Participants and data sources

A prospective population-based cohort study was conducted. The study population was defined as all female twins, i.e. women with a twin sister, born between 1959 and 1990 in Sweden, and recorded in the Swedish Twin Registry (STR) [22]. The STR is the largest population-based register of twin births in the world, with information such as birth date, sex, zygosity, and pair identification [23], [24]. After excluding women who delivered their first child before 1994 (or women whose twin sister had her first delivery before 1994) or before the age of 16, and twins where none in the pair had their first delivery between 1994 and 2009 (n = 7 304), the final cohort comprised 5 118 women. The selection of the study population is illustrated in Figure 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow chart for the study population from the Swedish Twin Register.

https://doi.org/10.1371/journal.pone.0101566.g001

Swedish nationwide registers make it possible to perform linkage of data between different registers on an individual level. The unique personal identity number assigned to each Swedish resident [25] was used to link information from several population-based registers. The Causes of Death Register was used to obtain information on death date [26]. The Medical Birth Register, established in 1973, includes information on almost all births in Sweden [27] and was used to obtain information on deliveries from 1994 until 2010. In order to increase coverage on deliveries, we also used the National Patient Register (NPR) to obtain information on delivery. This register includes all individuals admitted to any psychiatric or general hospital since 1987 [28]. Information on hospitalization with a principal diagnosis for delivery (as defined by the International Classification of Disease (ICD) [29]: O80-84) was obtained. The NPR was further used to obtain information on other hospitalizations.

From Statistics Sweden, we obtained information on educational level and year of emigration [30]. Finally, information on DP (date and diagnoses) for the years 1994 to 2009 was obtained from the Social Insurance Agency. DP diagnoses were based on the 9^th and 10^th revisions of the ICD.

Time in relation to childbirth

The studied women were followed in relation to year of first child birth, irrespective of birth outcome, referred to as T₀. In order to compare those who gave birth to those who did not, T₀ for the women who did not give birth was defined as the year of the twin sister's first childbirth.

Disability pension and social insurance in Sweden

All residents in Sweden aged 16–64 years are covered by the public sickness insurance and can be granted disability pension (DP), if their work capacity is reduced permanently or for long time due to disease or injury. The DP benefit covers at least 65% of lost income. In Sweden, the customary age of old-age retirement is at 65 years.

Variables

Hospitalization in terms of inpatient care was studied during the period six years prior to, and six years after T₀. In a descriptive part, we summarized the total number of hospitalization days per year. In the regression analyses, exposure to hospitalization after T₀ was considered. In order to consider subsequent pregnancies, we excluded future hospitalizations related to pregnancy, childbirth, and the puerperium (ICD-10: O80-84).

In the descriptive part of the study, total number of DP days each year during the period six years prior to, and six years after T₀ was summarized. In the analyses on cause-specific DP, diagnoses were classified according to the ICD-9 and ICD-10 at chapter levels for the two diagnostic groups mental disorders (ICD-9: 290–319; ICD-10: F00–F99) and musculoskeletal disorders (ICD-9: 710–739; ICD-10: M00–M99). Also, all participants were followed from year 2–5 after T₀ with respect to being granted DP.

Co-twin design

Depending on the zygosity, twins in a pair may share either 100% (monozygotic, MZ) or on average 50% (dizygotic, DZ) of their common genetic make-up. In addition to a similarity depending on the common genes, twins may also become more alike owing to the environment they experience together, i.e. shared (mainly childhood) environment. Co-twin design enables the possibility to take both genetic and shared environmental (also called familial) factors into account by investigating twin pairs discordant for factors of interest. In the discordant twin pair analyses, the follow-up time of the twin mother (on DP) in relation to the follow-up time of the co-twin without DP will be analyzed. Stratification by twin pairs allows each twin pair to have their own baseline hazard to control for the effects of potentially confounding familial factors in the models. That would control for familial factors, meaning that the observed associations found between child birth and hospitalization or DP would not be explained by familial factors shared by the twin sisters in a pair. If the familial factors would play a role, then the association should be present only in the analyses of the whole cohort but not between discordant twin sisters.

Statistical analysis

Descriptive analyses, based on mean number of days per year in hospital or on DP, were calculated. Cox proportional hazard models with constant time-at-risk [31] were applied to estimate hazard ratios (HR) with 95% confidence intervals (CI) for being granted DP during year 2–5 after T₀. In the regression analyses, we excluded individuals who were granted DP before T₀ and their twin (n = 156). The different sub populations in the regression analyses are described in Figure 1. When analyzing the association between childbirth, hospitalization after T₀ and DP, based on 4 962 twins (referred to as Population 1), three regression models were fitted, where we in Model a only adjusted for birth year. In Model b we made additional adjustments for delivery year and educational level at T₀. Educational level was classified into four categories; up through 9 years of compulsory school, 10–12 years of education (equivalent to senior high school), ≥13 years of education (i.e. college or university education), and one additional group for missing. In Model c finally, we also adjusted for previous hospitalization, i.e. inpatient care prior to T₀. Potential familial confounding was controlled for by analyzing twin pairs that were discordant with respect to being granted DP during the follow-up. Conditional Cox proportional models were applied and the sample was stratified by twin pairs [32]. Influence of familial factors (genetic and shared environment) will be indicated if the association found in the analyses of the whole cohort will disappear or change considerably in the analyses of discordant twin pairs. These co-twin analyses included 1 378 complete twin pairs (referred to as Population 3). In order to account for the sampled twin pairs rather than independent individuals, all analyses were clustered on twin-pair identity [33].

Similarly, we estimated HR with 95% CI for the associations between hospitalizations prior to T₀, after T₀, and DP, respectively. In these analyses, above described the regression models a and b were used.

All statistical analyses were performed with SAS 9.3 and STATA 12.1.

Interpretation and comparison with other studies

To our knowledge, this is the first study to examine the long-term associations between childbirth, morbidity and DP. One recent study analyzing the influence of low social interaction and socioeconomic risk conditions on the risk of DP in a cohort of all Swedish women born in 1960 to 1979 and not on DP, found that women not living with children and especially women living alone, had higher risk for DP when compared to women with children [7]. Similar findings were reported in another Swedish study of the same cohort and the DP risks were still lower among those having three or more children [34]. In some of the analyses, women with children had higher risk of future DP [34]. Worth noting, however, is that these two studies focused on motherhood rather than on giving birth and that also women not born in Sweden were included.

Our results also revealed that women who did not give birth had higher risk for DP due to mental diagnoses; this has not been studied before. An Icelandic study on DP due to mental disorders pointed out single mothers with poor financial status to be of greater risk of mental disorder that can lead to DP [35]. There are also findings suggesting that childlessness is not linked to mental disorder [36], [37]. However, these two studies did not use DP diagnoses but rather self-reported information on disorders [36], [37]. It is of course important to be aware of that most women, in both groups, were not granted DP at all.

In general, women who did not give birth had a higher risk for future DP. One could expect that a health selection, where some of the women who do not give birth may have worse health and therefore cannot get pregnant or choose not to have a child. However, several women choose not to have children due to other reasons. Studies comparing death rates in pregnant women with death rates in other women have showed a health selection possibly combined with an apparent protective effect of pregnancy on women's health [38]–[40]. Women who have given birth may thus have a greater proportion of healthy women.

The analyses of risk for DP due to musculoskeletal diagnoses showed the opposite, i.e. that women who gave birth had a higher risk for such DP. This is in line with previous research, showing that childbirth may be associated with higher risk for future musculoskeletal disorders [10], [11].

The risk of future DP was especially elevated in women who had experienced a hospitalization the years following the first childbirth or equivalent. Hospitalization, that is, our measure of morbidity, remained a significant predictor for being granted DP even when previous hospitalization, i.e. inpatient care prior to T₀, was accounted for. As it is known that low socioeconomic position is a risk factor for being granted DP [7], [41], we adjusted for educational level. However, this adjustment barely affected the estimates, leading to the conclusion that morbidity in terms of hospitalization is a strong risk factor for future DP. This might seem self-evident, nevertheless, often questioned in public discussions of the ill-health content of DPs. Previous research on the association between hospitalization and subsequent DP is scarce, and even though studies have shown that morbidity is associated with DP, most studies have focused on self-rated health and self-reported conditions rather than hospitalization as an indicator of morbidity [7], [34], [42], [43]. Also, repeated hospitalization, which could be regarded as a more severe measure of morbidity, was associated with an elevated risk for future DP.

Familial factors seemed to influence the association between childbirth and DP, between hospitalization and DP for those who gave birth, and also between repeated hospitalizations and DP. One possible explanation could be that more severe morbidity implies more severe disorders, and pathology generally shows higher heritability estimated than symptoms. For example, heritability of mental disorders varies between 28–85% [17], with higher estimates for major depression and bipolar disorder vs depressed mood [44], whereas genetic effects explained approximately 30–60% [13], [14] of the variation in musculoskeletal disorders.

Strengths and limitations

The strengths of this study include the population-based prospective design, using nationwide registers with high completeness and no loss to follow up, and the large cohort. The register-based data counteracts recall or selection bias. The National Patient Register has close to complete coverage of all inpatient care [28]. Other strengths include the use of a co-twin design, which enables to take into account the influence of familial factors, the prospective design with a long follow-up period, and that also information on the situation before T₀ could be included. That we had access to information about DP diagnoses was another strength.

There are, however, some weaknesses that should be mentioned. First of all, the terms for inpatient care have changed in Sweden during the studied period, and inpatient care has to a large extent been replaced by outpatient treatment, thus the rates of hospitalization have steadily decreased. Hence, in our study, some patients who in the later studied years were treated in outpatient care are classified as unexposed, as there is no record of their outpatient visit. However, the impact of this would be the same for those giving and not giving birth. Another limitation was that there were relatively few twin pairs who were discordant with respect to DP. Precision was relatively low, as indicated by the wide CIs. Hence, the results of the discordant twin pair analyses for DP should be interpreted with caution. Moreover, it is important to remember that the results concern women who gave birth, irrespective if the child survived or not or if the woman lived with the child. Other studies include information on being a mother, and might also include women who live with children she did not give birth to.

Ethical approval

The study was approved by the Regional Ethical Review Board of Karolinska Institutet, Stockholm, Sweden (2007/524-31).