Methods

Study population

The study was based on data from a nationwide survey of 49 321 Swedish males, born 1949–51, who were conscripted for compulsory military service in 1969–70. The background to the Swedish conscription surveys and the variables included have been presented in detail elsewhere.1213 At that time, only 2–3% of all Swedish men were exempted from conscription, in most cases as a result of severe handicaps or congenital disorders. Those included in this study accounted for 97.7% of all conscripts in 1969–70; the remaining 2.3% were born before 1949. The data collection is described in table 1.

Information collected at the 1969–70 conscription examination

The IQ tests performed included tests on logic/general intelligence; verbal test of synonym detection; visuospatial/geometric perception; and technical/mechanical skills with mathematical/physics problems, which have been described in detail elsewhere.1415 The outcome of each test was ranked 1–9.14 These standard nine values were transformed into a new standard nine scale as a measure of general ability and corresponding to approximate IQ bands of: <74, 74–81, 82–89, 90–95, 96–104, 105–110, 111–118, 119–126, >126.15 Of the men, 49 262 (99.9%) had a score on cognitive ability.

All conscripts were seen by a physician who diagnosed any disorders according to the Swedish version of the International Classification of Diseases, 8th revision (ICD-8). All the men were seen by a psychologist for structured interviews. Conscripts reporting or presenting psychiatric symptoms were seen by a psychiatrist, and any diagnoses were recorded according to ICD-8.

At conscription, all men went through a health examination where height and weight were measured. The following variables were used as possible confounders of the cognitive ability–mortality association. Body mass index (BMI) was calculated using body weight (kg) divided by height (m) squared. In the analyses, BMI as well as body height were used as continuous variables.

During conscription, the men completed questionnaires concerning information on social background, behaviour and adjustment, psychological factors, health and substance use, eg, alcohol consumption and tobacco smoking. In the analyses, smoking was divided into the following four categories: 1 = >20 cigarettes/day, 2 = 11–20 cigarettes/day, 3 = 1–10 cigarettes/day and 4 = non-smokers. Alcohol consumption in g 100% alcohol/week was calculated on the basis of the answers to the questions on frequency and average consumed volume of beer, wine and strong spirits. A composite variable, risky use of alcohol, included at least one of the following indicators of problem drinking: consumption of at least 250 g 100% alcohol/week, to have taken an eye-opener during a hangover, to have been apprehended for drunkenness or to have often been drunk (alternatives given in the questionnaire were “often”, “rather often”, “sometimes” and “never”). The variable “contact with police and child welfare authorities” (at least once) indicated problem behaviour and has been shown to be strongly related to later psychiatric diagnosis.16 The variable “emotional control” was assessed by the psychologist in the structured interview. Low emotional control, defined as a score of 1 or 2 on a scale from 1 (lowest) to 5 (highest), was reported for 20% of the cohort, and served as a summary assessment of mental stability, emotional maturity and tolerance of stress and frustration.

Census data on childhood social circumstances

The conscripts and their parents, and other heads of household when different from the parents, were linked to each other between censuses through their personal identification numbers by Statistics Sweden. Information on “childhood socioeconomic position” and on “crowded housing” was obtained from the National Population and Housing Census of 1960 (response rate 99%), ie, when the subjects were 9–11 years old. The classification, into the following six socioeconomic groups, was based on information on the occupation of the head of the household: (1) unskilled workers; (2) skilled workers; (3) assistant non-manual employees; (4) non-manual employees at intermediate or higher level; (5) farmers; (6) those not classified in a socioeconomic group. We used here a measure of “crowded housing” (in the 1960 census, >2 people/room (kitchen not included) was classified as crowded).

Data on adulthood socioeconomic position, attained education and income

Information on adulthood socioeconomic group for each conscript was obtained by record linkage with the National Population and Housing Census of 1990 held by Statistics Sweden. This census had a response rate of over 98%. The classification into the following eight socioeconomic groups in 1990 was conducted at Statistics Sweden and is based on information on occupation and the educational level required for the occupations: (1) unskilled workers; (2) skilled workers; (3) assistant non-manual employees; (4) non-manual employees at intermediate level; (5) non-manual employees at higher level; (6) farmers; (7) self-employed (mostly skilled workers or drivers); (8) those for whom no occupation was reported (eg, unemployed, early retired or disabled). The data were also linked to the Longitudinal Database of Education, Income and Occupation (LOUISE) of 1990–2002 held by Statistics Sweden in order to obtain information on achieved education for each study member and income for the year 1990. In the analyses, achieved education was divided into five categories (⩽9 years, 10–11 years, 12–13 years, 14 years, 15 or more years), and income was divided into quartiles.

Information on outcomes

Information on mortality and age at death in the subjects between 1970 and 2003 was obtained from the National Cause of Death Register administered by the Centre for Epidemiology at the National Board of Health and Welfare in Sweden.

Data analysis

Associations between cognitive ability and mortality from 1991 to 2003 were estimated using Cox’s proportional hazards models, yielding hazard ratios (HRs) with 95% confidence intervals (95% CI), in both univariate and multivariate (including the potential confounders) models using the PHREG procedure in the SAS computer package. Cognitive ability was modelled as a continuous variable, with HRs given for a decrease of 1 in the nine-point scale, as well as with dummy variables for each of nine categories using the highest test score as the reference category. In the adjusted models, the relative hazard associated with a particular level of IQ score was estimated adjusting for the effect of the risk factors added to the model in order of their appearance in the life course. In order to calculate the proportions of the increased relative risk of mortality associated with IQ that were attenuated after adjustment for other risk factors, we used the formula ((RR_crude−RR_adjusted)/(RR_crude−1))*100. Possible interaction effects between IQ (dichotomised IQ 1–4 vs IQ 5–9) and other early life factors were investigated according to the method proposed by Rothman.1718

Results

Of the 49 321 men conscripted for military service in 1969–70, 48 283 men were still alive in 1990. Of these, 47 678 (98.7%) participated in the census that year. Our presentations are based on those 43 834 men (94.0%) for whom there was information on all the variables included in the final analyses. In table 2, the childhood and adolescent characteristics of the full cohort of those included in the analytical sample are shown. The differences in the prevalence of each risk factor between the full cohort and the sample are small.

In table 3, the associations between cognitive ability, measured at conscription, and mortality between 40 and 54 years of age are shown. IQ was inversely associated with mortality (RR = 1.15, 95% CI 1.12 to 1.18, for each decrease of one on the nine-grade scale). In the multivariate analyses, risk factors were added to the model in the order of their appearance in the life course, ie: (1) childhood social circumstances; (2) adolescence behaviour and mental health; (3) adult social circumstances. Adjustments for indicators of poor social circumstances in childhood (low childhood socioeconomic position (SEP), crowded housing and low body height) did not contribute to lower the increased risk found in the crude analyses. Adjustment for indicators of adolescent mental health and social adjustment, as well as for smoking and risky use of alcohol, each contributed to some attenuation of the association between IQ and mortality, while adjustment for overweight did not. After adjustment for all early life factors, the relative risk was attenuated by 40%. Adjustment for all indicators of social circumstances in adulthood by themselves lowered the increased relative risk substantially (HR = 1.04, 95% CI 1.01 to 1.07, for each decrease of one on the nine-grade scale, ie –73%). The single factor that contributed most to attenuate the increased relative risk found in the crude analyses was adult socioeconomic position. In the final model, taking all risk factors into account, the risks were strongly attenuated, and the relative risks associated with a decrease of one on the nine-point scale was on the limit of significantly increased (HR = 1.02, 95% CI 0.99 to 1.06, ie – 87%). We investigated possible interaction effects between IQ and each confounding or mediating factor adjusting for the other factors. There were no indications of interaction effects.

A strong and graded increase in HR for mortality was seen on the nine-grade scale from high to low test scores (table 4). After adjustment for risk factors over the life course, no significantly increased risk remained. The HR in the lowest of nine IQ groups, in comparison with the highest, decreased from 3.3 (95% CI 2.3 to 4.6) to 1.2 (95% CI 0.9 to 1.8) after adjustment for all risk factors. Excluding information on achieved education from the fully adjusted model did not change the results.

Discussion

This is one of very few studies on the association between IQ and mortality with the possibility to adjust the analyses for a large number of risk factors for mortality over the life course. It shows, in a cohort of Swedish men, that IQ test scores at ages 18–20 years were associated with mortality between 40 and 54 years of age. A graded risk increase was seen across all IQ score groups. The increased risk was markedly attenuated by simultaneous adjustment for indicators of poor mental health and behavioural factors measured in late adolescence. Further adjustment for measures of adult social position almost entirely attenuated the increased risk found in the crude analyses, indicating a strong mediating effect from adult circumstances.

Only a very small proportion of Swedish men are exempted from conscription, ie, the cohort is highly representative of men born in Sweden around 1950.

Many studies on early life factors and health outcomes in adult life rely on retrospective information collected at some point during the subject’s adult life, which may introduce bias.19 In this study, all information was collected at the time point when the conditions reported were actually present. This concerns socioeconomic position and crowded housing in childhood as well as the behavioural factors measured at conscription (overweight, smoking, risky use of alcohol) and the information obtained when the subjects were 39–41 years of age. Hence, non-differential misclassification in this study is most probably smaller than in previous studies, leading to less dilution of relative risk estimates associated with the risk factors measured in early life and resulting in a higher explanatory power of each potential risk factor.

The psychiatric diagnoses given at conscription were categorised according to ICD-8. At present, version 10 of ICD is in use. We believe that the measure of psychiatric illness used in this study captures quite well what would have been found using the updated version of ICD, although the specific diagnoses would differ somewhat.

All conscripts met a psychologist for a face-to-face interview in order to assess their level of emotional control. We are not fully aware of any similar measure used in personality research today, but we have previously reported strong associations between the measure of low emotional control and smoking,20 alcoholism,21 early retirement22 and mortality.13

This study only includes men, and the reported associations may not be similar among women. In the study based on the British 1946 birth cohort,4 men but not women showed an increased risk of mortality with low IQ. The men were followed for mortality between 40 and 53 years of age when mortality rates are still low. In a study by Hart et al,23 there was an association between IQ and mortality among people younger than 65 years but not among those older than 65 years. In a previous study based on the 1969 conscription cohort, we found that the strength of the association between IQ and mortality was the same over 35 years of follow-up after the conscription examination.6

Comparison with previous studies

Only a limited number of previous studies have reported the association between childhood IQ and mortality.8 A Scottish study based on 922 subjects born in 1921 found an association, including males and females in the same analyses, between IQ measured at age 11 years and mortality during a follow-up from middle age. An increased relative risk of mortality was found only in the lowest, compared with the highest, IQ quartile (RR = 1.47), and for those who died before the age of 65 years, but not among those who died later in life.323 This study did not include information on other early life circumstances. In the British 1946 birth cohort, IQ was measured at age 8 years, and the study population was divided into IQ quartiles and followed for mortality between 9 and 54 years of age.4 In the crude analyses, a doubled risk of mortality was found among men in the lowest IQ quartile. After adjustment for indicators of social circumstances in childhood and adulthood, the increased risk was reduced by 63% and was no longer significantly increased. No association between IQ and mortality was found among women. In a Scottish cohort based on just over 2000 men and women born in Aberdeen in 1921, a stronger association among women than among men was found between IQ at age 11 years and mortality up to the age of 76 years.1 In a Danish cohort based on men born in Copenhagen in 1953, a graded association was found between IQ measured at age 12 years and mortality followed up to the age of 47 years.5 The study did not include information on the participants’ adult life circumstances, and adjustment for childhood social class only slightly attenuated the association. Batty et al11 investigated the association between IQ and mortality in a cohort consisting of 4316 male former Vietnam-era US army personnel. The men were tested for IQ at entry into the service in late adolescence and were interviewed in middle age concerning potential risk factors for mortality, eg, smoking and social circumstances over the life course. The association between IQ and mortality found in the crude analysis was not affected by adjustment for childhood social circumstances, but was almost completely attenuated by adjustment for adult social circumstances. Very similar results have been reported from three other US cohorts with data on IQ collected in adolescence and mortality up to middle age.10 Our finding of a strong mediating effect from socioeconomic factors in middle age on the association between childhood IQ and mortality is in strong agreement with those recently published findings.

Is IQ a risk factor for mortality?

We found a gradually increased risk of mortality by decreasing test score on all nine levels among men aged 40–54 years of age. Adjustment for all risk factors almost entirely attenuated the increased risk of mortality found in the crude analysis.

The indicators of poor childhood social circumstances (socioeconomic position and crowded housing measured at ages 9–11 years, and body height measured at ages 18–20 years) did not contribute to a decrease in the association between IQ and mortality. This is in accordance with findings in previous studies.14567 In this study, we were able to adjust for psychiatric diagnoses, social misbehaviour (self-reported contact with police and child care authorities) and low emotional control at conscription. All three factors contributed somewhat to attenuate the increased risk found in the crude analyses. It may be that those factors are indeed indicators of negative childhood experiences not captured by our measures of childhood social circumstances. There is evidence that a poor psychosocial home environment may be associated with negative health outcomes,24 negative health-related behaviours,25 as well as hampering IQ development.262728 It may also be that people with lower IQ have difficulty in interpreting signs of upcoming psychiatric symptoms and therefore have a lower ability to take preventive actions.15

It has been suggested that those with lower IQ may have more difficulty in understanding and interpreting health prevention messages, eg, related to smoking, risky use of alcohol and physical activity.29 Previous investigations on the association between IQ and lifestyle habits often point to the importance of socioeconomic factors. In this study, smoking and risky use of alcohol, but not overweight, contributed somewhat to attenuate the increased risk found in the crude analyses. Taylor et al30 found that childhood IQ was not associated with starting smoking, but with smoking cessation in adulthood among persons born in 1921. However, after adjustment for adult social class and deprivation category, the association was no longer significant. We have shown recently in this conscription cohort an association between IQ and smoking at age 18 years, which almost disappeared after adjustment for other predictors of smoking. No association was found between IQ and smoking cessation later in life.31 It is therefore unlikely that the misclassification of smoking status during follow-up is related to IQ. It has been reported that people with lower IQ more often report high alcohol consumption.32 In other studies, high alcohol consumption seems to be more common among people with higher IQ.33 In our study, a measure of risky use of alcohol contributed somewhat to attenuate the association between IQ and mortality.

Cognitive ability is strongly associated with indicators of achieved social position. In the study by Kuh et al4 on the association between IQ and all-cause mortality, 63% of the increased risk among men disappeared after adjustment for education and achieved socioeconomic circumstances, indicating the great importance of a pathway mechanism. In this study, we adjusted in the analyses for risk factors for mortality measured at different points in time during the life course. It turned out that adjustment in the analyses for indicators of adult social position contributed greatly to reducing the associations between IQ and mortality (−73% and fully adjusted −87%). Very similar results were shown in three US cohorts with data on IQ collected in early life.1011 This result is in line with a pathway effect where IQ determines adult social circumstances and those circumstances in turn determine the risk of mortality.3435

As IQ and education are strongly correlated, one might question the logic of adjusting the association between IQ and mortality for education. However, in this study, the introduction of achieved education in the analytical model after accounting for all other factors did not change the results at all.

Conclusions

In our analyses of the association between childhood IQ score and later mortality, we adjusted for indicators covering information on childhood circumstances, health behaviours and adult social circumstances. We found that the association between IQ and mortality disappeared entirely after adjustment in the analyses for adult social circumstances and social and behavioural factors measured in late adolescence. The association between IQ and mortality was strongly mediated by adult social circumstances.