Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Intelligence and all-cause mortality in the 6-Day Sample of the Scottish Mental Survey 1947 and their siblings: testing the contribution of family background

Intelligence and all-cause mortality in the 6-Day Sample of the Scottish Mental Survey 1947 and... Background: Higher early-life intelligence is associated with a reduced risk of mortality in adulthood, though this association is apparently hardly attenuated when accounting for early-life socio-economic status (SES). However, the use of proxy measures of SES means that residual confounding may underestimate this attenuation. In the present study, the potential confounding effect of early-life SES was instead accounted for by examining the intelligence–mortality association within families. Methods: The association between early-life intelligence and mortality in adulthood was assessed in 727 members of the 6-Day Sample of the Scottish Mental Survey 1947 and, for the first time, 1580 of their younger siblings. These individuals were born between 1936 and 1958, and were followed up into later life, with deaths recorded up to 2015. Cox regression was used to estimate the relative risk of mortality associated with higher IQ scores after adjusting for shared family factors. Results: A standard-deviation advantage in IQ score was associated with a significantly reduced mortality risk [hazard ratio¼ 0.76, p< 0.001, 95% confidence interval (CI) (0.68– 0.84)]. This reduction in hazard was only slightly attenuated by adjusting for sex and shared family factors [hazard ratio¼ 0.79, p¼ 0.002, 95% CI (0.68–0.92)]. Conclusions: Although somewhat conservative, adjusting for all variance shared by a family avoids any potential residual confounding of the intelligence–mortality associa- tion arising from the use of proxy measures of early-life SES. The present study demon- strates that the longevity associated with higher early-life intelligence cannot be ex- plained by early-life SES or within-family factors. V The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association 89 This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 90 International Journal of Epidemiology, 2018, Vol. 47, No. 1 Key words: mortality, intelligence, socio-economic status Key Messages A standard-deviation increase in IQ score was associated with a 24% decrease in mortality risk. After adjusting for all shared family factors, a standard-deviation increase in IQ score was associated with a 21% de- crease in mortality risk. Early-life socio-economic circumstances are not sufficient to explain the intelligence–mortality association. Introduction confounding, resulting from measurement error in early- life SES, may therefore be driving previously observed as- Identifying predictors of mortality and chronic disease has sociations to some extent, even after adjusting for apparent been the goal of many epidemiological studies from across early-life SES. various countries and populations. In cognitive epidemi- In a study of the link between intelligence in youth and ology, intelligence (general cognitive function) as measured later income inequality, Murray attempted to tackle this using psychometric tests in middle- and older-aged popula- SES-assessment problem by examining associations within tions has emerged as a predictor of longevity, with higher families. Having matched individuals to their nearest sib- intelligence test scores associated with reduced mortality 1–5 lings, Murray then examined whether within-sibling-pair risk. More recently, pre-morbid measures of cognitive intelligence differences could predict income differences ability in cohorts of children and young people have been within pairs. Examining the contribution of intelligence shown to be related to mortality risk up to seven decades 6–10 within families circumvented the need to obtain a range of later. SES measures, as it removed the variance accounted for by A key consideration in interpreting these results is the shared family environment (e.g. parental income, occu- understanding the role of early-life socio-economic status pation and education). Examining outcomes within fami- (SES). Early-life SES may be a key confounder in the lies therefore allows researchers to account for early-life cognition–mortality relation owing to its association with 11 12 SES without the need to operationalize and measure it. both childhood and adult intelligence, and its link to 7,13–16 The present study adopts a within-family method in later-life mortality and health. However, the causal order to examine the association between early-life intelli- relationship between early-life SES and intelligence, and gence and mortality while accounting for early-life SES. thus the nature of the resulting association with later-life 17,18 Families were established by linking members of the 6-Day mortality, is unclear. Researchers have typically at- Sample of the Scottish Mental Survey 1947 with their tempted to account for early-life SES by including it as a younger siblings. The 6-Day Sample is a group of individ- predictor in multivariable models, and many have uals (N¼ 1208) representative of the whole Scottish popula- observed that doing so does not attenuate the contribution tion born in 1936 and whose cognitive ability was tested at of early-life intelligence to mortality risk. 22,23 age 11 years old. Previous work with this sample has However, attempts to account for any contribution of shown early-life intelligence to be a significant predictor of early-life SES to the intelligence–mortality association are mortality from up to 67 years old, even after early-life SES limited by the way in which SES is operationalized. Given (interviewer-rated parental intelligence and personality, the breadth and complexity of SES, it is difficult to fully household cleanliness, etc.) had been accounted for. characterize early-life social circumstances. Researchers Importantly, the younger siblings of these individuals were commonly include proxy measures for early-life SES, such 19 7 tested on the same IQ-type test when they reached 11 years as parental occupation, parental income or participant’s old, and were recently linked to records of mortality. These own education. This has also resulted in a lack of consist- siblings shared similar early-life circumstances and upbring- ency between cohort studies in terms of the SES measure ings to their 6-Day Sample probands (e.g. parental SES and used. Given these shortcomings, it is likely that not all of household size), but may differ in their cognitive ability and the variance associated with early-life SES has been fully longevity. By examining the intelligence–mortality associ- captured or accounted for in studies examining the ation within families, each consisting of a 6-Day Sample association between intelligence and mortality. Residual Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 1 91 member and their siblings, it is possible, more comprehen- removed due to missing data (N¼ 17) resulted in a total sively than previously, to partition out the potential con- sample of 2307 individuals (6-Day Sample: N¼ 727; founding effect of early-life SES. Sibling Sample: N¼ 1580) from 728 families (mean family size¼ 3.13, SD¼ 1.57). Methods Study sample Assessments On 4 June 1947, almost all children born in 1936 and at- Intelligence tending school in Scotland sat the Moray House Test No. For both the 6-Day Sample and their siblings, intelligence 12 test of intelligence. This cohort of 70 805 individuals, was measured using the Terman-Merill test, Form L —an the Scottish Mental Survey 1947, comprised 88% of the adapted version of the Binet-Simon test of intelligence. Scottish population born in 1936 and 94% of the avail- This test included 129 items of both verbal and non-verbal able school population at the time. A subsample of this reasoning. Raw correct scores were converted into standar- 1936 birth cohort, the 6-Day Sample (n¼ 1208; 618 fe- dized IQ-type scores (M¼ 100, SD¼ 15). For the 6-Day males), was created by selecting individuals born in Sample, this test was administered in 1947, following their Scotland on the first day of every even-numbered month in completion of the Scottish Mental Survey 1947. Siblings, 1936, whether or not they completed the Moray House on the other hand, completed the test at age 11 years, be- Test of intelligence. The mean intelligence and geograph- tween 1948 and 1969. ical distribution of the 6-Day Sample has been shown to be similar to the full Scottish Mental Survey 1947 cohort. Mortality Members of the 6-Day Sample were given a second, indi- Vital status and date of death were obtained for each of vidually administered test of intelligence in 1947—the 22,28 the 6-Day Sample members and their siblings. This was Terman-Merrill revision of the Binet Test —and were achieved by linking both the 6-Day Sample and their sib- subsequently resurveyed about every year up to the age of lings to their respective administrative records held by the 27 years to collect information on education, family life, National Records of Scotland (NRS). This linkage was home environment, leisure activities, health and early- 23,25 approved by the Scotland-A Research Ethics Committee adulthood occupation. (Ref: 12/SS/0024), the National Services Scotland NHS Younger siblings (n¼ 1655; 798 females) of the 6-Day Privacy Advisory Committee and the Confidentiality Sample members were tested for intelligence using the Advisory Group of the Health Research Authority. same Terman-Merrill Binet Test as they approached age 11 Approval covered linkage without consent, up to years, and were also resurveyed into early adulthood. Of November 2015, under section 251 if the NHS Act 2006. the whole 6-Day Sample, 748 had siblings included in the The NRS used automated and manual tracing methods to follow-up surveys. The Sibling Sample, born between 1937 link identifiable information (date of birth, surname, fore- and 1958, ranged from first siblings (n¼ 748) to tenth sib- name and National Health Service number) for each indi- lings (n¼ 3), and siblings were on average 6.13 years vidual with their respective National Health Service younger than their 6-Day Sample probands [standard devi- Central Register records. ation (SD)¼ 3.96, Min.¼ 0.39 years, Max.¼ 22.36 years]. Individuals were censored at the end of mortality sur- Fourteen of the Sibling Sample were twins, though none veillance (30 November 2015 for most individuals). was twinned with their 6-Day Sample proband. Survival time was calculated as the number of days be- Combining the two cohorts resulted in a sample of tween the date of birth and either the date of death or cen- 2863 individuals. Of these, 79 individuals (6-Day Sample: soring date as appropriate. Individuals who had emigrated N¼ 4, Sibling Sample: N¼ 75) for whom follow-up data after completing the intelligence test but before the end of were not available were removed from any further ana- the surveillance period were retained in the sample, but lysis. As part of previous work, formal comparisons have were censored at the start of the month in which they shown that individuals lost to follow-up demonstrate embarked. higher childhood IQ scores (a difference of 4.7 IQ points), higher levels of schooling and higher SES relative to those retained in the 6-Day Sample. In order to ensure that Statistical analyses family-related factors could be shared, the present study focused on multiple-child families. Excluding those 6-Day Survival analyses were conducted using Cox proportional Sample members from single-child families (N¼ 460) or hazards regression. Hazard ratios were calculated for each those 6-Day Sample members whose siblings were predictor included in the model to indicate the Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 92 International Journal of Epidemiology, 2018, Vol. 47, No. 1 Table 1. Descriptive characteristics of the 6-Day Sample members and their younger siblings, and comparisons between groups 6-Day Sample (N¼ 727) Sibling Sample (N¼ 1580) Mean SD Mean SD Sex (N male/female) 363/364 819/761 IQ score* 100.28 19.05 98.31 17.60 Mortality status (N alive/dead) 437/290 1144/436 Time to death (years, from birth)** 63.99 11.65 57.93 14.35 Time to censor (years, from birth)** 79.43 0.28 72.93 4.12 Missing values were deleted listwise in each of the variable estimates. Time to death is calculated only for those who have died before the censor date; time to censor is calculated only for those still alive at the censor date. *t-test conducted between the 6-Day Sample and Sibling Sample, p¼ 0.019; **p< 0.001. proportionate change in mortality risk for a unit change in Table 2. Descriptive characteristics of the whole sample the predictor. In the first set of analyses, individual univari- (N¼ 2307) of 6-Day Sample members and their younger sib- able regression models were created to predict survival lings according to mortality status using either standardized (z-transformed) IQ scores, sex or Alive (N¼ 1581) Dead (N¼ 726) family size. Each model additionally included a random ef- fect of family to account for the fact that observations Mean SD Mean SD within each family are correlated. All of the predictors con- Sex (N male/female) 735/846 447/279 formed to the proportional hazards assumption (all IQ score* 100.73 18.20 95.09 17.28 ps> 0.28). In the second set of analyses, the effect of stand- Family size (people) 5.36 2.75 5.40 2.63 ardized IQ scores, sex and family size were assessed after Survival time (years, from birth)* 74.59 5.10 60.34 13.66 adjusting for the effects of all other predictors, including Missing values were deleted listwise in each of the variable estimates. the random effect of family. In the third set of analyses, the Survival time for those dead individuals represents the time until death; sur- mutually adjusted effects of standardized IQ scores and sex vival time for those alive represents the time until the censor date. *t-test con- were assessed in a fixed-effects model that was estimated ducted between those alive and dead, p< 0.001. by stratifying the analysis on family. This allows a different baseline hazard for each family, and allows the effects of sibling IQ distribution, thus lowering the mean IQ score shared family factors to be absorbed without having to be for siblings. Indeed, a comparison between 6-Day Sample estimated or even measured. Thus, the contribution of members and their nearest siblings by age demonstrated no standardized IQ scores to survival is assessed independ- significant difference in IQ scores (p¼ 0.368). Similarly, ently of all shared family factors. However, the stratified 6-Day Sample members exhibited significantly longer sur- model cannot estimate the effect of family size, as family vival times, both for those alive at the censor date and for size is constant within families. In all cases, missing values those who had died, than members of the Sibling Sample were deleted in a listwise manner. (Table 1). This reflects the fact that all individuals included Analyses were conducted in R (v3.3.1) using the in the Sibling Sample were younger than those in the 6-Day ‘psych’ package (v1.6.6). Cox regression analyses were Sample, and therefore did not have the opportunity to ac- conducted using the ‘survival’ (v2.40–1) and ‘coxme’ crue the same length of exposure period. (v2.2–5) packages. Table 2 shows the descriptive statistics of the whole sam- ple (6-Day Sample individuals and Sibling Sample individuals combined) according to mortality status. Those individuals Results who were still alive at the censor date demonstrated signifi- Table 1 shows the descriptive characteristics of the 6-Day cantly higher IQ scores than those who had died. Sample and Sibling Sample members. Members of the 6- Table 3 shows three sets of Cox regression analyses: Day Sample exhibited significantly higher IQ scores than first, the hazard ratios of standardized IQ score, sex and the members of the Sibling Sample, though this only equa- family size individually, with family entered as a random ted to a mean difference of 1.97 IQ points (Table 1). This effect in each univariable model; second, the mutually ad- difference was likely due to the combination of family size justed hazard ratios from a multivariable mixed-effects 22,27 effects and the way in which the comparison was model of standardized IQ score, sex and family size with weighted: as all younger siblings were retained in the com- family entered as a random effect; and third, the mutually parison, larger families contributed more IQ scores to the adjusted hazard ratios from a multivariable fixed-effects Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 1 93 Table 3. Hazard ratios (HRs) showing the mortality risk associated with a 1 standard-deviation increase in IQ score, with being female and with a one person increase in family size. Shown are the HRs including the random effect of family (in the univari- able models), adjusted for other predictors and including the random effect of family (in the multivariable models), and adjusted for other predictors and the stratifying effect of family (in the stratified multivariable model; N¼ 2228) Univariable Multivariable Stratified multivariable HR 95% CI p HR 95% CI p HR 95% CI p Standardized IQ score 0.76 0.68–0.84 <0.001 0.73 0.64–0.82 <0.001 0.79 0.68–0.92 0.002 Sex (Female) 0.57 0.41–0.72 <0.001 0.53 0.37–0.68 <0.001 0.47 0.38–0.58 <0.001 Family size 1.03 1.00–1.06 0.043 0.99 0.96–1.03 0.760 – – – in mortality risk for a one-member increase in family size. This association was attenuated once the effects of standar- dized IQ score and sex were adjusted for. The association between IQ scores and mortality re- mained relatively consistent across three further analyses (see Supplementary Material, available as Supplementary Data at IJE online). First, including single-child families (N¼ 2784 individuals) in the analyses demonstrated a 22% reduction in mortality risk with a standard-deviation advantage in IQ score after adjusting for shared family fac- tors. Second, including only those multiple-child families Figure 1. Kaplan–Meier survival curves for members of the full sample in which siblings were born within 7 years of their 6-Day based on IQ scores. Lines show survival probability for those with Sample probands (N¼ 1713 individuals) demonstrated a mean IQ scores and for those with IQ scores 1 standard deviation above or below the mean. Shaded areas represent 95% confidence intervals. 23% reduction in mortality risk with a standard-deviation advantage in IQ score after adjusting for shared family fac- tors. Third, repeating the survival analyses instead adjust- model of standardized IQ score and sex stratified by ing for an explicit proxy measure of SES, father’s family. occupational social class, demonstrated a 26% reduction Figure 1 shows the change in survival probability for in mortality risk with a standard-deviation advantage in those with a standard-deviation advantage or disadvantage IQ score. in IQ score. In the univariable model, after accounting for the correlation between family members, a standard- deviation advantage in IQ score was significantly associ- Discussion ated with a 24% reduction in mortality risk (Table 3). Once the effects of sex and family size were accounted for, The present study uses, for the first time to our knowledge a standard-deviation advantage in IQ score was associated in cognitive function-survival analyses, proband and sib- with a 27% decrease in mortality risk. In the stratified ling data. These valuable sibling data accompany the multivariable model, in which a different baseline mortal- 6-Day Sample of the Scottish Mental Survey 1947, and ity risk was specified for each family, the hazard ratio for were used to examine the association between early-life in- standardized IQ scores was somewhat attenuated, but telligence and mortality independently from family-related there remained a significant association between a early-life SES. The role of early-life SES in predicting mor- 14,16 standard-deviation advantage in IQ scores and a 21% re- tality has been well established, and previous studies duction in mortality risk. of early-life intelligence have tended to take and account Mortality risk was significantly lower for females in the for some measure of SES (e.g. parental occupation) in sub- univariable model. The hazard ratio associated with being sequent survival analyses to try to test for possible con- female was not attenuated when adjusted for standardized founding by family background. However, such attempts IQ score and family size, nor by stratifying the analyses by are limited by the lack of consensus on which measure of family. In the stratified multivariable model, being female SES to account for and by the residual confounding effect was associated with a 53% lower mortality risk. Family resulting from using select or few measures of SES. The size, on the other hand, was associated with a higher risk present study tackles these limitations by examining the as- of mortality in the univariable model, with a 3% increase sociation between intelligence and mortality within Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 94 International Journal of Epidemiology, 2018, Vol. 47, No. 1 families of the 6-Day Sample and their siblings, thus ac- life course outcomes within families. Notably, the present counting for shared family factors without the need to op- study replicated the significant association between IQ erationalize SES. After accounting for family-related SES in score and survival time previously reported in the 6-Day this fashion, there remained a significant and only slightly Sample, albeit with a much longer follow-up time (68 attenuated association between early-life intelligence and years). However, the present study is the first to demon- longevity. In the most conservative of the analyses, we strate that the association between higher IQ scores and observed a 21% decrease in mortality risk with each lower mortality risk extends to the younger siblings of the standard-deviation advantage in IQ score. This observa- 6-Day Sample. Even in these younger individuals, individ- tion is consistent with previous reports of SES-adjusted ual differences in cognitive ability appear to have import- hazard ratios in the Scottish Mental Survey 1947 (20% de- ant implications for longevity. crease in mortality risk) and in the 6-Day Sample more specifically (from 16% to 26% decrease in mortality 24 Limitations risk). The persistence of the intelligence–mortality associ- ation after taking into account early-life SES is also consist- By using family to represent SES, the present study only ac- ent with a recent meta-analysis of nine prospective cohort counts for SES that is related to family circumstances. studies, in which adjusting for various measures of SES Individual factors related to early-life SES, such as birth 19 13 15,36 did not attenuate the reduction in mortality risk associated weight, height and the person’s own education, with higher intelligence test scores (23% decrease in mor- may yet account for some of the association between intel- tality risk after adjustment). Also consistent with previous ligence and mortality. Previous work has suggested that work was the observed associations between male sex and adult SES may play a larger role in determining mortality increased mortality risk and between larger family size risk than childhood SES, and that adjusting for adult SES and increased mortality risk. and education provides the largest attenuation of the The somewhat attenuating effect of adjusting for intelligence–mortality association. However, education family-related SES, although small, is broadly consistent and adult SES are phenotypically and genetically correlated with previous work using discrete proxy measures of early- with, and at least partly confounded by, childhood intelli- life SES. However, the fact that a substantial IQ- gence such that lower early-life cognitive ability may result mortality association persists beyond such adjustment is in lower educational attainment, which leads to employ- 37,38 interesting, particularly given that the present study ac- ment in riskier and generally lower-paid occupations. counts for early-life SES in a different and more compre- Although more conservative than adjusting for discrete hensive way. Adjusting for within-family variance is proxy measures of SES, by adjusting for family-related fac- notably more conservative than simply accounting for an tors, the present study assumes that the important aspects explicit measure of SES, as it likely captures shared factors of the early-life environment are shared equally by siblings. not directly related to SES such as genetic factors, environ- In some cases, this assumption may not hold, e.g. where mental health (air pollution, etc.), childhood diet and ex- siblings live in different households (e.g. with a divorced posure to passive smoking. Notably, adjusting for shared parent or after migration) or where siblings live in the family factors led to a slightly larger attenuation of the same household at very different points in time (e.g. with intelligence–mortality association than adjusting for an ex- much later siblings). Indeed, the birth distance between plicit proxy measure of SES (see Supplementary Material, 6-Day Sample members and their siblings was as much as available as Supplementary Data at IJE online). However, 22 years in some exceptional cases. However, when the the advantage of the family-based approach adopted in the analyses were repeated using only siblings born within present study is that it avoids any measurement error or re- 7 years of their 6-Day Sample proband, the pattern of re- sidual confounding associated with using proxy measures sults was very similar. The effect of family, even in these such as parental occupation and income. Where previous individuals most likely (temporally) to share a family envir- studies have used proxy measures, residual confounding onment, only slightly attenuated the association between may erroneously underestimate the attenuating effect of higher intelligence and reduced mortality risk. early-life SES on the intelligence–mortality association. The present study demonstrates that the association be- As well as addressing the role of early-life SES in the tween early-life cognitive ability and mortality risk remains intelligence–mortality association, the present study is the even after the confounding effect of early-life SES is ac- first to describe mortality risk in the siblings of the 6-Day counted for. Notably, the present study takes the conserva- Sample. The 6-Day Sample was formed to be representa- tive approach of treating all variance shared within families tive of the Scottish nation born in 1936, and their younger as relevant to SES. Examining the role of intelligence within siblings were followed up with the intention of examining families allows adjustment of the intelligence–mortality Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 1 95 9. Batty GD, Wennerstad KM, Smith GD et al. IQ in early adult- association without the need to operationalize or measure hood and mortality by middle age: cohort study of 1 million SES, and so avoids the potential residual confounding asso- Swedish men. Epidemiology 2009;20:100–9. ciated with adjusting for proxy measures of early-life SES. 10. Calvin CM, Deary IJ, Fenton C et al. Intelligence in youth and The present study demonstrates that, whereas they are un- all-cause-mortality: systematic review with meta-analysis. Int J doubtedly important for longevity, early-life family condi- Epidemiol 2011;40:626–44. tions are not sufficient to explain the intelligence–mortality 11. Lawlor DA, Batty GD, Morton SM et al. Early life predictors of association frequently reported in cognitive epidemiology childhood intelligence: evidence from the Aberdeen children of research. Future work should focus on elucidating the mech- the 1950s study. J Epidemiol Commun H 2005;59:656–63. 12. Bates TC, Lewis GJ, Weiss A. Childhood socioeconomic status anisms by which early-life cognitive ability is associated amplifies genetic effects on adult intelligence. Psychol Sci 2013; with longer life. 24:2111–16. 13. Elo IT, Preston SH. Effects of early-life conditions on adult mor- Supplementary Data tality: a review. Popul Index 1992;58:186–212. 14. Galobardes B, Lynch JW, Davey Smith G. Childhood socioeco- Supplementary data are available at IJE online. nomic circumstances and cause-specific mortality in adulthood: sys- tematic review and interpretation. Epidemiol Rev 2004;26:7–21. Acknowledgements 15. Batty GD, Der G, Macintyre S, Deary, IJ. Does IQ explain socio- This work was supported by a UK cross-council Lifelong Health and economic inequalities in health? Evidence from a population Wellbeing Initiative (Grant Number: MRCG1001401), for which I.J.D. based cohort study in the west of Scotland. BMJ 2006;332: is the principal investigator. M.H.I., I.C. and I.J.D. are members of the 580–3. University of Edinburgh Centre for Cognitive Ageing and Cognitive 16. Galobardes B, Davey Smith G, Lynch JW. Systematic review of Epidemiology, part of the cross-council Lifelong Health and Wellbeing the influence of childhood socioeconomic circumstances on risk Initiative (Grant Number: MR/K026992/1). G.D. is a member of the for cardiovascular disease in adulthood. Ann Epidemiol 2006; MRC/CSO Social and Public Health Sciences Unit (Grant Number: 16:91–104. SPHSU13). The authors would like to thank the Scottish Council for 17. Batty GD, Deary IJ. Early life intelligence and adult health. BMJ Research in Education for use of the Scottish Mental Survey data and the National Records of Scotland for conducting the mortality record 2004;329:585–6. linkages. 18. Gottfredson LS. Intelligence: is it the epidemiologists’ elusive ‘fundamental cause’ of social class inequalities in health? J Pers Conflict of interest: The authors have no conflicts of interest to Soc Psychol 2004;86:174–99. declare. 19. Osler M, Anderson AN, Due P, Lund R, Damsgaard MT, Holstein BE. Socioeconomic position in early life, birth weight, References childhood cognitive function, and adult mortality: a longitudinal study of Danish men born in 1953. J Epidemiol Commun H 1. O’Toole BI, Adena MA, Jones MP. Risk factors for mortality in Australian Vietnam-era national servicemen: a case–control 2003;57:681–6. 20. Galobardes B, Lynch J, Davey Smith G. Measuring socioeco- study. Community Health St 1988;12:408–17. 2. Korten AE, Jorm AF, Jiao Z et al. Health, cognitive, and psycho- nomic position in health research. Brit Med Bull 2007;81–82: 21–37. social factors as predictors of mortality in an elderly community sample. J Epidemiol Commun H 1999;53:83–8. 21. Murray C. IQ and income inequality in a sample of sibling pairs from advantaged family backgrounds. Am Econ Rev 2002;92: 3. Pavlik VN, de Moraes SA, Szklo M, Knopman DS, Mosley TH Jr, Hyman DJ. Relation between cognitive function and mortal- 339–43. 22. The Scottish Council for Research in Education. The Trend of ity in middle-aged adults: the Atherosclerosis Risk in Communities Study. Am J Epidemiol 2003;157:327–34. Scottish Intelligence. London: University of London Press Ltd, 1949. 23. Maxwell J. Sixteen Years On: A Follow-Up of the 1947 Scottish 4. Deary IJ, Der G. Reaction time explains IQ’s association with death. Psychol Sci 2005;16:64–9. Survey. London: University of London Press, 1969. 24. Deary IJ, Batty GD, Pattie A, Gale CR. More intelligence, more 5. Batterham PJ, Christensen H, Mackinnon AJ. Fluid intelligence is independently associated with all-cause mortality over 17 dependable children live longer: a 55-year longitudinal study of a representative sample of the Scottish nation. Psychol Sci 2008; years in an elderly community sample: an investigation of poten- tial mechanisms. Intelligence 2009;37:551–60. 19:874–80. 25. Deary IJ, Whalley LJ, Starr JM. A Lifetime of Intelligence: 6. Whalley LJ, Deary IJ. Longitudinal cohort study of childhood IQ and survival up to age 76. BMJ 2001;322:819–22. Follow-Up Studies of the Scottish Mental Surveys of 1932 and 1947. Washington, DC: American Psychological Association, 7. Lager A, Bremberg S, Va ˚ gero ¨ D. The association of early IQ and education with mortality: 65 year longitudinal study in Malmo, 2009. 26. Cukic I, Brett CE, Calvin CM, Batty GD, Deary IJ. Childhood Sweden. BMJ 2009;339:b5282. 8. Leon DA, Lawlor DA, Clark H, Batty GD, Macintyre S. The as- IQ and survival to 79: follow-up of 94% of the Scottish Mental Survey 1947. Intelligence 2017;63:45–50. sociation of childhood intelligence with mortality risk from ado- lescence to middle age: findings from the Aberdeen Children of 27. MacPherson JS. Eleven-Year-Olds Grow Up. London: University of London Press, 1958. the 1950s cohort study. Intelligence 2009;37:520–8. Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 96 International Journal of Epidemiology, 2018, Vol. 47, No. 1 28. Terman LM, Merrill MA. Measuring Intelligence: A Guide to 34. Therneau TM. coxme: Mixed Effects Cox Models. R package the Administration of the New Revised Stanford-Binet Tests of version 2.2–5, 2015. Intelligence. Oxford: Houghton Mifflin, 1937. 35. Deary IJ. Follow-up studies of the Scottish Mental Surveys of 1932 29. Calvin CM, Batty GD, Brett CE, Deary IJ. Childhood club par- and 1947. In: Peel RA, Zeki, M (eds). Human Ability: Genetic and ticipation and all-cause mortality in adulthood: a 65-year Environmental Influences. London: The Galton Institute, 2006. follow-up study of a population-representative sample in 36. Davey Smith G, Hart C, Hole D et al. Education and occupa- Scotland. Psychosom Med 2015;77:712–20. tional social class: which is the more important indicator of mor- 30. Maxwell J. The Level and Trend of National Intelligence. tality risk? J Epidemiol Commun H 1998;52:153–60. London: University of London Press, 1961. 37. Davey Smith G. Health Inequalities: Lifecourse Approaches. 31. R Core Team. R: A Language and Environment for Statistical Bristol: Policy Press, 2003. Computing. Vienna: R Foundation for Statistical Computing, 2016. 38. Deary IJ, Johnson W. Intelligence and education: causal percep- 32. Revelle W. psych: Procedures for Personality and Psychological tions drive analytic processes and therefore conclusions. Research. Evanston, IL: Northwestern University, 2016. International Journal of Epidemiology 2010;39:1362–9. 33. Therneau TM. survival: A Package for Survival Analysis in r.R package version 2.40–1, 2015. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Journal of Epidemiology Oxford University Press

Intelligence and all-cause mortality in the 6-Day Sample of the Scottish Mental Survey 1947 and their siblings: testing the contribution of family background

Download PDF
8 pages

Loading next page...
1
 
/lp/ou_press/intelligence-and-all-cause-mortality-in-the-6-day-sample-of-the-R4sXIXq8HU

References (67)

Publisher
Oxford University Press
Copyright
Copyright © 2022 International Epidemiological Association
ISSN
0300-5771
eISSN
1464-3685
DOI
10.1093/ije/dyx168
Publisher site
See Article on Publisher Site

Abstract

Background: Higher early-life intelligence is associated with a reduced risk of mortality in adulthood, though this association is apparently hardly attenuated when accounting for early-life socio-economic status (SES). However, the use of proxy measures of SES means that residual confounding may underestimate this attenuation. In the present study, the potential confounding effect of early-life SES was instead accounted for by examining the intelligence–mortality association within families. Methods: The association between early-life intelligence and mortality in adulthood was assessed in 727 members of the 6-Day Sample of the Scottish Mental Survey 1947 and, for the first time, 1580 of their younger siblings. These individuals were born between 1936 and 1958, and were followed up into later life, with deaths recorded up to 2015. Cox regression was used to estimate the relative risk of mortality associated with higher IQ scores after adjusting for shared family factors. Results: A standard-deviation advantage in IQ score was associated with a significantly reduced mortality risk [hazard ratio¼ 0.76, p< 0.001, 95% confidence interval (CI) (0.68– 0.84)]. This reduction in hazard was only slightly attenuated by adjusting for sex and shared family factors [hazard ratio¼ 0.79, p¼ 0.002, 95% CI (0.68–0.92)]. Conclusions: Although somewhat conservative, adjusting for all variance shared by a family avoids any potential residual confounding of the intelligence–mortality associa- tion arising from the use of proxy measures of early-life SES. The present study demon- strates that the longevity associated with higher early-life intelligence cannot be ex- plained by early-life SES or within-family factors. V The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association 89 This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 90 International Journal of Epidemiology, 2018, Vol. 47, No. 1 Key words: mortality, intelligence, socio-economic status Key Messages A standard-deviation increase in IQ score was associated with a 24% decrease in mortality risk. After adjusting for all shared family factors, a standard-deviation increase in IQ score was associated with a 21% de- crease in mortality risk. Early-life socio-economic circumstances are not sufficient to explain the intelligence–mortality association. Introduction confounding, resulting from measurement error in early- life SES, may therefore be driving previously observed as- Identifying predictors of mortality and chronic disease has sociations to some extent, even after adjusting for apparent been the goal of many epidemiological studies from across early-life SES. various countries and populations. In cognitive epidemi- In a study of the link between intelligence in youth and ology, intelligence (general cognitive function) as measured later income inequality, Murray attempted to tackle this using psychometric tests in middle- and older-aged popula- SES-assessment problem by examining associations within tions has emerged as a predictor of longevity, with higher families. Having matched individuals to their nearest sib- intelligence test scores associated with reduced mortality 1–5 lings, Murray then examined whether within-sibling-pair risk. More recently, pre-morbid measures of cognitive intelligence differences could predict income differences ability in cohorts of children and young people have been within pairs. Examining the contribution of intelligence shown to be related to mortality risk up to seven decades 6–10 within families circumvented the need to obtain a range of later. SES measures, as it removed the variance accounted for by A key consideration in interpreting these results is the shared family environment (e.g. parental income, occu- understanding the role of early-life socio-economic status pation and education). Examining outcomes within fami- (SES). Early-life SES may be a key confounder in the lies therefore allows researchers to account for early-life cognition–mortality relation owing to its association with 11 12 SES without the need to operationalize and measure it. both childhood and adult intelligence, and its link to 7,13–16 The present study adopts a within-family method in later-life mortality and health. However, the causal order to examine the association between early-life intelli- relationship between early-life SES and intelligence, and gence and mortality while accounting for early-life SES. thus the nature of the resulting association with later-life 17,18 Families were established by linking members of the 6-Day mortality, is unclear. Researchers have typically at- Sample of the Scottish Mental Survey 1947 with their tempted to account for early-life SES by including it as a younger siblings. The 6-Day Sample is a group of individ- predictor in multivariable models, and many have uals (N¼ 1208) representative of the whole Scottish popula- observed that doing so does not attenuate the contribution tion born in 1936 and whose cognitive ability was tested at of early-life intelligence to mortality risk. 22,23 age 11 years old. Previous work with this sample has However, attempts to account for any contribution of shown early-life intelligence to be a significant predictor of early-life SES to the intelligence–mortality association are mortality from up to 67 years old, even after early-life SES limited by the way in which SES is operationalized. Given (interviewer-rated parental intelligence and personality, the breadth and complexity of SES, it is difficult to fully household cleanliness, etc.) had been accounted for. characterize early-life social circumstances. Researchers Importantly, the younger siblings of these individuals were commonly include proxy measures for early-life SES, such 19 7 tested on the same IQ-type test when they reached 11 years as parental occupation, parental income or participant’s old, and were recently linked to records of mortality. These own education. This has also resulted in a lack of consist- siblings shared similar early-life circumstances and upbring- ency between cohort studies in terms of the SES measure ings to their 6-Day Sample probands (e.g. parental SES and used. Given these shortcomings, it is likely that not all of household size), but may differ in their cognitive ability and the variance associated with early-life SES has been fully longevity. By examining the intelligence–mortality associ- captured or accounted for in studies examining the ation within families, each consisting of a 6-Day Sample association between intelligence and mortality. Residual Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 1 91 member and their siblings, it is possible, more comprehen- removed due to missing data (N¼ 17) resulted in a total sively than previously, to partition out the potential con- sample of 2307 individuals (6-Day Sample: N¼ 727; founding effect of early-life SES. Sibling Sample: N¼ 1580) from 728 families (mean family size¼ 3.13, SD¼ 1.57). Methods Study sample Assessments On 4 June 1947, almost all children born in 1936 and at- Intelligence tending school in Scotland sat the Moray House Test No. For both the 6-Day Sample and their siblings, intelligence 12 test of intelligence. This cohort of 70 805 individuals, was measured using the Terman-Merill test, Form L —an the Scottish Mental Survey 1947, comprised 88% of the adapted version of the Binet-Simon test of intelligence. Scottish population born in 1936 and 94% of the avail- This test included 129 items of both verbal and non-verbal able school population at the time. A subsample of this reasoning. Raw correct scores were converted into standar- 1936 birth cohort, the 6-Day Sample (n¼ 1208; 618 fe- dized IQ-type scores (M¼ 100, SD¼ 15). For the 6-Day males), was created by selecting individuals born in Sample, this test was administered in 1947, following their Scotland on the first day of every even-numbered month in completion of the Scottish Mental Survey 1947. Siblings, 1936, whether or not they completed the Moray House on the other hand, completed the test at age 11 years, be- Test of intelligence. The mean intelligence and geograph- tween 1948 and 1969. ical distribution of the 6-Day Sample has been shown to be similar to the full Scottish Mental Survey 1947 cohort. Mortality Members of the 6-Day Sample were given a second, indi- Vital status and date of death were obtained for each of vidually administered test of intelligence in 1947—the 22,28 the 6-Day Sample members and their siblings. This was Terman-Merrill revision of the Binet Test —and were achieved by linking both the 6-Day Sample and their sib- subsequently resurveyed about every year up to the age of lings to their respective administrative records held by the 27 years to collect information on education, family life, National Records of Scotland (NRS). This linkage was home environment, leisure activities, health and early- 23,25 approved by the Scotland-A Research Ethics Committee adulthood occupation. (Ref: 12/SS/0024), the National Services Scotland NHS Younger siblings (n¼ 1655; 798 females) of the 6-Day Privacy Advisory Committee and the Confidentiality Sample members were tested for intelligence using the Advisory Group of the Health Research Authority. same Terman-Merrill Binet Test as they approached age 11 Approval covered linkage without consent, up to years, and were also resurveyed into early adulthood. Of November 2015, under section 251 if the NHS Act 2006. the whole 6-Day Sample, 748 had siblings included in the The NRS used automated and manual tracing methods to follow-up surveys. The Sibling Sample, born between 1937 link identifiable information (date of birth, surname, fore- and 1958, ranged from first siblings (n¼ 748) to tenth sib- name and National Health Service number) for each indi- lings (n¼ 3), and siblings were on average 6.13 years vidual with their respective National Health Service younger than their 6-Day Sample probands [standard devi- Central Register records. ation (SD)¼ 3.96, Min.¼ 0.39 years, Max.¼ 22.36 years]. Individuals were censored at the end of mortality sur- Fourteen of the Sibling Sample were twins, though none veillance (30 November 2015 for most individuals). was twinned with their 6-Day Sample proband. Survival time was calculated as the number of days be- Combining the two cohorts resulted in a sample of tween the date of birth and either the date of death or cen- 2863 individuals. Of these, 79 individuals (6-Day Sample: soring date as appropriate. Individuals who had emigrated N¼ 4, Sibling Sample: N¼ 75) for whom follow-up data after completing the intelligence test but before the end of were not available were removed from any further ana- the surveillance period were retained in the sample, but lysis. As part of previous work, formal comparisons have were censored at the start of the month in which they shown that individuals lost to follow-up demonstrate embarked. higher childhood IQ scores (a difference of 4.7 IQ points), higher levels of schooling and higher SES relative to those retained in the 6-Day Sample. In order to ensure that Statistical analyses family-related factors could be shared, the present study focused on multiple-child families. Excluding those 6-Day Survival analyses were conducted using Cox proportional Sample members from single-child families (N¼ 460) or hazards regression. Hazard ratios were calculated for each those 6-Day Sample members whose siblings were predictor included in the model to indicate the Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 92 International Journal of Epidemiology, 2018, Vol. 47, No. 1 Table 1. Descriptive characteristics of the 6-Day Sample members and their younger siblings, and comparisons between groups 6-Day Sample (N¼ 727) Sibling Sample (N¼ 1580) Mean SD Mean SD Sex (N male/female) 363/364 819/761 IQ score* 100.28 19.05 98.31 17.60 Mortality status (N alive/dead) 437/290 1144/436 Time to death (years, from birth)** 63.99 11.65 57.93 14.35 Time to censor (years, from birth)** 79.43 0.28 72.93 4.12 Missing values were deleted listwise in each of the variable estimates. Time to death is calculated only for those who have died before the censor date; time to censor is calculated only for those still alive at the censor date. *t-test conducted between the 6-Day Sample and Sibling Sample, p¼ 0.019; **p< 0.001. proportionate change in mortality risk for a unit change in Table 2. Descriptive characteristics of the whole sample the predictor. In the first set of analyses, individual univari- (N¼ 2307) of 6-Day Sample members and their younger sib- able regression models were created to predict survival lings according to mortality status using either standardized (z-transformed) IQ scores, sex or Alive (N¼ 1581) Dead (N¼ 726) family size. Each model additionally included a random ef- fect of family to account for the fact that observations Mean SD Mean SD within each family are correlated. All of the predictors con- Sex (N male/female) 735/846 447/279 formed to the proportional hazards assumption (all IQ score* 100.73 18.20 95.09 17.28 ps> 0.28). In the second set of analyses, the effect of stand- Family size (people) 5.36 2.75 5.40 2.63 ardized IQ scores, sex and family size were assessed after Survival time (years, from birth)* 74.59 5.10 60.34 13.66 adjusting for the effects of all other predictors, including Missing values were deleted listwise in each of the variable estimates. the random effect of family. In the third set of analyses, the Survival time for those dead individuals represents the time until death; sur- mutually adjusted effects of standardized IQ scores and sex vival time for those alive represents the time until the censor date. *t-test con- were assessed in a fixed-effects model that was estimated ducted between those alive and dead, p< 0.001. by stratifying the analysis on family. This allows a different baseline hazard for each family, and allows the effects of sibling IQ distribution, thus lowering the mean IQ score shared family factors to be absorbed without having to be for siblings. Indeed, a comparison between 6-Day Sample estimated or even measured. Thus, the contribution of members and their nearest siblings by age demonstrated no standardized IQ scores to survival is assessed independ- significant difference in IQ scores (p¼ 0.368). Similarly, ently of all shared family factors. However, the stratified 6-Day Sample members exhibited significantly longer sur- model cannot estimate the effect of family size, as family vival times, both for those alive at the censor date and for size is constant within families. In all cases, missing values those who had died, than members of the Sibling Sample were deleted in a listwise manner. (Table 1). This reflects the fact that all individuals included Analyses were conducted in R (v3.3.1) using the in the Sibling Sample were younger than those in the 6-Day ‘psych’ package (v1.6.6). Cox regression analyses were Sample, and therefore did not have the opportunity to ac- conducted using the ‘survival’ (v2.40–1) and ‘coxme’ crue the same length of exposure period. (v2.2–5) packages. Table 2 shows the descriptive statistics of the whole sam- ple (6-Day Sample individuals and Sibling Sample individuals combined) according to mortality status. Those individuals Results who were still alive at the censor date demonstrated signifi- Table 1 shows the descriptive characteristics of the 6-Day cantly higher IQ scores than those who had died. Sample and Sibling Sample members. Members of the 6- Table 3 shows three sets of Cox regression analyses: Day Sample exhibited significantly higher IQ scores than first, the hazard ratios of standardized IQ score, sex and the members of the Sibling Sample, though this only equa- family size individually, with family entered as a random ted to a mean difference of 1.97 IQ points (Table 1). This effect in each univariable model; second, the mutually ad- difference was likely due to the combination of family size justed hazard ratios from a multivariable mixed-effects 22,27 effects and the way in which the comparison was model of standardized IQ score, sex and family size with weighted: as all younger siblings were retained in the com- family entered as a random effect; and third, the mutually parison, larger families contributed more IQ scores to the adjusted hazard ratios from a multivariable fixed-effects Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 1 93 Table 3. Hazard ratios (HRs) showing the mortality risk associated with a 1 standard-deviation increase in IQ score, with being female and with a one person increase in family size. Shown are the HRs including the random effect of family (in the univari- able models), adjusted for other predictors and including the random effect of family (in the multivariable models), and adjusted for other predictors and the stratifying effect of family (in the stratified multivariable model; N¼ 2228) Univariable Multivariable Stratified multivariable HR 95% CI p HR 95% CI p HR 95% CI p Standardized IQ score 0.76 0.68–0.84 <0.001 0.73 0.64–0.82 <0.001 0.79 0.68–0.92 0.002 Sex (Female) 0.57 0.41–0.72 <0.001 0.53 0.37–0.68 <0.001 0.47 0.38–0.58 <0.001 Family size 1.03 1.00–1.06 0.043 0.99 0.96–1.03 0.760 – – – in mortality risk for a one-member increase in family size. This association was attenuated once the effects of standar- dized IQ score and sex were adjusted for. The association between IQ scores and mortality re- mained relatively consistent across three further analyses (see Supplementary Material, available as Supplementary Data at IJE online). First, including single-child families (N¼ 2784 individuals) in the analyses demonstrated a 22% reduction in mortality risk with a standard-deviation advantage in IQ score after adjusting for shared family fac- tors. Second, including only those multiple-child families Figure 1. Kaplan–Meier survival curves for members of the full sample in which siblings were born within 7 years of their 6-Day based on IQ scores. Lines show survival probability for those with Sample probands (N¼ 1713 individuals) demonstrated a mean IQ scores and for those with IQ scores 1 standard deviation above or below the mean. Shaded areas represent 95% confidence intervals. 23% reduction in mortality risk with a standard-deviation advantage in IQ score after adjusting for shared family fac- tors. Third, repeating the survival analyses instead adjust- model of standardized IQ score and sex stratified by ing for an explicit proxy measure of SES, father’s family. occupational social class, demonstrated a 26% reduction Figure 1 shows the change in survival probability for in mortality risk with a standard-deviation advantage in those with a standard-deviation advantage or disadvantage IQ score. in IQ score. In the univariable model, after accounting for the correlation between family members, a standard- deviation advantage in IQ score was significantly associ- Discussion ated with a 24% reduction in mortality risk (Table 3). Once the effects of sex and family size were accounted for, The present study uses, for the first time to our knowledge a standard-deviation advantage in IQ score was associated in cognitive function-survival analyses, proband and sib- with a 27% decrease in mortality risk. In the stratified ling data. These valuable sibling data accompany the multivariable model, in which a different baseline mortal- 6-Day Sample of the Scottish Mental Survey 1947, and ity risk was specified for each family, the hazard ratio for were used to examine the association between early-life in- standardized IQ scores was somewhat attenuated, but telligence and mortality independently from family-related there remained a significant association between a early-life SES. The role of early-life SES in predicting mor- 14,16 standard-deviation advantage in IQ scores and a 21% re- tality has been well established, and previous studies duction in mortality risk. of early-life intelligence have tended to take and account Mortality risk was significantly lower for females in the for some measure of SES (e.g. parental occupation) in sub- univariable model. The hazard ratio associated with being sequent survival analyses to try to test for possible con- female was not attenuated when adjusted for standardized founding by family background. However, such attempts IQ score and family size, nor by stratifying the analyses by are limited by the lack of consensus on which measure of family. In the stratified multivariable model, being female SES to account for and by the residual confounding effect was associated with a 53% lower mortality risk. Family resulting from using select or few measures of SES. The size, on the other hand, was associated with a higher risk present study tackles these limitations by examining the as- of mortality in the univariable model, with a 3% increase sociation between intelligence and mortality within Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 94 International Journal of Epidemiology, 2018, Vol. 47, No. 1 families of the 6-Day Sample and their siblings, thus ac- life course outcomes within families. Notably, the present counting for shared family factors without the need to op- study replicated the significant association between IQ erationalize SES. After accounting for family-related SES in score and survival time previously reported in the 6-Day this fashion, there remained a significant and only slightly Sample, albeit with a much longer follow-up time (68 attenuated association between early-life intelligence and years). However, the present study is the first to demon- longevity. In the most conservative of the analyses, we strate that the association between higher IQ scores and observed a 21% decrease in mortality risk with each lower mortality risk extends to the younger siblings of the standard-deviation advantage in IQ score. This observa- 6-Day Sample. Even in these younger individuals, individ- tion is consistent with previous reports of SES-adjusted ual differences in cognitive ability appear to have import- hazard ratios in the Scottish Mental Survey 1947 (20% de- ant implications for longevity. crease in mortality risk) and in the 6-Day Sample more specifically (from 16% to 26% decrease in mortality 24 Limitations risk). The persistence of the intelligence–mortality associ- ation after taking into account early-life SES is also consist- By using family to represent SES, the present study only ac- ent with a recent meta-analysis of nine prospective cohort counts for SES that is related to family circumstances. studies, in which adjusting for various measures of SES Individual factors related to early-life SES, such as birth 19 13 15,36 did not attenuate the reduction in mortality risk associated weight, height and the person’s own education, with higher intelligence test scores (23% decrease in mor- may yet account for some of the association between intel- tality risk after adjustment). Also consistent with previous ligence and mortality. Previous work has suggested that work was the observed associations between male sex and adult SES may play a larger role in determining mortality increased mortality risk and between larger family size risk than childhood SES, and that adjusting for adult SES and increased mortality risk. and education provides the largest attenuation of the The somewhat attenuating effect of adjusting for intelligence–mortality association. However, education family-related SES, although small, is broadly consistent and adult SES are phenotypically and genetically correlated with previous work using discrete proxy measures of early- with, and at least partly confounded by, childhood intelli- life SES. However, the fact that a substantial IQ- gence such that lower early-life cognitive ability may result mortality association persists beyond such adjustment is in lower educational attainment, which leads to employ- 37,38 interesting, particularly given that the present study ac- ment in riskier and generally lower-paid occupations. counts for early-life SES in a different and more compre- Although more conservative than adjusting for discrete hensive way. Adjusting for within-family variance is proxy measures of SES, by adjusting for family-related fac- notably more conservative than simply accounting for an tors, the present study assumes that the important aspects explicit measure of SES, as it likely captures shared factors of the early-life environment are shared equally by siblings. not directly related to SES such as genetic factors, environ- In some cases, this assumption may not hold, e.g. where mental health (air pollution, etc.), childhood diet and ex- siblings live in different households (e.g. with a divorced posure to passive smoking. Notably, adjusting for shared parent or after migration) or where siblings live in the family factors led to a slightly larger attenuation of the same household at very different points in time (e.g. with intelligence–mortality association than adjusting for an ex- much later siblings). Indeed, the birth distance between plicit proxy measure of SES (see Supplementary Material, 6-Day Sample members and their siblings was as much as available as Supplementary Data at IJE online). However, 22 years in some exceptional cases. However, when the the advantage of the family-based approach adopted in the analyses were repeated using only siblings born within present study is that it avoids any measurement error or re- 7 years of their 6-Day Sample proband, the pattern of re- sidual confounding associated with using proxy measures sults was very similar. The effect of family, even in these such as parental occupation and income. Where previous individuals most likely (temporally) to share a family envir- studies have used proxy measures, residual confounding onment, only slightly attenuated the association between may erroneously underestimate the attenuating effect of higher intelligence and reduced mortality risk. early-life SES on the intelligence–mortality association. The present study demonstrates that the association be- As well as addressing the role of early-life SES in the tween early-life cognitive ability and mortality risk remains intelligence–mortality association, the present study is the even after the confounding effect of early-life SES is ac- first to describe mortality risk in the siblings of the 6-Day counted for. Notably, the present study takes the conserva- Sample. The 6-Day Sample was formed to be representa- tive approach of treating all variance shared within families tive of the Scottish nation born in 1936, and their younger as relevant to SES. Examining the role of intelligence within siblings were followed up with the intention of examining families allows adjustment of the intelligence–mortality Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 1 95 9. Batty GD, Wennerstad KM, Smith GD et al. IQ in early adult- association without the need to operationalize or measure hood and mortality by middle age: cohort study of 1 million SES, and so avoids the potential residual confounding asso- Swedish men. Epidemiology 2009;20:100–9. ciated with adjusting for proxy measures of early-life SES. 10. Calvin CM, Deary IJ, Fenton C et al. Intelligence in youth and The present study demonstrates that, whereas they are un- all-cause-mortality: systematic review with meta-analysis. Int J doubtedly important for longevity, early-life family condi- Epidemiol 2011;40:626–44. tions are not sufficient to explain the intelligence–mortality 11. Lawlor DA, Batty GD, Morton SM et al. Early life predictors of association frequently reported in cognitive epidemiology childhood intelligence: evidence from the Aberdeen children of research. Future work should focus on elucidating the mech- the 1950s study. J Epidemiol Commun H 2005;59:656–63. 12. Bates TC, Lewis GJ, Weiss A. Childhood socioeconomic status anisms by which early-life cognitive ability is associated amplifies genetic effects on adult intelligence. Psychol Sci 2013; with longer life. 24:2111–16. 13. Elo IT, Preston SH. Effects of early-life conditions on adult mor- Supplementary Data tality: a review. Popul Index 1992;58:186–212. 14. Galobardes B, Lynch JW, Davey Smith G. Childhood socioeco- Supplementary data are available at IJE online. nomic circumstances and cause-specific mortality in adulthood: sys- tematic review and interpretation. Epidemiol Rev 2004;26:7–21. Acknowledgements 15. Batty GD, Der G, Macintyre S, Deary, IJ. Does IQ explain socio- This work was supported by a UK cross-council Lifelong Health and economic inequalities in health? Evidence from a population Wellbeing Initiative (Grant Number: MRCG1001401), for which I.J.D. based cohort study in the west of Scotland. BMJ 2006;332: is the principal investigator. M.H.I., I.C. and I.J.D. are members of the 580–3. University of Edinburgh Centre for Cognitive Ageing and Cognitive 16. Galobardes B, Davey Smith G, Lynch JW. Systematic review of Epidemiology, part of the cross-council Lifelong Health and Wellbeing the influence of childhood socioeconomic circumstances on risk Initiative (Grant Number: MR/K026992/1). G.D. is a member of the for cardiovascular disease in adulthood. Ann Epidemiol 2006; MRC/CSO Social and Public Health Sciences Unit (Grant Number: 16:91–104. SPHSU13). The authors would like to thank the Scottish Council for 17. Batty GD, Deary IJ. Early life intelligence and adult health. BMJ Research in Education for use of the Scottish Mental Survey data and the National Records of Scotland for conducting the mortality record 2004;329:585–6. linkages. 18. Gottfredson LS. Intelligence: is it the epidemiologists’ elusive ‘fundamental cause’ of social class inequalities in health? J Pers Conflict of interest: The authors have no conflicts of interest to Soc Psychol 2004;86:174–99. declare. 19. Osler M, Anderson AN, Due P, Lund R, Damsgaard MT, Holstein BE. Socioeconomic position in early life, birth weight, References childhood cognitive function, and adult mortality: a longitudinal study of Danish men born in 1953. J Epidemiol Commun H 1. O’Toole BI, Adena MA, Jones MP. Risk factors for mortality in Australian Vietnam-era national servicemen: a case–control 2003;57:681–6. 20. Galobardes B, Lynch J, Davey Smith G. Measuring socioeco- study. Community Health St 1988;12:408–17. 2. Korten AE, Jorm AF, Jiao Z et al. Health, cognitive, and psycho- nomic position in health research. Brit Med Bull 2007;81–82: 21–37. social factors as predictors of mortality in an elderly community sample. J Epidemiol Commun H 1999;53:83–8. 21. Murray C. IQ and income inequality in a sample of sibling pairs from advantaged family backgrounds. Am Econ Rev 2002;92: 3. Pavlik VN, de Moraes SA, Szklo M, Knopman DS, Mosley TH Jr, Hyman DJ. Relation between cognitive function and mortal- 339–43. 22. The Scottish Council for Research in Education. The Trend of ity in middle-aged adults: the Atherosclerosis Risk in Communities Study. Am J Epidemiol 2003;157:327–34. Scottish Intelligence. London: University of London Press Ltd, 1949. 23. Maxwell J. Sixteen Years On: A Follow-Up of the 1947 Scottish 4. Deary IJ, Der G. Reaction time explains IQ’s association with death. Psychol Sci 2005;16:64–9. Survey. London: University of London Press, 1969. 24. Deary IJ, Batty GD, Pattie A, Gale CR. More intelligence, more 5. Batterham PJ, Christensen H, Mackinnon AJ. Fluid intelligence is independently associated with all-cause mortality over 17 dependable children live longer: a 55-year longitudinal study of a representative sample of the Scottish nation. Psychol Sci 2008; years in an elderly community sample: an investigation of poten- tial mechanisms. Intelligence 2009;37:551–60. 19:874–80. 25. Deary IJ, Whalley LJ, Starr JM. A Lifetime of Intelligence: 6. Whalley LJ, Deary IJ. Longitudinal cohort study of childhood IQ and survival up to age 76. BMJ 2001;322:819–22. Follow-Up Studies of the Scottish Mental Surveys of 1932 and 1947. Washington, DC: American Psychological Association, 7. Lager A, Bremberg S, Va ˚ gero ¨ D. The association of early IQ and education with mortality: 65 year longitudinal study in Malmo, 2009. 26. Cukic I, Brett CE, Calvin CM, Batty GD, Deary IJ. Childhood Sweden. BMJ 2009;339:b5282. 8. Leon DA, Lawlor DA, Clark H, Batty GD, Macintyre S. The as- IQ and survival to 79: follow-up of 94% of the Scottish Mental Survey 1947. Intelligence 2017;63:45–50. sociation of childhood intelligence with mortality risk from ado- lescence to middle age: findings from the Aberdeen Children of 27. MacPherson JS. Eleven-Year-Olds Grow Up. London: University of London Press, 1958. the 1950s cohort study. Intelligence 2009;37:520–8. Downloaded from https://academic.oup.com/ije/article/47/1/89/4085882 by DeepDyve user on 16 July 2022 96 International Journal of Epidemiology, 2018, Vol. 47, No. 1 28. Terman LM, Merrill MA. Measuring Intelligence: A Guide to 34. Therneau TM. coxme: Mixed Effects Cox Models. R package the Administration of the New Revised Stanford-Binet Tests of version 2.2–5, 2015. Intelligence. Oxford: Houghton Mifflin, 1937. 35. Deary IJ. Follow-up studies of the Scottish Mental Surveys of 1932 29. Calvin CM, Batty GD, Brett CE, Deary IJ. Childhood club par- and 1947. In: Peel RA, Zeki, M (eds). Human Ability: Genetic and ticipation and all-cause mortality in adulthood: a 65-year Environmental Influences. London: The Galton Institute, 2006. follow-up study of a population-representative sample in 36. Davey Smith G, Hart C, Hole D et al. Education and occupa- Scotland. Psychosom Med 2015;77:712–20. tional social class: which is the more important indicator of mor- 30. Maxwell J. The Level and Trend of National Intelligence. tality risk? J Epidemiol Commun H 1998;52:153–60. London: University of London Press, 1961. 37. Davey Smith G. Health Inequalities: Lifecourse Approaches. 31. R Core Team. R: A Language and Environment for Statistical Bristol: Policy Press, 2003. Computing. Vienna: R Foundation for Statistical Computing, 2016. 38. Deary IJ, Johnson W. Intelligence and education: causal percep- 32. Revelle W. psych: Procedures for Personality and Psychological tions drive analytic processes and therefore conclusions. Research. Evanston, IL: Northwestern University, 2016. International Journal of Epidemiology 2010;39:1362–9. 33. Therneau TM. survival: A Package for Survival Analysis in r.R package version 2.40–1, 2015.

Journal

International Journal of Epidemiology – Oxford University Press

Published: Feb 1, 2018

Keywords: relationship - sibling

There are no references for this article.