Methods

The data material used is the Stockholm Birth Cohort Study, created in 2004/2005 by a probability matching of two longitudinal studies: The Stockholm Metropolitan Study (SMS) and The Swedish Work and Mortality Data Base (WMD).21 The Metropolitan cohort comprised all children born in 1953 and living in the greater Stockholm metropolitan area in 1963 (n = 15 117). The SMS data cover the period 1953–1986, while the follow-up data from the WMD cover the period 1981–2003. Approximately 96% (n = 14 294) of the SMS cohort was matched successfully. Ethical permission was obtained from the Stockholm Regional Ethics Committee.

Peer status was measured in sixth grade, when the children were 12–13 years of age (1966). It was established by the question: “Whom in this class do you best like to work with at school?” This question includes dimensions of school performance as well as of friendship, likeability and social inclusion. For example, Stütz finds a high positive correlation between children’s nominations of work partners and their nominations of best friends.22 All pupils were instructed to nominate three classmates in no particular order. School classes with less than 10 pupils were excluded from the analyses. The nominations were categorised into five status groups: marginalised (0 nominations), peripheral (1 nomination), accepted (2–3 nominations), popular (4–6 nominations) and favourite (7 or more nominations).14 This categorisation follows the recommendations that have been made for sociometric information based on three positive nominations, with the mean value as guideline.22 (For a further description, see 13 14.) Because the status distribution does not vary systematically between school classes of different size, it is assumed that the categorisation of peer status reflects a child’s relative standing in the school class irrespective of class size.14

Parental social class was included as a control variable. This information was derived from the 1963 Population Register and, in most cases, refers to the male head of the family.23 The categories were: upper and upper middle class; entrepreneurs; officials and non-agricultural employees; skilled workers; unskilled workers; and others (eg, homemakers, pensioners and students).

Information on overall morbidity (ie, any event leading to inpatient care) was derived from the Swedish Hospital Discharge Register, covering the years 1973–2003. This register contains data about inpatient care, based on all discharges from Swedish hospitals (overnight patients). The diagnoses contained in these records are primarily based on the judgement of the doctor24 and classified using the 8th Revision (1973–1986), the 9th Revision (1987–1996) and the 10th Revision (1997–2003) of the International Classification of Diseases (ICD). In the present study, underlying diagnoses were grouped according to the European Shortlist for Causes of Death as established by Eurostat (see Spijker25 for a further description). The shortlist was chosen not only to facilitate a meaningful grouping of diseases, but also to enable comparisons with similar previous studies (eg, Erikson and Torssander’s work on social class and cause-specific mortality26). However, as it focuses on causes of death, some important morbidity outcomes with low mortality (such as depression) are not identified. The shortlist is largely based on the ICD chapters (although somewhat more compressed) and includes 65 hierarchically arranged groups, of which 17 constitute first-level groups and the rest second-level groups.27 Inpatient care due to complications of pregnancy, childbirth and the puerperium were not included in the present study. Furthermore, inpatient care due to certain conditions originating in the perinatal period, and for congenital malformations and chromosomal abnormalities, were excluded as they precede the establishment of peer status. For the analysis, individuals were first categorised into a dichotomous variable indicating overall inpatient care (ie, the presence or the absence of a hospital discharge between 1973 and 2003, regardless of diagnosis). They were then categorised as discharged or not for all disease-specific groups. Consequently, any given individual may appear in multiple groups. Although this must be kept in mind when interpreting the results, if only the very first event had been included (in line with previous studies analysing multiple years of hospital data24 28), this would, first, have disqualified diseases with a later onset and, second, resulted in several disease groups not being large enough to analyse.

The study population was defined as those with full data for all included variables who attended a school class with at least 10 children, leaving 12 504 individuals eligible for analysis. The procedure of analysing the associations between peer status and cause-specific inpatient care was the following: first, associations between peer status and inpatient care were analysed using Cox’s proportional hazards regression models, producing hazard ratios with the highest peer status group (“favourite”) as the reference category. Subjects entered the study on 1 January 1973 and were censored when dying from any disease, on the date of first discharge (for each disease-specific group separately) or at the end of follow-up (31 December 2003). In both cases, only information on year is available; month of discharge or death is therefore set at the last day of June in the appropriate year. The issue of tied failure times has been handled by controlling the results through the exactm command in Stata, an exact marginal-likelihood method that is appropriate when tied failures are numerous. In a second step, a summary measure of the size of the peer status differences was constructed, using a procedure adopted from Erikson and Torssander.26 Slopes were calculated by means of ordinary least squares (OLS) regressions of the logged parameter values of the five categories of peer status. These slopes were produced for each first-level, as well as for those second-level disease-specific groups that contained at least 30 discharges. The significance of the slopes (5% level) is based on Kendall’s rank order tests of the assumption of a monotonous increase in the hazard ratios from the highest peer status category to the lowest, as opposed to an assumption of no association. Finally, for comparative purposes, additional analyses of peer status and morbidity were conducted. Here, beta coefficients from Cox regressions, with peer status included as a continuous measure (ie, number of nominations, ranging from 0 to 22), were calculated.

Results

The hazard ratios for peer status and overall inpatient care (table 1) demonstrate a clear gradient for men and women alike. This suggests that the lower the peer status, the higher the risk of overall inpatient care in adulthood. These associations remain when childhood social class is adjusted for. The slope is estimated at 0.0935 for men and 0.0928 for women, with a marginal decrease when parental social class is added to the analyses.

Table 2 (men) and table 3 (women) present hazard ratios from Cox regressions for the disease-specific groups, sorted according to slope size. First-level and second-level disease groups are shown separately. Based on the results of the rank order tests, it is shown that linearity can be assumed, with only a few exceptions, for the relationship between peer status and the various disease-specific groups.

Large differences by peer status are found in mental and behavioural disorders, and in the second-level groups alcohol abuse and drug dependence. For example, drug dependence demonstrates a very steep slope, for both men (0.24) and women (0.17). Second-level groups related to external causes of injury and poisoning, such as suicide and intentional self-harm, accidental poisoning and events of undetermined intent, also demonstrate steep gradients. For example, the slope for accidental poisoning is 0.44 for men and 0.23 for women.

Another finding concerns diseases of the circulatory system. Although there is no evident slope for this first-level group, the second-level groups ischaemic heart disease (IHD) and cerebrovascular diseases for men and women, as well as other heart diseases for women, show some of the steepest gradients of all the included disease-specific groups. For example, the slope for IHD is 0.46 for men and 0.29 for women. Slopes related to endocrine, nutritional and metabolic diseases are also very steep. For instance, the slope for diabetes is 0.26 for men and 0.18 for women.

Most of the remaining first-level groups demonstrate slopes in the range 0.10 to 0.20. Some gender differences can be noted: for women, there is no observable slope for neoplasms (0.05) (which is not surprising, as breast cancer is more common among more highly educated women29), while the opposite result is found for men (0.20). This pattern can also be observed for infectious and parasitic diseases and diseases of the genitourinary system. For diseases of the digestive system, however, the slope is fairly steep for women (0.13) but not for men (0.06). The same pattern is found for diseases of the skin and subcutaneous tissue. Finally, diseases of the blood(-forming) organs and immunological disorders demonstrate little evidence of slopes for men and women alike.

Beta coefficients from Cox regressions, with peer status included as a continuous measure, are presented in the last column of table 2 (men) and table 3 (women). Although these coefficients are based on a different scale from the slopes, the results are very similar (with only some exceptions) to the slopes based on the categorised peer status measure.

Discussion

The results suggest that childhood social status in school is associated with overall adult morbidity leading to hospital care: the lower the peer status in school, the higher the risk of inpatient care in adulthood. This gradient exists for women and men alike. When inpatient care is divided into disease-specific groups, however, there are substantial differences in the degree and magnitude to which they vary with peer status, suggesting that peer status position in childhood is associated with most—but not all—types of morbidity. Thus, these findings are, to some extent, in line with theories of “general susceptibility” among individuals in lower social positions, suggesting that there is a common set of general risk factors underlying the association between social position and health.30 However, the variation in associations also emphasises the importance of interpreting the heterogeneous contributions in order to find the mechanisms and pathways through which peer status is linked to disease.11

The gradient for mental and behavioural disorders (including depression, anxiety, alcohol abuse and drug use) and the large relative differences for external causes (such as suicide attempts and intentional self-harm) are consistent with the results from previous studies.31 32 For example, Östberg and Modin found an association between childhood peer status and self-reported health problems in adulthood.13 Despite the fact that the degree of severity most probably differs between diseases influencing self-reported health problems and diseases leading to inpatient care, the results from Östberg and Modin’s study are quite similar to the results found in the present study. The most interesting finding in the present study, however, is the steep gradients for ischaemic heart disease and cerebrovascular diseases for men and women, other heart diseases for women, and diabetes. These types of disease are known to be socially patterned,2 33 and they have previously been linked to childhood socioeconomic status, but not to the child’s status position in school. A plausible mechanism may involve lifestyle factors, health-related behaviours and perhaps stress. To verify this, further research is needed.

The results for peer status and disease-specific morbidity correspond quite well with previous analyses of adult social class and cause-specific mortality,26 which may lead to the conclusion that peer status merely mediates the association between childhood socioeconomic status and morbidity. However, when parental social class is controlled for, the estimates are overall somewhat attenuated, but the main findings remain unaltered. This suggests that the differences in morbidity by peer status are not accounted for by parental social class. Hence, although these types of status structure share important similarities, inasmuch as they are both hierarchical structures of social position, there are also important differences due to the different structural levels on which these status positions are distributed. While the child’s peer status position is created in the classroom (micro-level), thus directly affecting the child in everyday face-to-face interaction, childhood social status in society (macro-level) influences the child primarily via the parents. Thus, it may be assumed that psychosocial mechanisms (eg, social support, social influence and social engagement) are more directly linked to the former type of structure.34 It has been suggested that these mechanisms influence health through three pathways: the psychological, the behavioural and the physiological.35 In the present study, some of the largest relative differences in disease-specific morbidity by peer status were found in disease-specific groups that were psychologically related (eg, mental disorders, suicide attempts and intentional self-harm) or affected by behavioural risk factors (eg, diabetes and circulatory diseases such as ischaemic heart disease).7 Hence, this study adds support to the notion of peer status being linked to health through behavioural and psychological pathways and, as such, influencing diseases with these particular kinds of aetiology. The results can also be seen to support the physiological pathway, for instance through obesity.

Peer status is thought of here as influencing morbidity, but one may argue that diseases that may have an early onset, such as mental disorders, precede the establishment of peer status. However, previous studies have found that peer status predicts depressive symptoms irrespective of symptoms at an earlier point in time.36 Nevertheless, peer status and health are perhaps best seen as parts of a process of mutual influence, evolving over time.20 In a similar way, peer status appears to be continuously renegotiated. Status mobility is, however, restrained, particularly for individuals in the lowest and highest peer status positions.37 Thus, one may assume that the assignment of social positions in the peer status hierarchy is a fairly stable phenomenon. To the extent that peers constitute significant others,38 the exposure to expectations and access to resources that accompany any given peer status position are likely to have a long-term impact on the child’s identity, behaviour and ambitions, as well as the choices they make for themselves. This may in turn affect health development across the life course.

Few data materials permit a life-course approach to peer status in school and morbidity. The Stockholm Birth Cohort study used in this paper offers the opportunity to examine how sociometrically assessed childhood peer status at the age of 12–13 years relates to a wide variety of morbidity outcomes from the age of 20–50 years. The present study does, however, have some important limitations. First, the use of hospital discharge register data to indicate morbidity produces a possible bias if status position in school also predicts differences in help-seeking behaviour. Second, the cohort members are relatively young, meaning that diseases with a late onset are not well represented, which precludes the analysis of some second-level disease groups (eg, types of cancer). A somewhat different overall picture may have emerged if it had been possible to include all second-level groups. A third issue concerns comorbidity. For instance, alcohol-related disorders co-occur with other psychiatric disorders such as anxiety and depression,39 as well as with external causes such as suicide attempts.40 41 In addition, heart disease commonly co-occurs with diabetes.42 To complicate matters further, diabetes and heart diseases have been found to co-occur with depression,43 44 schizophrenia45 and anxiety.46 Thus, one may claim that a small group of comorbid individuals account for a large proportion of disease.26 As a consequence, they contribute to the gradients in a wide range of diseases, equalising differences in the magnitude of the effect of peer status between various diseases and leading to an overestimation of the results. When controlling for comorbid diseases in the analysis by including dummies, however, the associations between peer status and health-related diseases, and between peer status and psychologically related diseases, were reduced but only to a limited extent. Thus, the main findings remained unaltered (data not presented). Furthermore, if the association between peer status and morbidity was due to such circumstances, we would probably see a threshold effect for the marginalised group rather than a gradient across all peer status categories.

In conclusion, the results of this study, especially with regard to peer status being associated with such a wide range of morbidity outcomes, underline the need to understand how and why peer status has such a seemingly strong impact on health outcomes across the life course.

This is an important area of research and policy-making, given that most children attend school for several years and are thus subjected to the social ordering of this micro-level social world. However, if the universality of status distributions constitutes “a social law”,47 which seems to be the case, we should perhaps rather focus on how the psychosocial mechanisms between social status in school and health operate, and how they may be influenced in order to reduce the negative impact of peer status on health across the life course.