IJE Advance Access originally published online on November 12, 2005
International Journal of Epidemiology 2006 35(2):458-465; doi:10.1093/ije/dyi239
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Does the primary school attended influence self-reported health or its risk factors in later life? Aberdeen Children of the 1950s Study
1 MRC Social and Public Health Sciences Unit, University of Glasgow, Glasgow, UK.
2 London School of Hygiene and Tropical Medicine, London, UK.
* Corresponding author. MRC Social and Public Health Sciences Unit, University of Glasgow, 4 Lilybank Gardens, Glasgow G12 8RZ, UK. E-mail: r.dundas{at}msoc.mrc.gla.ac.uk
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Background Adult health and its determinants are influenced by the environment in childhood. The school attended is known to affect the health behaviours of pupils while still at school. Little is known about the long-term influence of school attended on health.
Methods A total of 7095 respondents (mean age 47 years) to a follow-up questionnaire who attended primary school in Aberdeen, UK, provided information on self-reported health; self-reported high blood pressure; GHQ-4; smoking status; alcohol intake; and obesity. Variance partition coefficients (VPCs) summarized the variation in adult health outcomes and behaviours across schools. Multilevel logistic regression was used to estimate the contribution of school to variation in the outcomes taking into account individual-level and school-level factors.
Results There was some variation across schools in the proportion of adults reporting poor self-rated health (VPC = 0.020) and smoking (0.019). Higher VPCs were found for factors potentially confounded with school: paternal social classes (I&II) (0.45) and gender (0.44). Age at leaving secondary education (0.28) and income (0.10) varied across schools. The effects of primary school diminished after adjusting for individual-level childhood risk factors. The further addition of adult risk factors attenuated these childhood effects. After full adjustment there was no effect of the primary school attended for high blood pressure, current smoking, alcohol intake, and obesity, and negligible effects for the other outcomes.
Conclusions Contrary to our expectations, we found little evidence of any relationship between primary school and adult self-reported health or behaviour. This is surprising given the extent to which characteristics known to be associated with adult health were clustered within schools.
Keywords Schools, adult, health, variance partition coefficient
Accepted 10 October 2005
Adult health and its determinants are known to be influenced by the environment in childhood. Parental social class, indicative of childhood conditions, has been shown to be related to mortality in adulthood,1,2 with social class differentials increasing with age. The inverse relationship between birthweight and coronary heart disease in later life is well documented.3,4 Childhood intelligence is known to be related to adult mortality,5 adult social class, and the relative affluence/deprivation of the area of residence in adulthood.6 These relationships suggest that inequalities in adult health may have their origins in childhood.
Adult factors are also associated with adult ill health or mortality, but it is not always clear whether these relationships are independent of the relationship with childhood factors or mediate the same. The relationship between birthweight and CHD appears to be influenced by adult body mass index (BMI),7 and higher maternal height has been shown to be protective against being in the highest risk group (low birthweight/high adult BMI).8 Educational attainment and adult social class appear to eclipse the effect of parental class on mortality,9 and educational attainment attenuates the effect of parental class on cardiovascular disease risk factors.10 Among men cardiovascular disease and stroke mortality have been shown to be similarly related to childhood and adult social circumstances,11 but smoking-related cancer mortality was considerably more strongly related to adult than childhood circumstances.12
The school attended is known to affect the health behaviours of pupils while still at school. A recent international review of the literature reported large differences in smoking prevalence between similar schools.13 School effects have been shown to exist for other health-related behaviours, notably drinking and drug use.14 Such findings are in line with studies showing the importance of school on educational attainment.15,16 In terms of education achievement, there is some evidence that primary school is more influential than secondary school.17,18
Little is known, however, about the long-term effects of the school attended on health. A recent 35 year follow-up of a large cohort of children first studied in primary schools in the 1960s has allowed us to investigate such long-term effects of primary school on adult health. Since we expect there to be differences between schools in their intake (in terms of their social composition), this paper investigates the extent to which there are differences in adult health outcomes or behaviours between primary schools, and the extent to which any such differences may be explained by differences in school composition or influenced by adult factors which may themselves be mediated by the primary school attended (e.g. income).
|
Methods |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
The Children of the 1950s Study is been described elsewhere.19 Briefly, it is based on a survey of the total population of all children born in Aberdeen (1950–56) who attended primary school in Aberdeen, Scotland, in December 1962. The aim of the initial survey was to ascertain the population prevalence of ‘mental subnormality’ (learning disability) in this population of 5- to 12-year olds.20 A total of 52 primary schools in Aberdeen took part in the original study. The cohort was revived in 1998 and the original 12 150 subjects were traced through the National Health Service Central Registry. Vital status and whereabouts were established for 98.5% and a follow-up questionnaire was mailed to 11 321 surviving participants, starting in May 2001.21 The main focus of this paper is the influence of school on adult health. Subjects who were in schools for blind, deaf, disabled, and intellectually impaired were not included in this analysis.
In this paper we consider three adult health outcomes based on the questionnaire: self-rated health (poor or fair health versus good health), self-reported prevalence of doctor diagnosed high blood pressure, and mental health [measured using 4 items from the 12 item General Heath Questionnaire (GHQ-4)].22,23 The four questions were as follows: In the past few weeks have you been (i) feeling reasonably happy, all things considered; (ii) able to enjoy your normal day to day activities; (iii) losing confidence in yourself; and (iv) feeling unhappy and depressed? GHQ-4 caseness was recorded as yes if any of the responses was adverse, and no otherwise.22 There were four adult health risk factors: current and ever smoking, alcohol use (drinking more than 21 units per week), and BMI (based on self-reported height and weight). As well as a continuous measure BMI was categorized in two ways: overweight (BMI > 25) and obese (BMI > 30).
The childhood measures used were paternal social class at birth; birthweight; IQ test scores at ages 7, 9, and 11; and age at leaving secondary education. IQ tests were routinely administered to children in schools in Aberdeen.10,19 We also adjusted for adult social class and income. Smoking, alcohol use, and BMI were considered both as outcomes in themselves and as adjustment factors for other outcomes. The childhood and adult measures were separated into two groups: school compositional factors and mediating factors. The compositional factors are the individual-level explanatory factors which cannot be influenced by the school the child attended but might confound the association of school with health outcomes. These were sex, paternal social class, maternal height, more than 3 children in the family in 1962, and birthweight. Age at the time of questionnaire was also controlled for in the compositional factor models as it may confound the relationships with the outcome measures. The mediating factors were the individual-level explanatory factors which might be influenced by primary school—IQ test scores at ages 7, 9, and 11; age at leaving secondary education; adult social class; income; smoking; alcohol use; and BMI.
Variance partition coefficients (VPCs) were calculated to summarize the extent to which the adult health outcomes and risk factors varied across schools. (The VPC is equivalent to the intraclass correlation coefficient in variance components models.24) VPCs were also calculated for the compositional and mediating factors. The VPC measures the extent to which values of the dependent variable are similar for respondents attending the same primary school.
It accounts for the clustering within the data by comparing the variance within primary schools with the variance between primary schools. A small value for the VPC implies that the within-school variance is much greater than the between-school variance (i.e. little of the total variability between individuals is due to the school attended). Categorical variables with more than two categories were progressively grouped to allow for the calculation of a VPC. For example, income is categorized into three categories: 
£10 000, £10–20 000, and £30 000+. A VPC was calculated for income £10 000 compared with everyone earning >£10 000; then £10 000 was grouped with £20 000 and a VPC calculated for £20 000 compared with those earning >£20 000.
Multilevel logistic regression was used to assess the effect of primary school attended on adult health and risk factors for adult health. Each model had three levels: level 1, child; level 2, family; and level 3, school. Family was included as a level to ensure that the school-level variance was not the result of children from the same family attending the same school. The aim was to estimate the school-level variance unadjusted, and then adjust the variance for compositional factors, and then calculate the amount of variance that could be explained by mediating factors. Initial models were fitted for each outcome variable using only the compositional factors and age and sex. Further models were fitted retaining significant compositional factors and adding significant mediating factors. For binary outcomes, we used a second order penalized quasilikelihood approximation.25
To take into account the context of the school—and to see whether there was a school ‘social class’ effect independent of the effect of paternal social class—an aggregated variable comprising the proportion of children in each school with fathers in the professional and managerial social classes (social classes I&II) was also considered. The aggregated school variables were calculated using original 12 150 subjects from the 1962 survey. We tried to obtain true contextual data on the schools, such as pupil–teacher ratio, state of school buildings, absenteeism rate, and quality of teachers, but these were not available in systematic or comparable formats across all schools.
|
Results |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Follow-up questionnaires were received from 7095 individuals (63% response rate). The sample size for each of the 42 schools ranged from 4 to 435; 3 schools had <40 study subjects. There were 7 single sex schools: 3 for boys and 4 for girls. Responses were received from 3387 males and 3708 females with a mean age at follow-up of 47 years. The 7095 individuals were in 6016 families; 352 families had 2 children, 79 had 3 children; and 3 had 4 children. A total of 323 siblings from the same family attended different schools. There was little mobility between schools; 91% of individuals were at the same school from the ages 5 to 12 years.
Table 1 shows the school aggregated prevalence, interquartile range, and range for the school compositional factors and test scores. The table shows that there is considerable variation between the schools in terms of percentage of children with fathers in social classes I&II and social classes IV&V, family size >3, and average test scores at age 7 and 9 years. There was some variation between schools in percentage of children with mothers of height <150 cm, but less variation between schools in percentage of children with birthweight <2500 g and average test score at age 11 years.
|
Table 2 summarizes the extent to which adult health outcomes and determinants and confounding and mediating factors clustered within schools. There was some variation across schools in the proportion of adults reporting poor self-rated health (VPC = 0.020), obesity (0.010), current smoking (0.019), and drinking >21 units per week (0.012). (This means that, for example, 2% of the variation in poor self-rated health was accounted for by the primary school level). Higher VPCs were found for child (individual)-level factors potentially confounded with primary schools: paternal social class (0.451 for children whose fathers were not social class I or II, professional, or managerial), family size (0.139), and gender (0.442). The high VPC for paternal social class is indicative of the strong social patterning distinguishing between many schools at the time. Test scores and age at leaving secondary education also varied across schools.
|
Health outcomes
The school compositional factors that were significantly related to poor self-rated health were lower paternal social class, family size >3 children in 1962, and birthweight <2500 g. Older adult age was also significantly related to poor self-rated health. When the mediating variables were included in the model none of the compositional variables remained significant. Income, smoking, age at leaving secondary school, and IQ test score at age 7 years were significantly associated with poor health (Table 3). The lower the income the more likely the respondent was to have poor self-rated health. Smokers and ex-smokers were more likely to have poor self-rated health. Adjusting for the compositional factors reduced the school level variance by 83% (from 0.06 to 0.01) (Table 5). On inclusion of the mediating variables the variance fell by a further 90%.
|
Sex, age, maternal height, and birthweight were all associated with self-reported high blood pressure. These remained significant when the mediating variables were included in the model. The mediating variables significantly associated with high blood pressure were health behaviour variables—smoking, drinking >21 units of alcohol per week, and overweight and obesity. Adults who had a low birthweight (2500g) were more likely to have self-reported high blood pressure. Males were more likely to have high blood pressure than females. Current smokers were less likely to have high blood pressure than non-smokers (a relationship reported elsewhere).26 The compositional variables explained all of the variation between schools (Table 5).
The compositional factors associated with GHQ-4 caseness were being a woman and being in a family with >3 children in 1962. When the mediating factors were included these were no longer significant. Decreasing income, smoking, alcohol use, and obesity were associated with GHQ-4 caseness. The school level variance was similar for the null model (0.019) and the model including compositional factors (0.020). It decreased to 0.012 when the mediating factors were included.
Health behaviours
Paternal social class and family size were associated with current smoking in adulthood. Children from large families were more likely to be smokers in adulthood compared with those from smaller families. Once mediating variables were included in the model paternal social class was no longer significant. Instead adult socioeconomic measures were significant; age on leaving secondary education, adult social class, and income were associated with smoking. People with a lower BMI were more likely to smoke (Table 4). When only significant compositional variables were included in the model they explained all of the school level variance (Table 5). Including the mediating variables, the school-level variance remained zero.
|
|
The results for the variables that were associated with ever smoking were similar to those for current smoking apart from IQ test score at age 7 years being significantly related to ever smoking. The variance between schools decreased to 0.016 when adjusted for compositional and mediating factors (Table 5).
Alcohol consumption of >21 units per week was associated with sex and maternal height <150 cm. These remained significant when mediating factors were included. Income, current and former smoking, BMI, and IQ age 9 years were all associated with alcohol use, and these factors explained all of the variation between schools.
Paternal social class and family size were significantly associated with obesity. These variables explained all of the variation between schools (Table 5). Paternal social class remained significant when the mediating variables were included in the model. Age at leaving secondary school and smoking were associated with obesity; smokers and ex-smokers were less likely to be obese than non-smokers. Those leaving secondary school at age 16 years were more likely to be obese than those who were of age 
17 years. There was a weak association with adult social class, and those with a lower IQ test score at age 9 years were more likely to be obese. There was still no variance between schools following the addition of the mediating variables.
The contextual effect of school social class on self-rated health was examined. Significant relationships were found with self-rated health, current smoking, ever smoking, and obesity. Adults who in childhood attended a school comprising a higher proportion of children with fathers from social classes I&II were less likely to report poor self-rated health, less likely to be current or ever smokers, and less likely to be obese. The relationship with smoking and obesity disappeared when fully adjusted for the mediating factors. For self-rated health a weak effect remained (P = 0.061); for a 10% increase in proportion of children with fathers from social classes I&II at a school the odds ratio of reporting poor self-rated health in adults was 0.94 (95% confidence interval: 0.88–1.003).
|
Discussion |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
The VPCs for the individual-level factors that could not be affected by the school the child attended but may confound the school-level effects—sex, paternal social class at birth, maternal height, family size, and birthweight—are quite high. The VPCs for the health outcomes and determinants are low; the highest was for self-rated health. However, only 2% of the variation in self-rated health is due to the primary school attended. This is surprising given the pattern of clustering in schools of factors which are associated with poor health and its determinants—paternal social class, adult social class, educational attainment, and income. The highest unadjusted VPC is for paternal social class, but paternal social class is no longer significantly associated with poor self-rated health following adjustment for adult income and smoking status. An advantage of the design of this study is that father's social class was recorded when the participant was in primary school so is not subject to recall bias.
The findings of such small effects of primary school on adult health contradict the current research on the relationship between schools and contemporary health behaviours,13,14 and suggest to us three possibilities. First, it may be that the effect of school diminishes over time; 35 years after the event the differing social influences throughout life have balanced out to leave no residual effect of the primary school. Second, it could be that the primary school is the wrong unit of analysis; health behaviours may be learnt—and health generated or destroyed—in other social contexts such as secondary school, neighbourhood, or other social network. Third, it may be that children or schools in Aberdeen from 35 years ago differ from those of today such that had research been conducted into children's health behaviours at the time there may not have been school effects—which would suggest that it is not possible to extrapolate this research to give insights into the future health of today's schoolchildren. An alternative explanation that mobility of students between schools may have caused an underestimation of the between-school variance is unlikely as the mobility between schools in the study was low (9%).
The response rate to the adult questionnaire was 63%. Although this is fairly high for a postal questionnaire some non-response bias exists. It has been shown that adults whose fathers' social class was higher and those with a higher IQ were more likely to respond.21 This means that adults with lower paternal social class and lower IQ are under-represented in the study. Since both paternal social class and IQ are patterned by primary school it is possible that our findings of only small or non-existent school effects would be affected if the relationship between these childhood factors and adult health or health determinants were different among the non-respondents.
The models were fitted using family as an intermediate level. This nested structure assumes that all children from the same family attend the same school. About 5% of children from the same family attended different schools, implying that a cross-classified model may be more appropriate.27 The simpler, strictly hierarchical model that we adopted treats such children as though they were from different families. This may result in a slight underestimate of the variance between schools if the variance between families is over-estimated, but given the low level of cross-classification and the small variances involved it is unlikely that this would have a material impact on any of the results. The children were in different classes in each school and this could have been used as a level. Unfortunately school class was not adequately recorded and so was not used. If class had been used the variance of interest would be the variance between classes plus the variance between schools. The consequence of ignoring the between-class variance is to potentially underestimate the variance between schools since some of the variance between classes would be estimated as being due to differences between families.
The relationship between childhood and adult factors, adult health behaviours, and health outcomes varied depending on the outcome considered. Although differences in self-rated health and GHQ-4 caseness were related to childhood factors, these relationships were mediated by adult income and health behaviours. Self-reported high blood pressure on the other hand was independently related to maternal height and birthweight after adjusting for elevated rates associated with alcohol use and high BMI and decreased odds among current smokers.
Current and ever smoking and adult obesity were related to many of the childhood social indicators in addition to the adult factors and health behaviours. Simultaneous adjustment showed that independent effects remained relating to educational attainment and adult social class (also with income for smoking) after adjustment for the other health behaviours. However, the effect of family size remained significant for smoking and of paternal social class for obesity. Weekly alcohol use in excess of 21 units was associated with maternal height, adult income, and other health behaviours and all these relationships remained significant when adjusted simultaneously. The mixed message is then that whilst childhood factors appear to be related to adult health outcomes, they are mediated by the effects of adult social factors and health behaviours. There is more evidence that the adult health behaviours are related to childhood social circumstances in addition to adult circumstances.
Even in the unadjusted model the effects of primary school on adult health were small. This was unanticipated given the extent to which characteristics known to be associated with adult health—paternal social class, age at leaving secondary education, and income—were clustered within schools. The inclusion of the childhood factors indicating school composition explained much of the variation between schools, and any remaining variation all but disappeared once the mediating adult factors were taken into account. Although estate agents and parents place emphasis on the quality of primary school as a reason for moving house, this paper implies there are only minor differences across primary schools in adult self-reported health; it may be that the neighbourhood and parental environments are the important contexts.
|
Acknowledgments |
|---|
The authors are very grateful to Raymond Illsley for providing them with the data from the Aberdeen Child Development Survey and for the advice about the study. Graeme Ford played a crucial role in identifying individual cohort members and in helping initiate the process of revitalization. Doris Campbell, George Davey Smith, Marion Hall, Bianca de Stavola, Susan Morton, David Batty, David Godden, Diana Kuh, Debbie Lawlor, Glyn Lewis, and Viveca Östberg collaborated with D.A.L. to revitalize the cohort. Heather Clark managed the study at the Dugald Baird Centre, Aberdeen with the assistance of Margaret Beveridge. The authors would also like to thank the staff at ISD (Edinburgh) and GRO (Scotland), who were also involved in the revitalisation. The authors acknowledge the study participants who responded to the questionnaire 36 years after the original survey was conducted. The MRC Social and Public Health Sciences Unit is supported by the Chief Scientist Office of the Scottish Executive Health Department and the Medical Research Council.
KEY MESSAGES
|
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
1 Osler M, Andersen A-MN, Due P, Lund R, Damsgaard MT, Holstein BE. Socioeconomic position in early life, birth weight, childhood cognitive function, and adult mortality. A longitudinal study of Danish men born in 1953. J Epidemiol Community Health 2003;57:681–86.
2 Pensola TH, Valkonen T. Mortality differences by parental social class from childhood to adulthood. J Epidemiol Community Health 2000;54:525–29.
3 Barker DJP. Mothers, Babies, and Diseases in Later Life. London: BMJ, 1994.
4 Lawlor DA, Ben-Shlomo Y, Leon DA. Pre-adult influences on cardiovascular disease. In: Kuh D, Ben-Shlomo Y (eds). Life Course Influences on Adult Disease. Oxford: Oxford University Press, 2004.
5 Whalley LJ, Deary IJ. Longitudinal cohort study of childhood IQ and survival up to age 76. BMJ 2001;322:819–22.
6 Hart CL, Deary IJ, Taylor MD et al. The Scottish Mental Health Survey 1932 linked to the Midspan studies: a prospective investigation of childhood intelligence and future health. Public Health 2003;117:187–95.[CrossRef][Medline]
7 Frankel S, Elwood P, Sweetman P, Yarnell J, Davey Smith G. Birthweight, body-mass index in middle age, and incident coronary heart disease. Lancet 1996;348:1478–80.[CrossRef][ISI][Medline]
8 Power C, Li L, Manor O, Davey Smith G. Combination of low birth weight and high adult body mass index: at what age is it established and what are its determinants? J Epidemiol Community Health 2003;57:969–73.
9 Pensola TH, Valkonen T. Effect of parental social class, own education and social class on mortality among young men. Eur J Public Health 2002;12:29–36.
10 Lawlor DA, Batty GD, Morton SMB, Clark H, Macintyre S, Leon DA. Childhood socioeconomic position, educational attainment and adult cardiovascular risk factors: findings from the Aberdeen Children of the 1950s cohort study. Am J Public Health 2005;95:1245–51.
11 Næss Ø, Claussen B, Davey Smith G. Relative impact of childhood and adulthood socioeconomic conditions on cause specific mortality in men. J Epidemiol Community Health 2004;58:597–98.
12 Claussen B, Davey Smith G, Thelle D. Impact of childhood and adulthood socioeconomic position on cause specific mortality: the Oslo Mortality Study. J Epidemiol Community Health 2003;57:40–45.
13 Aveyard P, Markham WA, Cheng KK. A methodological and substantive review of the evidence that schools cause pupils to smoke. Soc Sci Med 2004;58:2253–65.[CrossRef][ISI][Medline]
14 West P, Sweeting H, Leyland AH. School effects on pupils' health behaviours: evidence in support of the health promoting school. Research Papers in Education 2004;19:261–91.[CrossRef]
15 Cuttance P. Intra-system variation in the effectiveness of schooling. Research Papers in Education 1988;3:183–219.
16 Thomas S, Sammons P, Mortimore P, Smees R. Stability and consistency in secondary schools' effects on students' GCSE outcomes over three years. School Effectiveness and School Improvement 1997;8:169–97.
17 Goldstein H, Sammons P. The influence of secondary and junior schools on sixteen year examination performance: a cross-classified multilevel analysis. School Effectiveness and School Improvement 1997;8:219–30.
18 Mortimore P, Sammons P, Stoll L, Lewis D, Ecob R. School Matters: The Junior Years. London: Open Books, 1988.
19 Batty GD, Morton SMB, Campbell D et al. The Aberdeen Children of the 1950s cohort study: background, methods and follow-up information on a new resource for the study of life course and intergenerational influences on health. Paediatr Perinat Epidemiol 2004;18:221–39.[CrossRef][ISI][Medline]
20 Birch HG, Richardson SA, Baird D, Horobin G, Illsley R. Mental Subnormality in the Community: A Clinical and Epidemiologic Study. Baltimore, MD: Williams & Wilkins Co, 1970.
21 Nishiwaki H, Clark H, Morton SM, Leon DA. Early life factors, childhood cognition and postal questionnaire response in middle age: the Aberdeen Children of the 1950s study. BMC Med Res Methodol 2005;5:16.[CrossRef][Medline]
22 Wiles N, Peters T, Leon DA, Lewis G. Birth weight and psychological distress at age 45 to 51 years: Results from the Aberdeen Children of the 1950s cohort study. Br J Psychiatry 2005;187:21–28.
23 Jacobsen B, Hasvold T, Hoyer G, Hansen V. The General Health Questionnaire: How many items are really necessary in population surveys? Psychol Med 1995;25:957–61.[ISI][Medline]
24 Merlo J, Chaix B, Yang M, Lynch J, Råstam L. A brief conceptual tutorial of multilevel analysis in social epidemiology: linking the statistical concept of clustering to the idea of contextual phenomenon. J Epidemiol Community Health 2005;59:443–49.
25 Goldstein H, Rasbash J. Improved approximations for multilevel models with binary responses. J R Stat Soc [Ser A] 1996;159:505–13.
26 Green MS, Jucha E, Luz Y. Blood-pressure in smokers and nonsmokers—epidemiologic findings. Am Heart J 1986;111:932–40.[CrossRef][ISI][Medline]
27 Browne W, Goldstein H, Rasbash J. Multiple membership multiple classification (MMMC) models. Stat Modelling 2001;1:103–24.
28 Snijders TAB, Bosker RJ. Multilevel Analysis. London: Sage, 1999.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||