Article Text

Download PDFPDF

The combined influence of parental education and preterm birth on school performance
  1. Marit Gisselmann1,
  2. Ilona Koupil1,
  3. Bianca L De Stavola2
  1. 1Centre for Health Equity Studies (CHESS), Stockholm University/Karolinska Institutet, Stockholm, Sweden
  2. 2Medical Statistics Unit (MSU), Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, University of London, London, UK
  1. Correspondence to Dr Marit Gisselmann, Centre for Health Equity Studies (CHESS), Stockholm University/Karolinska Institutet, 106 91 Stockholm, Sweden; marit.gisselmann{at}


Background Social background and birth characteristics are generally found to be independently associated with school achievements but the underlying mechanisms are not fully understood. This study aimed to explore how parental education and shorter gestational age jointly influence school performance in a cohort of Swedish children.

Methods 10 835 children born between 1973 and 1981 were studied, the third generation of the register-based Uppsala Multigenerational Birth Cohort. Ordinal logistic regression models were fitted to estimate OR of achieving middle and high grades in Swedish language at age 16 years, relative to low grade, by parental education and own gestational age, adjusting for potential confounders.

Results In children from families with lower parental education, the adjusted OR of receiving a higher grade was 0.54 (95% CI 0.41 to 0.71) for preterm (<37 completed weeks) compared with full-term births. This estimate did not change when adjusted for several potential confounders (0.59; CI 0.44 to 0.79). When different cut-points were selected to define preterm birth, the estimated OR for those with low parental education decreased linearly from 0.83 (CI 0.72 to 0.96) using less than 39 weeks as the cut-point, to 0.52 (CI 0.30 to 0.90) using less than 35 weeks. There was no evidence of significant effects of shorter gestational age for children with parents from other educational groups.

Conclusions The disadvantage of shorter gestational age on the chance of achieving higher grades in Swedish language was confined to children from families in which none of the parents had higher education. This suggests that the detrimental influence of shorter gestational age on school performance in language may be avoidable.

  • Educational status
  • gestational age
  • premature infants
  • school performance
  • social paediatrics
View Full Text

Statistics from

It has long been known that children from families with higher socioeconomic status are more likely to go on to higher education,1 2 which is often required to achieve jobs classified as higher social class. In Sweden, education has been claimed to be the most important mechanism in the transfer of socioeconomic capital from parents to children.3 This transfer is composed of two processes: the influence of social background on school achievement and the influence of social background on continuation to higher education net of school achievement.4 In addition to the demonstrated influence of social background on educational achievement, studies of birth characteristics and school achievement as well as studies of birth characteristics and cognitive ability almost invariably find associations between these factors.5 6 Drawing from several fields of research, we hypothesise that cognitive stimulation, cognitive ability and learning in the school situation are important mediators between socioeconomic background and school achievement (as measured by school grades and attained education), as well as between birth outcome and school achievement. Our theoretical model is based on the assumption that cognitive ability can be enhanced with cognitive stimulation.7–9 and that cognitive ability has an important influence on school grades.10

Socioeconomic background and school achievement

That lower socioeconomic status of the family is associated with a lower level of school achievement and cognitive ability is well established.11 A vital part of family life is communication, and the quality of early childhood interaction has been shown to influence literacy skills development several years later.12 13 In families with higher socioeconomic status, parents and children will be interacting in a richer vocabulary environment.14 A study of verbal communication and children's vocabulary found that the number of words spoken around and directly to the child in early childhood (between 12 months and 3.5 years of age) was indeed the most important single factor to increase vocabulary growth.15 By the age of 4 years, children in families of higher socioeconomic status had heard three times more words than those in families with lower socioeconomic status (30 millions and 10 millions approximately), and these also contained a richer vocabulary. The correlation between the number of words spoken and child vocabulary growth was r=0.92.15

In families with higher socioeconomic status, children also have higher aspirations both in terms of school grades and in terms of the length of total education.16 17 This may influence children's own effort in school and their will to engage in learning activities over a longer period of time, factors that have been shown to be of importance for learning at school, presumably through cognitive stimulation.10 The extent and quality of learning that takes place at school will also depend on the student's behaviour at school. Research has shown that feeling safe and comfortable in an environment is enhancing the learning process as well as performance.18 19 Children from families with lower socioeconomic status may be less likely to feel at ease at school and they are also more likely to have behavioural problems.20 21

Birth characteristics and school achievement

Very low birth infants and infants born very preterm have been consistently found to be at higher risk of lower cognitive scores or cognitive impairment,22–24 with the degree of cognitive impairment being directly proportional to differences in birth weight or gestational age,5 and stretching across the whole range of birth weights.25 Furthermore, being born full term but small for gestational age has been associated with impaired cognitive ability.6 Likewise, being low birth weight or born preterm was found to be associated with a higher degree of attention deficit, as well as with other behavioural and social problems.5

When cognitive measures and school achievement were examined together, birth outcome was found to be positively associated with school achievement by means of earlier cognitive ability.26 27

Aims of the study

Although the mechanisms leading to disadvantageous birth outcomes are not always fully understood and are likely to include multiple underlying causes for shorter gestational age, all preterm births share the challenge of a larger part of brain development taking place in the postnatal period. Family socioeconomic status can be seen as the most important early environment that influences children's cognitive development. Parental education in particular is likely to be strongly linked to cognitive stimulation in the home environment. The aim of this study is to examine how a more advantageous cognitive environment, as measured by higher parental education, modifies the effect of preterm birth on school performance.

Materials and methods

Study population

The analyses were based on the third generation of the Uppsala Birth Cohort Multigenerational Study (UBCoS Multigen), grandchildren to the original cohort born in Uppsala from 1915 to 1929.28 29 The Swedish Medical Birth Register was established in 1973, and children born between 1973 and 1981 were included in the analyses of five-scale grades at the end of the 9th year of school (age 15–16 years), recorded during 1988–97. Of the 11 124 third generation singleton children born alive in 1973–81, we excluded those who died (116) or emigrated (79) before June in the year of turning 16 years when grades are awarded, as well as adopted children (94) giving us a study population of 10 835.


All data on birth characteristics were obtained from the Swedish Medical Birth Register. Gestational age was based on the best measure available according to the Swedish Centre for Epidemiology, either ultrasound or the last menstrual date. Preterm birth was defined as birth before 37 completed weeks of gestation. In further analyses, we also used alternative definitions of short gestational age ranging from less than 35 weeks compared with 35+ weeks, up to less than 40 weeks compared with 40+ completed weeks. The fetal growth rate was calculated by assigning standard birth weight for gestational age (z-score) by sex from all births in the Medical Birth Register for the respective years. Year of birth, gender, maternal age and parity were included as potential confounders.

Information on parental education was used to capture educational resources and values within the household and was obtained from the Swedish Census 1990. We categorised education as higher or lower, in which higher education corresponded to at least 3 years of secondary education, which typically corresponds to the academically oriented profile required for admission to tertiary education. When information was available for both parents, families were classified into concordantly high (both parents with higher education), concordantly low (both parents with lower education) and discordant (one lower, one higher). When information was available for one parent only (8% of families), education was classified as concordantly low or high. The effects of mother's and father's education in particular was analysed among children from families with discordant education.

School performance was obtained from the Swedish National Agency for Education and measured by grades in various subjects. We used scores in Swedish language, ranging from 1 (lowest) to 5 (highest) recorded at the end of grade 9 (typically age 15–16 years). This is the last year of compulsory school, with grades being used to apply for admission to secondary school and is therefore a good marker of school achievement. Swedish, mathematics and English are generally considered the main subjects studied. Grades in Swedish only were used in our analysis, as those in the latter two subjects were not fully comparable across alternative teaching streams over the study period. For reasons explained below, we collapsed Swedish grades 1 and 2, and 4 and 5, to create a new scale with three categories (low, medium, high).

Statistical methods

To capture the information in our data most effectively we modelled the odds of achieving (collapsed) low, medium or high grades for children with different parental education and gestational age by ordinal logistic regression. This enabled us to present odds of receiving a higher grade in comparison with lower grades in one single OR, representing both the differential between low and medium grade as well as between medium and high grade. The assumption of constant effect across consecutive categories was tested and found to be met when using this categorisation, whereas it was violated when the original five level grades were used instead.

Confounding by maternal age, parity, gender and standardised size at birth was assessed by comparing crude and adjusted estimates. Year of birth was considered an a priori confounder because grades are known to increase with time, and year of birth was thus always controlled in models. Effect modification between the two exposures of interest—parental education and preterm birth—was tested by including an interaction term. Maternal age was included in the models as a quadratic function, birth year of the child as yearly age bands, standardised birth weight for gestational age as linear function and parity as a categorical. To find out if our main result was driven or distorted by particular subgroups of the study participants, analyses were repeated excluding in turn one of the following groups: very preterm births (<32 weeks of gestation); children born with congenital malformations; children born with more serious congenital malformations; children with information on education from only one parent and children with one foreign-born parent. From a model in which parental educational category and gestational age were allowed to interact we calculated the OR for achieving higher grades in Swedish for preterm births in comparison with term births by educational category, holding the rest of the variables constant. To summarise these results, the model-predicted probabilities of achieving each grade were then calculated by educational category and preterm/full-term birth.

To explore the influence of shorter gestational age beyond the common definition of preterm as less than 37 completed weeks of gestation, we repeated the analysis for different cut-points of gestational age. Less than 34 completed weeks contained too few cases of preterm, therefore the minimum and maximum cut-points we considered were less than 35 and less than 40 completed weeks of gestation.

Results were considered significant if p<0.05. All analyses were carried out in Stata 10.


Table 1 summarises the distribution of the variables of interest and, for each of their categories, their percentage distribution of preterm births and grade in Swedish language. The variables with the most missing values were grade in Swedish (3.2%) and parental education (1.8%), leading to 10 242 births (of 10 835; 94.5%) with complete data on all variables included in the regression analyses. The overall proportion of preterm births in the study participants who were singleton live births was 4% (N=437). The proportion of preterm births among the subset used in the regression analyses was also 4% (N=404). The concordant low parental educational category held approximately 44% of the study participants, with just under 23% of their children receiving higher grades in Swedish. The concordant high parental educational category included approximately 27% of the study participants, with approximately 58% of their children gaining higher grades in Swedish. For 60% in the discordant parental education group the father had higher education and the mother lower, with their children faring mid-way between the other two parental groups. In contrast, the proportion of preterm births among the children of these three parental groups were fairly similar (3.9–4.2%, table 1).

Table 1

Characteristics of boys and girls from the third generation Uppsala Birth Cohort Multigeneration Study born in 1973–81 (N=10 835)

The year of birth adjusted (‘minimally adjusted’ hereafter) OR of achieving higher grades in Swedish, separately estimated for each variable of interest, show that odds are greater in children with higher parental education, children born from full-term pregnancies, those with higher fetal growth rate and those who were female and mothers with lower parity (table 2). The minimally adjusted OR for achieving higher grades of children with preterm birth was 0.75 (CI 0.62 to 0.90), and changed to 0.81 (CI 0.66 to 0.98) when all the potential confounders were included (table 2, model 3). However, when educational categories and gestational age were allowed to interact, the odds of achieving higher grades were lower for preterm birth relative to full term only for the low educational category, OR 0.59 (CI 0.44 to 0.79, table 3, model 5). For those with discordant or concordantly high parental education, the OR for preterm birth were close to one (0.95, CI 0.66 to 1.37) or above one (1.21, CI 0.82 to 1.82) and were not statistically significant. These main results were not altered by the exclusion of any of the subgroups described in the Methods section.

Table 2

Minimally and multiply adjusted OR of achieving higher grades in Swedish by entry variables (n=10 242)

Table 3

Multiply adjusted OR of achieving higher grades in Swedish for children born preterm, calculated by parental education (n=10 242)

Figure 1 illustrates how children born into families with high education had the highest predicted probability of achieving the highest grades, irrespective of whether they were born preterm or full term (predicted probability ≈0.6). This probability of achieving the highest grades decreases with decreasing parental education. In particular, the predicted grades distribution for children born into families with lower education is also shifted to the left, relative to that of the other children. In the lower educational category, children born preterm have the highest probability of receiving the lowest grades (predicted probability ≈0.4), which is substantially higher than that predicted for full-term births in the same educational category (predicted probability ≈0.3).

Figure 1

The probability of achieving a certain grade in Swedish by parental education and length of gestation (predicted from model 5).

When ordinal logistic regression (calculated as for model 5, table 3) was fitted using different cut-points to define short gestation, significantly lower odds for achieving higher grades in Swedish were still found to be restricted to the lower parental educational category: OR decreased linearly from 0.83 (CI 0.72 to 0.96) when the cut-point was set at less than 39 completed weeks (in comparison with a gestational age 39+) to 0.52 (CI 0.30 to 0.90) when it was set at less than 35 compared with 35 + weeks (table 4), with no clear indications of a threshold effect.

Table 4

Multiply adjusted OR of achieving higher grade in Swedish for children with shorter gestational age in comparison to longer (n=10 242)

We repeated the analyses splitting the discordant parental education group to explore whether there were differential effects if the mother or the father had the lower educational level. Results did not identify any substantive differences (OR 1.04, CI 0.63 to 1.70 and OR 0.85, CI 0.49 to 1.47, respectively, using 37 weeks as the cut-point for gestational age) although they were based on small numbers of preterm babies.


Our main finding is that the long-term disadvantage of shorter gestational age on educational performance was confined to children in families in which both parents have lower education. Together with the evidence on social inequalities in preterm birth, our finding that social background might protect from negative consequences of preterm birth renders preterm birth a potentially important mechanism in the reproduction of social inequalities across generations.

How well founded then is our conclusion? The comparisons that are possible to make with population data inform us that individuals in our data are a little better off in terms of the proportion of preterm birth (of the live-born singletons 4.8% were preterm in Sweden 1973–81, and 4.3% in all UBCoS 11 124 births for these years), but doing about as well in school as the population (mean grade over the years 1990–7 was 3.2 for all nine graders in Sweden and in our study population). The other strengths of our register-based study are that we have a relatively small proportion of missing data and the quality of data is believed to be good.30 We thus believe our results can be generalised to the respective cohorts of children born in Sweden, with the possible exception of families in which both parents have immigrated as these are not included in our study.

The UBCoS Multigen data study has some limitations, however. First, we did not have any indication of cognitive ability although we used the grades taken at age 16 years, which are a prerequisite for securing a place in secondary education. On the other hand, using scores at such ages may miss important variations at earlier ages. Research indicates that cognitive development in early childhood is extremely important for later school achievements, particularly in terms of language development.15 There is also evidence that cognitive pathways in later childhood are modified by parental education, for example by parental support with homework, which was found to be advantageous for school performance only if the parents were well educated.31 Therefore, unfortunately, we could not discern at what time period during childhood the consequences of preterm birth were amended for the more educated group. Furthermore, parental education was measured in 1990 when children were 9–17 years old and is most likely a good proxy for childhood social environment. Our results are based on language scores in children, known to be especially influenced by social background. Perhaps this is not the case for all school subjects, or other measures of school or cognitive achievement such as ability or intelligence tests. Further register-based research extending over a longer period and more detailed information do seem a fruitful way to continue research into how early social and biological environment influences health and social career in later life.

When interpreting our results in terms of the specific importance of parental education, we need to consider the possibility of clustering and the accumulation of social problems (in addition to less cognitive stimulation) in a disadvantaged group contributing to the poorer school achievement. In the low educated group there is only one indication of this: a slightly larger proportion of grades are lacking in this group (3.9% in comparison to 2.3% in the other groups). As our study participants were descendants of a Swedish cohort, children could have no more than one foreign-born parent. This group constitutes only 6% of the analysed children and their parents are not overrepresented among the lower educated. More importantly, the lower educational group constitutes approximately 44% of our study population and is therefore not an extreme group.

Concerning potential biological differences between the groups of children investigated in our study, there is one important issue that cannot be directly addressed in ours or similar studies. Namely, as has been suggested for causes of low birth weight in the case of the low birth weight paradox, the composition of causes of shorter gestational age might differ between the educational groups.32 If this is the case, and if the degree to which these causes are associated with impaired cognitive development varies, this could be the underlying drive for our results. Another reason could lie in the role of fetal growth. We have investigated this by including interaction terms between standardised size at birth and parental education group, but these were not significant. We also included interaction terms with the other potential confounders, and again these did not affect our main results to any appreciable extent. We were also able to assess whether our results were influenced by any particular group: excluding very preterm births or children with more or less serious malformations did not alter our main results.

The nutritional status of the pregnant mothers and children during early life development could also be a potentially important mechanism linking poor birth outcome with later educational disadvantage in children. It has been hypothesised that anaemia and iron deficiency affect cognitive development in children.33 Clear social gradient in the prevalence of iron deficiency has been shown in the UK, and it is plausible that iron deficiency recently reported among infants and young children in Sweden is also more common among children from socially disadvantaged families.34 35

We conclude that in spite of the limitations of our study, our main finding that the environment provided by parents with higher education is more advantageous for school performance is likely to be robust. This result is also in agreement with the work of Wang et al.36 In their study of Taiwanese 15-year olds, the disadvantage of lower birth weight and preterm birth in terms of scores in Mandarin, and some scores in mathematics and science, decreased with increasing paternal education. Furthermore, if paternal education was classified as college or more (the highest group), there was no disadvantage for preterm normal birth weight children.36

The importance of education for life chances in the western world continues to increase. During the past decades survival rates have increased for preterm and especially very preterm children. With more children carrying the risks associated with adverse birth outcomes throughout childhood, the influence of birth outcome on school achievement is likely to be greater in the population and will contribute to reproducing social inequalities in the next generation. Most importantly, our results suggest that environmental interventions may be possible for tackling the long-term effects of preterm birth on school performance and related inequities.

What is already known on this subject

Shorter gestational age and lower parental education are associated with educational outcomes.

What this study adds

The negative influence of shorter gestational age on grade in language was confined to children from families in which none of the parents had higher education.


The authors would like to thank Rawya Mohsen and Bitte Modin for help with data management.


View Abstract


  • Funding The UBCoS Multigen study is supported by grants from the Swedish Council for Working Life and Social Research (grant no 2007-1010) and the Swedish Research Council (2006-7498). MG (grant no 2008-1309) and IK (grant no 2006-1518) are currently funded by the Swedish Council for Working Life and Social Research. BLdS was supported by a travel grant from the Swedish Council for Working Life and Social Research (grant no 2006-1276) while working on this study.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the Regional Ethics Ccommittee in Stockholm (Dnr 03-117, 04-944T).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.