Article Text
PDF
Abstract
Background Inverse associations between IQ and stroke have been reported in a few studies, but none have investigated subtypes of stroke, nor have they studied fatal and non-fatal stroke separately. Stroke is a heterogenic disease and strength of associations with IQ and putative causal pathways cannot be assumed to be identical for different subtypes.
Methods IQ was measured for 1.1 million Swedish men, born 1951 to 1976. Data from several national registers were linked and the cohort followed until the end of 2006 for non-fatal, and 2004 for fatal stroke. HRs with 95% CIs adjusted for age, body mass index, blood pressure and socioeconomic factors were estimated using Cox proportional hazards models.
Results Inverse associations were found between IQ and all stroke subtypes. The strength of the associations differed by subtype, with the strongest RR found for haemorrhagic stroke. In adjusted models using IQ as a continuous variable over a standard nine point scale, HR for mortality in all stroke was 0.89 (95% CI 0.85 to 0.93), that is an 11% decrease in stroke risk per unit increase in IQ. For non-fatal stroke, the corresponding HR was 0.92 (95% CI 0.91 to 0.93). The results were based on a rather young cohort, and results should therefore be generalised to early stroke events rather than the general population.
Conclusions Inverse associations were found between IQ and all stroke subtypes, fatal and non-fatal. For all types of non-fatal stroke, the inverse associations with IQ remained after adjustments for childhood and adult socioeconomic position.
- IQ
- ischaemic stroke
- haemorrhagic stroke
- epidemiology
- cohort ME
- epidemiology ME
- stroke DI
Statistics from Altmetric.com
Introduction
Intelligence (IQ) has been shown to be inversely associated with all-cause and cause-specific mortality in populations followed into middle-age and older, even after adjustment for childhood and adult socioeconomic position.1–8 An inverse relationship has also been demonstrated between offspring IQ and parental mortality.9 A recent study showed that even when IQ was measured as early as 5 years of age, it was associated with risk factors for premature mortality at age 30, such as smoking, obesity and hypertension.10 Coronary heart disease (CHD) is one of the conditions showing the strongest inverse association with IQ.7 However, for other cardiovascular diseases such as stroke, the evidence is less clear. Given the consistent findings for CHD,1 6 7 11 similar results might be expected for stroke, as CHD and stroke share many physiological, behavioural and socioeconomic risk factors.
Only five previous studies have reported associations between IQ in childhood or in early adulthood and mortality and/or morbidity from stroke.3 6 11–13 They all reported inverse associations between IQ and stroke; but after adjustment for social position in childhood and other potential risk factors for stroke, such as blood pressure, smoking and birth weight, the associations were attenuated and no longer statistically significant in any of these studies. In the last study, stroke was only separated from CHD in crude analysis. Morbidity and mortality were not analysed separately in any of the previous studies. In addition, no analyses were performed for subtypes of stroke, except in the study by Hart et al,12 in which haemorrhagic stroke was separated from all stroke, but the statistical power was low in that analysis.
Stroke is not a homogenous disease, with respect to risk factors and pathogenesis.14 For instance, high blood pressure seems to increase the risk for haemorrhagic stroke more than for ischaemic stroke.15–18 Several studies have reported positive associations between body mass index (BMI) and ischaemic stroke,19 20 but studies on BMI and haemorrhagic stroke have shown conflicting results.20–22
The absence of studies that have investigated the association between IQ and stroke subtypes, kept fatal and non-fatal stroke separate, and uncertainty regarding whether any association might be explained by confounding or mediation by physiological and/or socioeconomic factors, prompted the present population-based study based on a cohort of more than one million Swedish men.
Methods
Record linkage of registers and study cohort
The present study population consisted of all men born in Sweden 1951–1976 who were identified in the Military Service Conscription Register (MSCR) (1 252 008 men). Over the years covered by this study, military conscription was compulsory for male Swedish citizens and approximately 95% of the male population attended conscription examinations. Additional information was obtained by record-linkage of the following national registers: the Cause of Death Register, the Hospital Discharge Register, Statistics Sweden's Population and Housing Censuses, and the Education Register.
Conscription data
IQ was measured at military conscription at a mean age of 18.3 years (SD 0.55 years). The IQ test, which has been described elsewhere,23–25 consists of four subtests: a logical, a verbal, a spatial and a technical test. All tests were presented to the subjects in succession in written form. All test scores - including a global IQ score derived from a summation of the results of the four subtests - were standardised to give a Gaussian distributed score between 1 and 9. A higher value indicates greater cognitive ability. In this study, only the global IQ was used in the analyses. IQ was used as a continuous variable except in one analysis where IQ was trichotomised. BMI and systolic blood pressure were also measured during the conscription examination and were treated as continuous variables in the analyses.
Measures of socioeconomic position
Socioeconomic position (SEP) in childhood and adulthood is associated with IQ and also with risk factors for stroke.26 27 In the presented analyses three indicators of SEP were used; parental occupation, own attained education and own occupation. Parental occupation was assessed through Statistics Sweden's socioeconomic index and was extracted from the Population and Housing Censuses when the participants were 5-10 years of age. A six-stage classification was used as a categorical variable in the analyses6: higher level non-manuals,5 middle level non-manuals,4 lower level non-manuals,3 skilled workers,2 unskilled workers, and1 others including farmers, students, home makers and subjects with disability pension. Data on own occupation, classified according to the same categories, was obtained for a large subset (854 690 individuals) when the participants were age 28 years or older. Own attained education was extracted from the National Education Register (years 1990–2004). Level of education was classified into seven categories7: PhD education,6 other higher education >15 years,5 higher education 13–15 years,4 full secondary education (11–12 years),3 secondary education <11 years,2 9–10 years of primary school education and1 <9 years of primary school.
Outcome variables
The non-fatal stroke cases were collected through the Hospital Discharge Register and fatal cases through the Cause of Death Register. Stroke cases were classified according to the International Classification of Diseases (ICD) 8, 9 or 10 codes contained in the registers. Stroke was categorised as ischaemic stroke, 433 (ICD 8), 434 (ICD 9) and I63 (ICD 10), haemorrhagic stroke, 431 (ICD 8 and 9) and I61 (ICD 10), subarachnoid bleeding, 430 (ICD 8 and 9) and I60 (ICD 10), and other stroke 436.9 and 434 (ICD 8), 432, 437, 438, 435 (ICD 8 and 9), and 433 (ICD 9) and I62, I64-69, G45 (ICD 10). Analyses of non-fatal stroke included events registered in the Hospital Discharge Register and not in the Cause of Death Register, meaning that those who were hospitalised and died from stroke were considered as fatal stroke events. In the analyses of non-fatal stroke, all fatal stroke events were excluded and first event was used for all stroke. In the analyses of subtypes, first event by each subtype was used, resulting in a larger sum than for all stroke. Only 20% of the stroke cases in Sweden occur below the age of 65,28 which is why the stroke cases in this study can be considered early stroke cases.
From the study population of 1 252 008 men, 63 825 were excluded because they did not go through military conscription, 382 because they had been hospitalised for stroke before conscription, 27 299 because they lacked information on IQ and another 25 119 because they lacked information on any of the potential confounding factors included in the analyses. The final analytic cohort comprised 1 135 383 men (91% of the study population).
Statistical analysis
Data were analysed using Cox proportional hazards regression models with age as the time axis. The follow-up ended at the date of death or hospitalisation, date of emigration or on 31 December 2004 for mortality or 31 December 2006 for morbidity, whichever came first. For non-fatal stroke, first event was used for all stroke and first specific event for type-specific stroke. Fatal events were excluded from the analyses of non-fatal stroke. Analyses were adjusted for BMI and systolic blood pressure, childhood SEP (considered confounding factors) and own education and occupation (seen as mediating factors). Cohort effects were also tested for by using birth year in 5 year strata, as well as confounding from conscription age, test centre and the type of municipality (urban/rural) where they lived at the time of enlistment. As these potential confounding factors had no effect on the estimates they were not included in the final analyses. HRs were estimated together with their 95% CIs, using the phreg procedure in SAS version 9.1. The proportional hazards assumption was tested graphically and no evidence was found that it was violated. The analyses were first performed using IQ as a categorical variable with nine levels (tables not shown but available from the author upon request). The effect of IQ on stroke appeared to be linear over the whole IQ range, which led to the use of IQ as a continuous variable.
Results
Mean values of IQ, blood pressure and BMI were compared for those belonging to the study cohort but who were excluded from the main analyses due to missing data on any of the covariates. The estimates from unadjusted analyses were almost identical between individuals with complete data and those excluded due to missing data. In the final step of the analyses adjustments were made for education and own occupation. In this step, part of the study cohort was lost due to missing data on these two variables, either for technical reasons or because they were too young (below 28 years). However, there were no differences of estimates for the main analytic cohort and the subset when age-adjusted analyses or analyses with further adjustments were made.
Table 1 shows descriptive information on the distribution of IQ and potential confounders of the IQ-stroke association for subjects with and without stroke. For all the risk factors, there were obvious differences between subjects with stroke and those without, where IQ and the socioeconomic variables showed the largest differences, and BMI and blood pressure the smallest.
Distributions of study characteristics by stroke and death
In table 2, IQ was stratified into three classes: low, middle and high IQ. Number of events, mean age at a stroke event, total follow-up time and incidence rates are presented. The incidence rate of non-fatal stroke was about 14 times as high as the incidence rate for fatal stroke.
The mean follow-up time from conscription to first admission to hospital inpatient care for a non-fatal stroke was 24 years (SD 7.9) and for a fatal stroke 22 years (SD 7.7). During the follow-up period, 7661 non-fatal stroke events and 554 fatal stroke events occurred. Numbers of cases from specific underlying causes are presented in tables 3 and 4. The most common stroke subtype among fatal cases in the present study population was haemorrhagic stroke (40% of the cases as compared to 11% for ischaemic stroke). For non-fatal stroke, the order was the opposite (40% of the cases was ischaemic compared to 17% for haemorrhagic stroke).
Number of stroke events, follow-up time and incidence rate per IQ category
HRs with 95% CIs for IQ as a predictor for fatal stroke (N=1 135 383)
HRs with 95% CIs for IQ as a predictor for non-fatal stroke (N=1 134 829)
Tables 3 and 4 show crude and adjusted HRs for fatal and non-fatal stroke by stroke subtype. There were inverse associations between IQ and all stroke subtypes, with the strongest RR seen for fatal stroke of ‘other’ type (HRage-adjusted 0.83, 95% CI 0.71 to 0.97) and haemorrhagic stroke and subarachnoid bleeding (HRage-adjusted 0.86, 95% CI 0.81 to 0.92 and 0.86, 95% CI 0.80 to 0.91) respectively. There was little evidence of an association with fatal ischaemic stroke (HRage-adjusted 0.97, 95% CI 0.86 to 1.10). When adjusting for childhood SEP, BMI and systolic blood pressure (potential confounders), there was only a slight attenuation of the estimates. Controlling for own education and occupation (potential mediators) in the model further attenuated the estimates slightly, but inverse associations remained for all stroke subtypes. Significance was, however, lost for all fatal stroke outcomes except for all stroke and haemorrhagic stroke. For non-fatal stroke there were inverse association between IQ and all stroke outcomes, even after full adjustments, and the strongest RR was found for haemorrhagic stroke (HRfully adjusted 0.90, 95% CI 0.87 to 0.94). To make the meaning of the HR more apprehensible, it implies that changing from lowest IQ test score=1 to highest IQ test score=9 decreases the risk with 1-(exp(8*ln HR)), for example 1-(exp(8*ln 0.90)) = 57% for non-fatal haemorrhagic stroke. The results for fatal and non-fatal stroke were similar, although the larger number of events made the estimates for non-fatal stroke more precise. The analyses were also performed for all stroke, fatal and non-fatal taken together. The results were almost identical to the results for non-fatal stroke, HRmodel C 0.92, 95% CI 0.91 to 0.93 and HRmodel E 0.96, 95% CI 0.94 to 0.97.
Discussion
This is the first study of which the authors are aware that has analysed the association between IQ and stroke subtype and for fatal and non-fatal stroke separately. IQ has been shown to be inversely associated with stroke in previous studies,3 6 11 but in none of the studies was there any evidence of an association between IQ and stroke after adjustment for potential confounders. In two of the studies,6 11 there were few stroke events, which might explain the non-significant findings. In the third study,3 the adjusted estimates were statistically significant until the final model, where all the covariates were taken into account simultaneously. The association in the third study was based on fatal and non-fatal cases lumped together and the HR was 0.98 (95% CI 0.91 to 1.03). If any comparison is to be made it should be based on the estimate on all stroke events, fatal and non-fatal, where the HR was 0.96 (95% CI 0.94 to 0.97). However, the cases in the present study represent earlier stroke cases. One of the earlier studies on IQ and stroke11 separated early events from later (before and after 65 years), and found a higher risk of low IQ with early stroke than later stroke, which is in line with the present results.
Even though research within the field of cognitive epidemiology has started to explore the underlying mechanisms of the IQ-mortality association, the research conducted so far has mainly focused on the association of IQ with different disease outcomes. As mentioned earlier, few studies have been published with stroke as an outcome and most of them have had low power due to few stroke events. The present Swedish study shows an inverse association of IQ with all subtypes of stroke, with a stronger RR for haemorrhagic stroke than for ischaemic stroke. This finding is supported by the results of the only previous study separating haemorrhagic stroke from all stroke, in which Hart et al also found a stronger association for haemorrhagic stroke; however, there were only 12 haemorrhagic stroke events in that study.12 It can only be speculated why the risk differs by stroke subtype, but as stroke is a heterogenic disease with, at least partly, different risk factors for different subtypes, this finding is not implausible.
The causal pathways underlying the IQ-stroke mortality association are not well understood. In addition to unadjusted confounding due to unknown or unmeasured risk factors in childhood or adulthood and/or genetic confounding, the association may be biased by reverse causality.29 30 In the present study, IQ was measured in late adolescence whereas most disease events occurred many years later. As subclinical disease potentially influencing IQ is uncommon at these young ages, reverse causality may not be a plausible explanation for the association. Instead the IQ-stroke mortality association may be mediated by behavioural risk factors, such as smoking, dietary habits and physical activity.
The cases in the present study represent early stroke events. For ischaemic stroke, a large Swedish study14 showed that atherosclerosis in the aorta or the large arteries is less common among younger patients compared to older. Among the younger patients, no specific risk factors could be identified for a large percentage of the cases. A younger age of onset implies less exposure to behavioural and environmental risk factors, and genetic factors should consequently be more important.
Haemorrhagic stroke occurring at older ages is often influenced by amyloid angiopathy, that is accumulation of amyloid protein in vessel walls, decreasing elasticity and increasing fragility, making them more prone to rupture.31 However, as the present study cohort was relatively young at the end of the follow-up (28-53 years), hypertension and genetics, rather than fragile vessels caused by ageing, are likely contributing causes. Adjustments were made for blood pressure (and BMI) measured at the time of conscription, but there was no information on blood pressure in later life. However, it is well-known that blood pressure (and BMI) tracks from adolescence into middle age.32 33 If there had been access to information about later blood pressure (and BMI), adjustment might have attenuated the IQ-stroke association more than the very small attenuation seen in present estimates. However, it would be a matter of interpretation whether this was due to confounding or a mediating effect of IQ through blood pressure on risk of stroke. However, blood pressure and BMI were considered to be confounding factors as these physiological factors were measured at the same time as IQ. Interestingly, Batty et al found that IQ at age five was associated with high blood pressure in adulthood.10 It is possible that common genetic factors that increase the risk of stroke and are related to IQ and possibly also blood pressure, might partially explain the association between IQ and haemorrhagic stroke.
Epidemiological analyses should, of course, be adjusted for confounding, but adjustment for mediating factors might lead to overadjustment resulting in inappropriate attenuation of exposure-disease associations. The distribution of childhood SEP was different for stroke cases versus non-cases, and therefore taken into account in the present analyses.27 Model B in tables 3 and 4 demonstrates that adjustment for this confounding factor only had a marginal effect on the association between IQ and risk of stroke. Model C was additionally adjusted for BMI and blood pressure, considered potential mediators. But, as these variables were measured at the same time as IQ, they might also be seen as confounders, and model C would then demonstrate the strength of the IQ-stroke association without adjustment for mediators. It is debatable whether to adjust for adult SEP considered as a mediating factor and not a confounder. In particular, adjustment for own attained education is debatable when studying the effect of IQ on mortality, as IQ and education are highly correlated. On the other hand, controlling for education will leave the effect of IQ alone, that is without the mediating effect of education. Adjusting the IQ-stroke association for education and own occupation (mediators of the IQ-stroke association) will produce a more conservative estimate, and the associations seen in models D and E are probably overadjusted, that is underestimated. Even so, the association of IQ with stroke persisted after adjustment for both education and occupation for all non-fatal stroke outcomes indicating an effect of IQ on stroke not mediated by these socioeconomic factors. The results in the present study were based on a rather young cohort and results should therefore be generalised to early stroke events rather than the general population.
Study strengths and limitations
The present large, longitudinal dataset provides precise estimates, especially in the analyses of non-fatal stroke. As data on exposure, confounders and outcomes were collected from nationwide registers, that is independent sources, any classification bias should be non-differential.
It is a limitation that the study was based on men only, which made it impossible to generalise the findings to women. Furthermore, the results can only be generalised to stroke cases occurring up to middle-age and not to stroke occurring at older ages. Another limitation is the lack of information about risk factors for stroke other than BMI and systolic blood pressure in late adolescence, and SEP in childhood and adulthood. Information on dietary habits, physical activity, alcohol use, smoking and later hypertension would have been useful in the analyses, even if they are regarded as mediators rather than confounders. The possibility cannot be excluded that the IQ-stroke associations reported above are partially explained by unobserved factors not taken into account in the analysis. Therefore, it is not claimed that the IQ-stroke associations reported in this study are causal. Future research should explore the possible underlying pathophysiological and mediating processes of the IQ-stroke association for different types of stroke in more detail. As so much is still unknown about the causal pathways between IQ and stroke, the authors feel it would be premature to suggest a public health message or clinical advice.
Conclusion
The present results show that low IQ is a risk factor for both haemorrhagic and ischaemic non-fatal stroke. For all types of non-fatal stroke the inverse associations with IQ remained after adjustments for childhood and adult SEP. There was a stronger RR for haemorrhagic stroke than for ischaemic stroke. More information about the causal pathways linking IQ to stroke is needed before any clear public health recommendations or clinical advice can be suggested. The results were based on a rather young cohort, and results should therefore be generalised to early stroke events rather than the general population.
What is already known on this subject
There is strong evidence for an inverse association between IQ and cardiovascular morbidity and mortality.
However, less is known about associations between IQ and stroke. Those few studies on the relation of IQ with stroke that have been published so far have shown inconsistent results.
What this study adds
In this study, based on more than one million Swedish men, inverse associations were found between IQ and stroke, for haemorrhagic stroke as well as for ischemic stroke.
In addition inverse IQ-stroke associations were found for fatal as well as for non-fatal stroke.
References
Footnotes
Competing interests None.
Ethics approval This study was conducted with the approval of the Regional Ethics Committee in Stockholm, Sweden.
Provenance and peer review Not commissioned; externally peer reviewed.