants with low levels of work pace compared to high-level categories (data not shown).

There were distinct crude differences with higher absence risks in the public sector, among part-time employees, employees with schedules other than daytime work, and organisational and physical problems at work in a dose-dependent fashion (table 3). These associations became more moderate after adjustment, and risk differences between public and private sectors, work schedules except night work, and turning and bending at work were close to zero. The sums of risk differences for the highest exposure categories were 0.218 for the three organisational work environment factors and 0.101 for the two physical load factors in the adjusted model.

The adjusted population fraction estimates of absence for more than 2 weeks attributed to “job adjustment obtained” and “job adjustment not necessary” were −9.0% and −27.9%, respectively (table 4). The PAF estimates of other job characteristics and work environment factors were rather small, except for hectic work pace and monotony at work. Organisational factors had a much higher impact on absence than physical factors.

The probabilities of returning to work in week 30 in group 2 in association with job characteristics and work environment are shown in table 5. The pattern is the opposite of that for absence in group 1. A moderately higher probability of returning to work in the public sector was an exception to this pattern. Job adjustment was associated with a probability increase of 0.041 of returning to work in the adjusted model. No clear heterogeneity in job adjustment effects was found in separate strata of job characteristics and work environment, individual characteristics or background factors (data not shown). The trongest associations for work environment factors were seen for lack of hectic work pace, monotony and physical burden at work in the adjusted models. The sums of differences in fractions returning to work for the most favourable exposure categories were 0.146 for the organisational work environment factors and 0.058 for the physical load factors in the adjusted model.

The adjusted population fraction of returning to work in week 30 attributed to “job adjustment obtained” was 12.3%, whereas the corresponding estimate attributed to “job adjustment not necessary” was 33.5% (table 6). Most of the job characteristics and work environment factors had moderate impact on return to work, but the PAFs for both organisational and physical factors combined were substantial. The highest PAF estimate was found for heavy physical work. Several of the PAF estimates had wide confidence intervals.

Low self-esteem was weakly associated with increased absence in group 1, but no pattern was evident with return to work in group 2. Self-efficacy score was not associated with either outcome. Further, self-esteem and self-efficacy score had no measurable influence on the relationship between job characteristics and work environment, and study outcome in either group (data not shown).

DISCUSSION

We found a substantial level of absence from work in mid-pregnancy among employees in the Norwegian MoBa cohort. More than half of the participants had experienced spells of sick leave between weeks 13 and 30, and sick leave accounted for nearly 15% of all days in this period. Absence was strongly dependent on the pregnant woman’s own perception of the need for job adjustment. Among those who needed job adjustment, absence levels were lower when this was obtained. The association between job adjustment and absence was particularly strong among employees working shifts, evenings or nights, or doing part-time work. Absence was also associated with the quality of organisational and physical factors in the work environment. These associations were strongest and had the largest impact for organisational qualities, notably hectic work pace and monotony at work. However, heavy physical work had a considerable impact on the probability of returning to work. Personality factors (self-esteem, self-efficacy) did not seem to influence work attendance or the associations between work attendance, and job characteristics or work environment.

The MoBa study has several positive features, being large, prospective and population based. However, there are potential validity problems as well, and in particular two problems that could cause bias. First, the low participation could cause selection bias.17 Second, this particular study uses self-reported data on all determinants and outcomes, which could lead to information bias.

Participation in the study was low. The 27.4% employed in week 17 who were excluded did not seem to differ much from the participants. However, the 57.3% loss of participation in the MoBa cohort17 could be more important. It is clear that the participants’ educational attainment is differently distributed from that of Norwegian women of comparable age, with a higher proportion of women with third level education.23 MoBa participants are also slightly older, with fewer low birth weight children and fewer lone mothers than all mothers in the MBRN.17 It is therefore likely that key determinants (job adjustment, work environment, other job characteristics) have skewed distributions toward more favourable categories, and that absence from work is underestimated. This could suggest that the overall absence risk (0.308) is falsely low and that the probability of returning to work (0.137) is falsely high. This is a likely assumption considering the outcome distributions across educational categories (table 1). It is also supported by the fact that the fraction of days with absence during follow-up in the study population (14.9%) is lower than the national second trimester absence (about 20%).24 It is more difficult to assess the effect of such selective loss on associations. There are some studies showing clear effects by selective participation on descriptive characteristics (prevalences, risks) but only minor effects on associations.25 As yet, no such analyses have been performed in the MoBa cohort. Subset analyses among participants with different levels of education showed job adjustment effects that are not very heterogeneous. This could indicate that bias from selective loss of lowly educated mothers would not have a large impact on the job adjustment associations.

All information on key variables was based on self-report from the pregnant woman in our study, and this is clearly not the most appropriate design for our purpose. This could lead to common method bias26 because error in independent and dependent variables could correlate. Such bias would most likely lead to inflated estimates of associations. Simultaneous error in determinant and outcome data that has its origin in personality characteristics (eg, exaggeration in the description of adverse phenomena by persons with high negative affectivity) will cause a biased inflation of associations.27 We did not have data on negative affectivity level, but the total lack of influence by self-esteem and self-efficacy level could suggest that personality factors did not have substantial influence on the results. Others have found good agreement between self-reported and registered sickness absence, and also similar associations with self-reported health for both measures.28 29

The study is longitudinal, and information on determinants was provided before the outcome information. This was not the case for information on job adjustment which was provided in retrospect in week 30. Particular caution is therefore warranted concerning the job adjustment results which could be influenced by work attendance earlier in pregnancy. Job adjustment responses being influenced by degree of work attendance rather than the quality of the work environment and working hours could lead to a false job adjustment effect.

It is difficult to rule out confounding due to lack of comparability on factors not accounted for in an observational study. However, the socioeconomic and other factors we did consider had no measurable influence on the results when included in the analytical models. The increased absence and reduced probability of return to work in women with part-time work (⩽22.5 h) could possibly be explained by residual confounding from background factors or work environment, since this group had a particularly negative distribution on several factors associated with increased risk.

In general, the work environment results are in agreement with several other studies in the Nordic countries and elsewhere.5^–12 The finding of clear associations between organisational and physical qualities in the work environment, and the risk of absence and probability of returning to work when absent is also in agreement with several other studies.7^–12

Six out of ten pregnant employees considered there was a need for job adjustment, and about half of these obtained some re-arrangement of their work. These figures are quite similar to those of a Norwegian study in 1989.11 It seems that those in greatest need according to their own judgement were not the ones most likely to achieve such adjustment in the present study. This is in accordance with results from a national sample of births in France in 1981.10 Our results suggest that when job adjustment is needed, achieving some arrangement decreases the risk of an absence of more than 2 weeks by nearly 11%. Job adjustment increased the probability of attending work in week 30, when absent in week 17, by 4%. The perceived need for job adjustment among pregnant employees is frequent in this study, and the population fractions of reduced absence and increased attendance that are attributed to obtained job adjustment in this study are considerable. Job adjustment in pregnancy was also associated with reduced sick leave in France10 and Norway,11 but comparison with our result is difficult due to differences in study design and outcome categories. In the Norwegian study, the proportion of women who left work on sick leave before delivery was 11.3% lower in the “adjustment obtained” category than in the “adjustment needed, but not obtained” category.11

The results suggest that the high level of sick leave in pregnancy can be influenced by the quality of the work environment and by efforts to make job adjustments for pregnant women. However, the results should be interpreted cautiously because of low participation in MoBa and potential information bias from self-reported data. The job adjustment results are in accordance with the few published studies that have addressed this topic and support the role and importance of the European Union legislation15 on job adjustment in pregnancy. Despite the fact that the European Union directive15 has been active since 1992, research and evaluative efforts have been scarce. We based our results on self-reported data on work and work attendance in mid-pregnancy. Studies that provide more detailed information on the types of adjustment are warranted. We believe that more knowledge on the effects of job adjustment in pregnancy could be obtained from studies based on data from other sources (eg, company level data). Intervention studies in this field could be particularly interesting.