Study population

The study was based on a cohort of Swedish males who were conscripted into compulsory military service in 1969/1970. Only 2–3% of all Swedish men were exempted from conscription at this time, in most cases due to severe handicaps or congenital disorders. Approximately 98% were born 1949–1951. The men born 1949–1951 were 49 321 in total, were all aged 18–20 years at conscription, and were all between 40 and 57 years of age approximately during the follow-up period from 1991 to 2008. During conscription, at any of seven regional conscription centres, each conscript underwent a series of tests of physical and mental health status, psychosocial functioning, and intelligence; full medical examinations were carried out; and self-administered questionnaires on family, social background, behavior and adjustment, and health and substance use were completed.

Ethics statement

The study was approved by the Stockholm Regional Ethical Review Board in Sweden (Dnr 2010/604-32). Due to the character of the data base and the anonymization of all data, the Review Board waived the normal requirement for written consent.

Measures of intelligence and psychosocial functioning

Psychometric assessment of intelligence was conducted during conscription through the use of four subtests, measuring verbal ability, logical-inductive ability, visuospatial ability, and technical comprehension. Results were converted into normally-distributed standard-nine (stanine) scales for each subtest, with scores 1 to 9, and the scales were then combined and transformed onto a new stanine scale as a measure of general intelligence, corresponding to approximate IQ bands of <74, 74–81, 82–89, 90–95, 96–104, 105–110, 111–118, 119–126 and >126. This general intelligence variable was used in the analyses in the present study. In the full cohort of men, 49 262 (99.9%) had a score on general intelligence.

Assessment of psychosocial functioning was made through a semi-structured interview administered by a certified psychologist. The overall objective of the interview was to assess the conscript's ability to cope with the psychological requirements of military service and, ultimately, of armed combat. Willingness to assume responsibility, independence, having an outgoing character, persistence, emotional stability, and power of initiative were regarded as the requirements for ‘high ability’ [26]. In the interview, usually lasting between 20 and 30 minutes, the psychologist asked not only about adjustment problems and conflicts, but also about successes, responsibilities taken on, and initiatives shown or experienced, in school, at work, in sports or other leisure activities, and at home [22]. Each conscript's mental energy, stability of emotions, social maturity, and active/passive interests were rated by the psychologist, who then assigned the conscript a summary score between 1 and 9 on psychosocial functioning, a variable constructed to follow a normal distribution. A high ranking on psychosocial functioning could be argued to bear similarities with low neuroticism, high conscientiousness, and high extraversion [22], and would thus be similar to the ‘general factor of personality’ that is found, for example, among traits of the currently-popular five factor model of personality traits [27]. Inter-rater reliability for the assessment of psychosocial functioning was found to be high (r = 0.86) in a test where 30 recorded interviews from 1972/1973 were scored by 30 psychologists [28].

Measures of socioeconomic status

For the present study, the cohort of conscripts was linked to the Longitudinal Database of Education, Income and Occupation (LOUISE) of 1990–2002, held by Statistics Sweden, in order to obtain information on educational level in 1990 for each study member. In the study, educational level was divided into five categories: ≤9 years of education, 10–11 years, 12–13 years, 14 years, and ≥15 years. This categorization reflects the educational arrangements of the time and roughly corresponds to compulsory level, vocational secondary level, pre-academic secondary level, university degree after 2 years of study, and university degree after 3 or more years of study. The majority of the cohort had completed no more than two years of post-compulsory education by the time of the conscription examination in 1969/70.

Linkage with the National Population and Housing Census of 1990 (response rate >98%) provided information on occupational class. A classification into the following eight classes was conducted by Statistics Sweden: unskilled workers, skilled workers, non-manual employees at lower (assistant), intermediate, or higher level, farmers, self-employed, and those for whom no occupation was reported. In the present study, we used the first five classes, which are hierarchically ordered. The self-employed were allocated to any one of these classes, based on occupational information in the 1990 census, and a number of men who had no reported occupation in the 1990 census were allocated on the basis of occupational information from the corresponding census in 1985. The 1985 and 1990 censuses also provided register-based information on level of income – any taxable income – before the follow-up. The average income of each subject for these years was divided into income quintiles in the present study. The extreme categories represent a number of people within the study population with very high or very low incomes (by Swedish standards in the late 1980's), which means that the differences between these categories and the middle categories are greater than the differences among the three middle categories.

Measures of mortality

In order to obtain information on mortality, the cohort of conscripts was linked to the National Cause of Death Register 1991–2008, held by the National Board of Health and Welfare. The conscripts were followed with regard to all-cause mortality, and to major cause-specific mortality: cardiovascular disease (CVD) mortality [ICD codes, 9th (390–459), and 10^th (I00–I99) revisions], mortality from injuries/violent causes [ICD codes, 9^th (800–999) and 10th (V–Y) revisions], cancer mortality [ICD codes, 9th (139–209) and 10th (C) revisions], and alcohol-related mortality [ICD codes, 9th (291, 303) and 10th (F10, K70, K74) revisions].

Measures of childhood socioeconomic circumstances and somatic diagnosis at conscription

To some extent, lower IQ and psychosocial functioning could reflect poverty and poor health in childhood [29]. Therefore, parental SES, crowded housing in childhood, and having a somatic diagnosis at conscription were treated as possible confounders in the present study. Parental SES refers to the father's occupational position (or that of any other head of household), a classification into seven groups made at Statistics Sweden, and crowded housing refers to ≥2 people per room (kitchen not included). Both variables were based on information obtained by linking the cohort with the National Population and Housing Census of 1960. Somatic diagnosis refers to having any non-psychiatric diagnoses recorded at conscription (ICD-8), excluding some frequent diagnoses unlikely to have an impact on SES attainment or premature death (e.g., refractive error).

Statistical analyses

For descriptive purposes, we computed mean IQ and psychosocial functioning, with standard deviations, measured at age 18–20 by educational level, occupational class, and level of income measured at ∼39 years of age, along with Spearman's correlation coefficients. We also calculated cumulative incidence of all-cause and cause-specific mortality across levels of education, occupational class, and income, in order to display the SES-related inequalities that are hypothetically explained by intelligence and psychosocial functioning.

Associations of IQ and psychosocial functioning, respectively, with cause-specific and all-cause mortality were estimated using Cox proportional-hazards regression, as implemented in the SAS (version 9.3) PHREG procedure. The increase/decrease in hazard ratios (HR) by lower level of IQ/psychosocial functioning was estimated using variables divided into three groups. Proportionality of hazards was checked with the LIFETEST procedure in SAS (survivor functions).

Associations between SES and premature death were estimated using the relative index of inequality (RII), which is in line with most corresponding previous studies. RII is a widely used regression-based summary measure that takes the size of hierarchical SES groups into account [30]. It is obtained by assigning each SES group a value between 0 (lowest rank) and 1 (highest rank), derived from the group's proportionate size and corresponding to the midpoint of the SES group's range. The mortality rate of the SES groups is then regressed on the RII scores, in the present study using Cox proportional-hazards regression (thus, RII = hazard ratio). For interpretation, a given RII should be seen as the ratio of the mortality between extremes in the lowest-ranking SES group and the highest-ranking SES group [31].

To analyze the extent to which IQ and psychosocial functioning statistically explained the associations between SES and causes of premature death in the cohort, we compared different regression models. The reference model (“Base”) included, in addition to SES in early middle age as predictor and subsequent mortality as outcome, the three following covariates: childhood social class (categorical) and crowded housing (dichotomous), and having a somatic diagnosis in late adolescence (dichotomous); the second model included IQ in addition to these; the third model included, instead, psychosocial functioning as an additional covariate; the fourth model included both IQ and psychosocial functioning as additional covariates. Thus, we sought to minimize confounding by childhood background variables in the estimation of the contributions of IQ and psychosocial functioning; we have previously demonstrated modest correlations between childhood SES and IQ/psychosocial functioning [22].

Percentage attenuations (with 95% confidence intervals) of SES–mortality associations between the base model and other models were calculated in 1000 bootstrap samples. These analyses were conducted with R 2.15.2 statistical software and the ‘boot’ package [32]. The formula for percentage attenuation was (RII_{base model}-RII_{other model})/(RII_{base model}-1)x100. Incidence rates/differences of all-cause mortality for lowest vs. highest SES groups were calculated and multiplied by percentage attenuations to provide a rough approximation of absolute differences in all-cause mortality possibly accounted for by IQ and psychosocial functioning together (reported in text only).

Methodological considerations

The present study was based on a large population, highly representative of men born around 1950 in Sweden. Only a very small proportion among Swedish men, aged 18–19 years in most cases, was exempted from the 1969/70 conscription examinations. The study used prospectively measured data on childhood background factors, IQ and psychosocial functioning in late adolescence, SES in early middle age, and mortality between 40 and 57 years of age, obtained from sources with minimal loss of information. Earlier studies have to a greater extent been based on cross-sectional measurements of IQ, personality characteristics, and SES [7], [10], [16], [17], and have typically been of limited size [33]. Self-rating problems were reduced: IQ was measured through a comprehensive multidimensional test and psychosocial functioning was rated on the basis of a 20–30 minutes long interview with a psychologist. Psychosocial functioning cannot be straight-forwardly compared with personality inventories such as the NEO-PI (“Big Five”) [34] since it is a one-dimensional summary measure. However, it should account for significant individual variation in psychological characteristics like level of activity, power of initiative, independence, and emotional stability, which we argue have similarities with high extraversion, high conscientiousness, and low neuroticism combining to form what is called the general factor of personality [27]. Estimated correlations between this and general intelligence have been rather similar to the correlation between psychosocial functioning and IQ (ρ = 0.30) found in the present study [27]. At the same time, it is possible that adjustments for psychosocial functioning in our analyses captured some effects of IQ.

Women could not be studied, since a corresponding source of information on IQ and psychosocial functioning did not exist for them. Whether or not associations between IQ/psychosocial functioning, SES in adulthood and mortality in women would be similar to the ones found in the present study is uncertain [5], [9].

Comparison with previous studies

A number of previous studies have investigated the extent to which socioeconomic inequalities in mortality may be confounded by intelligence differences. Research results – and interpretations – have been conflicting. Analyses from the Whitehall II study showed that IQ statistically explained over 20% of socioeconomic differences in CHD and mental functioning, and 30–40% of socioeconomic differences in physical functioning and self-rated health. However, associations between IQ and a majority of the health outcomes were found to be statistically non-significant after adjusting for SES differences and, therefore, the authors rejected the hypothesis that intelligence differences are an important explanation of SES-related inequalities in health [7]. Furthermore, analyses of the U.S. National Longitudinal Survey of Youth study suggested that IQ could not explain the association between education/income and mortality before the age of 50 [8], and a similar conclusion was drawn from a study in Malmö, Sweden [9]. Analyses of the West of Scotland Twenty-07 study, on the other hand, showed a stronger statistical explanatory power of IQ; for example, the associations found between education/social class and mortality/CHD mortality completely disappeared after adjusting for IQ, even though the IQ-adjusted associations between SES indicators and other health outcomes, which were “softer”, generally remained statistically significant and showed less attenuation in effect sizes [10]. Analyses of the Vietnam Experience study also showed substantial statistical explanatory power of IQ with regard to socioeconomic differences in total and CVD mortality; IQ showed greater explanatory power than did traditional CVD risk factors [12]. Furthermore, analyses of data linked between the Scottish Mental Survey of 1932 and the Midspan Studies [11] indicated the importance of IQ in this context, as did a study of more than 1 million relatively young Swedish men with regard to inequalities in mortality due to injuries [13]. The analyses in the present study showed that IQ could statistically explain roughly 45–50% of educational or occupational class inequalities in premature death due to injuries and CVD, rather similar findings to those of earlier studies [7], [12], [13]. The analyses also showed that IQ accounted for somewhat less of the SES-related inequalities in alcohol-related mortality. Finally, IQ was shown not to explain the small SES-related inequalities observed in cancer mortality, and, partly because of this, the contribution of IQ to inequalities in all-cause mortality was less, about 25%.

In the present study, analyses showed that psychosocial functioning also could account for about 25% of the inequalities in all-cause mortality. Further, it explained roughly 20% of the educational and occupational class inequalities in CVD mortality, and about 25–30% of the inequalities in injury and alcohol-related mortality. The potential importance of personality differences for SES-related inequalities in health was previously indicated in two studies (lacking information on IQ, however). The French GAZEL study showed that some specific measures of personality could together statistically explain about 30% of the associations between SES indicators and total mortality among middle-aged men; and about 40% of the associations with CVD mortality in men was explained by a measure of “coronary prone personality” [16]. In the MIDUS cohort [17], analyses showed that about 20% of the SES gradient in total mortality was statistically explained by differences in general personality traits from the Five Factor Model [34]. In the Vietnam Experience study, however, the neuroticism dimension of personality appeared unrelated to income inequalities in mortality [18].

The analyses in the present study showed that psychosocial functioning and IQ both contributed partly to explaining socioeconomic differences in major causes of mortality, and that their contributions correlated to some extent. Previous studies have not been able to examine this issue, and more studies are needed to establish the extent to which intelligence, psychosocial functioning and personality dimensions overlap in terms of their explanatory contributions to socioeconomic differences in mortality. Correlations between intelligence and aspects of personality are known in the literature [19].

In the present study, we found that the statistical explanations by IQ of social inequalities in mortality were consistently smaller when SES was measured by income level than when measured by level of education or occupational class, which is in line with earlier studies [7], [10], [11], [12]. This was not seen for our measure related to personality, in agreement with the one previous study with which a comparison can be made [16]. Smaller contributions from IQ when SES was measured by income resulted, overall, in less explanation of income inequalities in mortality than when it was measured by education or occupational class. This may possibly be because income, as compared to the other measures of SES, is more variable over time in individuals and captures change, and, furthermore, that it is to a greater extent affected by circumstances in present time, such as problems with health and employment [23]. Psychosocial functioning may be relatively more predictive of changing circumstances like these than of SES attainment earlier in adulthood [22], [26].

Interpretation

Higher scores on psychosocial functioning at 18–20 years of age, and even more on IQ, were found to be associated with higher SES attained in early middle age. Both IQ and psychosocial functioning were also related to all-cause and cause-specific mortality from 40 to 57 years of age. When this was taken into account, the associations between SES and major causes of death were attenuated. This was demonstrated in models where childhood background variables had already been accounted for. Thus, we cannot reject the hypothesis that observed associations between lower attained SES and higher risk of mortality, illustrated in Figure 1, are in part accounted for by individual psychological differences, although the idea that intelligence is the fundamental cause of social class inequalities in health [6] is only partly supported; it might be better characterized as one of the possible causes. Reaching this conclusion on the basis of a Swedish population is, by the way, no paradox; welfare states with egalitarian (e.g. educational) policies provide an environment in which intrinsic resources more than social background may serve as a link to higher attained SES [35], [36]. Within such societies, egalitarian policies affecting children may have reduced the extent to which social disadvantage can account for intelligence and personality differences [29].

Individual psychological differences might reflect varying ability among individuals to deal with long-term risks of disease and death. With regard to intelligence, Gottfredson and Deary [20] have argued that its relationship to health may mainly be due to efficient self-care and safer behaviors among individuals higher in intelligence. Negative health behaviors, such as tobacco smoking, have been shown to be potentially important factors explaining associations between individual characteristics, SES in adulthood, and health outcomes [25], [37]. Over the life course, significant individual differences may thus drive accumulation of advantage/disadvantage in terms of both SES and health [35].

Of the SES-related inequalities in total mortality, around 60–70% remained unexplained by IQ and psychosocial functioning as measured in the present study. In other words, major parts of the inequalities seen in Figure 1 could have explanations unrelated to individual differences in personality and intelligence.

In conclusion, personality characteristics and intelligence might both contribute to cause SES-related inequalities in premature death, but the magnitude of their contributions likely varies with measure of socioeconomic status and cause of death. Both personality characteristics and intelligence should be considered in future studies.