Annals of Oncology 18: 21–28, 2007
doi:10.1093/annonc/mdl473
original article
IQ in early adulthood and later cancer risk: cohort
study of one million Swedish men
G. D. Batty1,2, K. Modig Wennerstad3, G. Davey Smith4, D. Gunnell4, I. J. Deary2,
P. Tynelius3 & F. Rasmussen3*
1
MRC Social and Public Health Sciences Unit, University of Glasgow, Glasgow; 2Department of Psychology, University of Edinburgh, Edinburgh, UK;
Child and Adolescent Public Health Epidemiology Group, Department of Public Health Sciences, Karolinska Institute, Stockholm, Sweden; 4Department of
Social Medicine, University of Bristol, Bristol, UK
3
Background: While several studies have reported an inverse relation between IQ and total mortality rates, little is
known about the association, if any, between IQ and disease-specific outcomes, particularly cancer.
of age, and who were followed for incident cancer. Hazards ratios for the relation between IQ and 20 cancer
outcomes were computed using Cox regression.
Results: During an average of 19.5 years of follow-up, there were 10 273 new cancer cases. IQ showed few
associations with the cancer end points studied. There was a suggestion that IQ was positively associated with lung
cancer, and inversely related to stomach, oesophageal and liver malignancies, although effects were modest. The
only robust gradient was found for IQ in relation to skin cancer (HRper one standard deviation advantage in IQ; 95% confidence
interval 1.18; 1.13, 1.24; P value for trend across categories: <0.01), which was attenuated but retained statistical
significance after adjustment for indices of socioeconomic position across the life course.
Conclusions: In this large cohort of Swedish men followed into middle age, IQ was related to very few of the
cancer outcomes under investigation. This indicates that the recent observation that low IQ is related to increased
mortality rates may not be generated by an IQ–cancer gradient. Given that the present analyses are among the first
to examine these associations, replication is required.
Key words: cancer, cohort, IQ, men
introduction
In the rapidly evolving field of cognitive epidemiology [1],
a recent systematic review of nine studies found that mental
ability (denoted here as IQ) is inversely associated with an
increased risk of all-cause mortality in populations followed for
up to six decades [2]. Thus, high IQ scores seem to confer
protection against premature mortality. This association appears
to be strong, consistent across populations and research groups,
and independent of early life socioeconomic circumstances [2].
Crucially, in the studies on which this review is based, IQ has
been assessed in childhood or early adulthood when scores are
unlikely to be influenced by existing morbidities, such as diabetes
and hypertension, which can reduce mental ability [3, 4] and, in
studies of middle- and older-age adults, potentially generate
a spurious inverse IQ-mortality relation [5].
While evidence for a negative IQ-mortality gradient is
therefore growing, the influence, if any, of IQ on chronic
diseases, such as cardiovascular disease and cancer, which may
*Correspondence to: Dr F. Rasmussen, Child and Adolescent Public Health
Epidemiology Group, Department of Public Health Sciences, Karolinska Institute,
SE-17176 Stockholm, Sweden. E-mail: [email protected]
ª 2007 European Society for Medical Oncology
generate this association, has been little examined. Observations
that low childhood IQ scores are associated with smoking [6],
obesity [7] raised blood pressure/hypertension [8], alcohol
binge drinking [9] and socioeconomic deprivation [10] in adult
life provide a strong prima facie case for a relationship between
this psychological trait and cardiovascular disease [5]. There
is a suggestion that high IQ is associated with reduced
coronary heart disease risk, but there appears to be no
relation with stroke [11, 12].
While the relation between low IQ and increased prevalence
of some of the afore mentioned risk factors also points to
a similar relation with some cancers, another body of evidence
indicates the IQ-malignancy relation might, in fact, be positive.
Thus, IQ is positively related to both height [13] and the insulinlike growth factor-I system [14, 15]. Given that these factors in
turn demonstrate positive associations with elevated rates of
selected cancer outcomes—including carcinoma of the
colorectum, haematopoietic and prostate [16–17]—the same
pattern of association for IQ and these malignancies would
be anticipated.
To the best of our knowledge only three studies have assessed
the relation between IQ and cancer. With cancer of all sites
original
article
Methods: A cohort of 959 540 Swedish men who underwent IQ testing at military conscription at around 19 years
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
Received 19 July 2006; revised 16 August 2006; accepted 18 August 2006
original article
combined, IQ reveals null [11, 19] and inverse associations
[20]. While there is a suggestion of an inverse relation of IQ with
both stomach and lung cancer [20], these analyses, and those
featuring total cancers, are hampered by a low number of cases.
By utilising data arising from a cohort of over one million
Swedish men who had their IQ assessed during medical
examination at military conscription in early adulthood and
were linked to cancer registers, we are able to address this paucity
of evidence and methodological weaknesses by examining the
association between IQ and cancer risk in a dataset several orders
of magnitude larger than previously utilised.
methods
study participants and record linkage
conscription examination
The Swedish military service conscription examination involves a full
medical assessment in which physical health, mental status and cognitive
function (IQ) are ascertained. During the years covered by this study, this
examination was required by law; only men of foreign citizenship or those
with a severe medical condition or disability are excused. The present dataset
covers examinations from 1st January 1970 to 31st September 1994, after
which a modified IQ protocol was introduced at various stages according to
the conscription centre.
During the conscription examination, IQ was measured by four subtests
representing logical, spatial, verbal and technical abilities [21]. All test
scores—including a global IQ score derived from a summation of the four
subtests results—were standardised to give a Gaussian distributed score
between one and nine. Higher values indicate greater intellectual capacity.
Due to military secrecy, a detailed description of the tests is not available to
persons outside the conscription board, however, the subtests can be
described in general terms. The logical test measures the capacity to
understand written instructions and apply them to solving a problem. Items
from the spatial test depict a plan drawing of an object in its preassembled,
two-dimensional state. Respondents were required to identify, from a series
of drawings of fully assembled, three-dimensional objects, which it
represented. The verbal test measures the knowledge of synonyms. The
subject was required to determine which out of four alternatives is the
synonym of a given word. The technical test measures knowledge of
chemistry and physics, so implying a component of general knowledge. All
tests are presented in succession in the form of written questionnaires.
Further variables extracted from the conscription exam and used in the
present analyses were conscription centre (six offices), height and weight. The
latter two variables were measured directly using standard protocols from
which an index of adiposity, body mass index (BMI) (weight/height2), was
computed.
socioeconomic variables
The highest socioeconomic index of either parent was based on census
assessments in the 1960s and 1970s [five categories: nonmanual (high/
intermediate), nonmanual (low), skilled, unskilled and other]. The study
participants’ attained education level, originally coded into five categories,
22 | Batty et al.
was collapsed into four (<9 years of primary school, 9–10 years of primary
school, full secondary school, higher education).
cancer ascertainment
Cancer was ascertained either from the cancer register or from mortality
records. The Swedish Cancer Register is regulated by law with clinicians and
pathologists obliged to record new cases of cancer. Diagnoses in the register
were coded according to the seventh revision of the International
Classification of Diseases (ICD) [22], with conversions from later revisions
made as necessary. In the cancer register, the category of all malignant
neoplasms—referred to as ‘all cancers’—was divided into subcategories:
buccal cavity and pharynx (140–148); oesophagus (150); stomach (151);
colorectum (153–154); liver (155); pancreas (157); larynx (161); trachea,
bronchus and lung (162–163; referred to as ‘lung cancer’); prostate (177);
testes (178); other and unspecified genital organs (179); kidney (180);
bladder (181); skin (190–191); eye (192); brain (193); thyroid and other
endocrine glands (194–195); bone and connective tissue (196–197) and
lymphatic and haematopoietic tissues (200–207). ‘Other’ or unspecified
cancers comprised those malignancies that did not fall into any of these
categories.
statistical methods
Our analyses are based on 959 540 conscripts (71.3% of the linked
population) with complete information on IQ and collateral characteristics,
and with no record of a prior cancer diagnosis (see later explanation).
Pearson’s correlation coefficients among the four IQ subtests were high, with
a mean of 0.56 (range 0.48–0.70; Table 1). A high correlation was observed
for global IQ in relation to the subtests. Given also that preliminary analyses
revealed that patterns of association between each of the IQ subtest scores
and various cancers was similar, for brevity, we report on the association
between cancer and global IQ scores only (denoted hereafter as IQ).
We examined the proportional hazards assumption graphically for IQ in
relation to all cancers combined and found no evidence for violation. We
therefore used a series of Cox proportional hazards models [23] to assess the
influence of global IQ on cancer incidence (registration or death). In these
analyses we adjust for a range of confounding or mediating variables as
follows. In view of the well-documented secular increases in IQ [24], and the
wide range of birth years (24 years) in the present analyses, we controlled for
birth year. IQ has also been shown to be related to both height (a marker of
early life socioeconomic position) [8] and BMI (an indicator of adiposity)
[8], and both are related to cancer risk [17, 18, 25, 26]. Height (a confounder)
and BMI (a mediator) were therefore also added to the multivariable model.
Given that it is plausible that the IQ testing protocol could have varied by
conscript testing centre we also included this factor as a potential confounder.
Hazards ratios with accompanying 95% confidence intervals (CIs) were
computed that were initially adjusted for age (in using age as the time axis,
we controlled for it), then parental social class, subjects’ own educational
level, and, finally, all covariates described above. In keeping with previous
analyses [27], we categorised global IQ data into three groups (‘low’ = 1–3,
Table 1. Correlation coefficients for the association between global
IQ and subtests (N = 959 540)
Global
Logic
Synonym
Spatial
Technical
Global
Logic
Synonym
Spatial
Technical
1.00
0.85
0.80
0.77
0.79
1.00
0.70
0.55
0.57
1.00
0.48
0.52
1.00
0.55
1.00
P value for all correlation coefficients is <0.001.
Volume 18 | No. 1 | January 2007
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
The study cohort comprised all nonadopted men born in Sweden from
1952 to 1976 for whom both biological parents could be identified in the
Multi-Generation Register (MGR). Using unique personal identification
numbers we were able to link these men to the Military Service Conscription
Register (MSCR), the Cancer Register, the Cause of Death Register,
Population and Housing Censuses records (1960, 1970 and 1990) and to the
Register of Education (2000). This resulted in 1 346 545 successful matches.
Study approval was obtained from the Ethics Committee, Stockholm.
Annals of Oncology
original article
Annals of Oncology
results
We examined the relationships between IQ and each of the
study covariates (Table 2). In general, the most favourable level
of each characteristic was evident in the higher scoring
IQ groups. Thus, in comparison to their lower scoring
counterparts, higher IQ-scoring men were slightly leaner and
markedly taller; they were also less likely to have a parent in an
unskilled occupation and basic educational credentials
themselves. The average age at conscription examination was
marginally higher in the higher IQ-scoring groups.
Unsurprisingly, the highest correlation coefficient was for the
relationship between IQ and education.
In Tables 3, 4 and 5, we report on the relation between IQ
and subsequent cancer risk. During a mean of 19.5 years of
follow-up (range 0.01–32.9), there were 10 273 cancer events. In
age-adjusted analyses, there was a weak, positive association
between IQ and all cancers combined (P for linear trend across IQ
groups = 0.01; Table 3). This gradient was attenuated markedly
when the two indicators of socioeconomic position (paternal
occupational social class and subjects’ educational level) were
added separately to the multivariable models. After control for all
study covariates, there was no evidence of an IQ–total cancer
relation (HRper one SD advantage in IQ; 95% CI 1.01; 0.98, 1.03).
In some analyses there was a suggestion of a weak, inverse
relation of IQ with cancer of the stomach and oesophagus, and
a similarly modest positive relation with lung malignancy. None
of these gradients, however, achieved statistical significance at
conventional levels.
In Table 4, high IQ scores were associated with an elevated
risk of carcinoma of the skin (1.18; 1.13, 1.24). In keeping with
most other analyses, adjustment for markers of socioeconomic
position led to a weakening of this gradient but a significant
effect remained. We also examined the relation between IQ and
type of skin cancer (data not tabulated). The age-adjusted
associations between IQ and melanoma (1.19; 1.13, 1.25; N =
1617 cases) and other forms of skin malignancy (1.18; 1.07, 1.29;
N = 460 cases) were the same, matching those for all skin
cancers combined (1.18; 1.13, 1.24; Table 4). Finally, we also
explored if there was any effect modification in this relationship
according to area of residence of the study participant at
conscription (coded into: principal city or suburb thereof; large
city or industrial region; or rural area), or duration of follow-up
(coded into: 10 year-bands). In each case there was no
suggestion of interaction (P value for both ‡0.41).
In age-adjusted analyses, there was an indication of a negative
relationship between IQ and liver malignancy (P value for trend
across IQ categories = 0.08) that generally held when other
covariates were added to the multivariable model. While IQ
did not reveal an association with incident testicular cancer,
there was some evidence of an inverse relation with other
malignancies of the genitals (Table 5). The apparent significant
influence of IQ on kidney cancer (0.83; 0.70, 0.96) was lost when
adjustment was made for educational attainment (0.81; 0.61,
1.08). Although there was some evidence of a positive gradient
for IQ in relation to ‘other and unspecified cancers’, IQ was
essentially unrelated to the remaining cancer sites examined
(Tables 2–5). When we excluded men who developed cancer in
the first 5 years of follow-up and repeated the above described
analyses, the association between IQ and each malignancy was
essentially the same.
Given the relation of smoking with both certain cancers [29]
and IQ [6, 8], the former is a candidate confounder in the
present analyses. For a subgroup of men (N = 45 710) in the
study population, data on smoking habits were collected.
Adjustment for this behaviour had essentially no impact on the
null relation between IQ and all cancers combined (N = 828
cases). There were too few site-specific cancers to facilitate
further analyses.
Table 2. Association between global IQ score and covariates (N = 959 540)
Age at testing (mean, SD)
Body mass index (kg/m2) (mean, SD)
Height (cm) (mean, SD)
Parent in unskilled occupation 1960–1970 (%)
Low educational level (£10 years) (%)
Birth year 1952–1955 (%)
Global IQ
Low (N = 191 668)
Medium (N = 524 293)
High (N = 243 579)
Correlation with
global IQ
18.22 (0.56)
21.87 (3.22)
177.81 (6.58)
35.59
30.90
17.68
18.23 (0.52)
21.69 (2.79)
179.20 (6.40)
25.69
12.90
16.93
18.32 (0.53)
21.38 (2.49)
180.42 (6.37)
17.17
3.69
19.17
0.068
ÿ0.060
0.134
ÿ0.244
0.453
ÿ0.016
P value for all correlation coefficients is <0.001.
Volume 18 | No. 1 | January 2007
doi:10.1093/annonc/mdl473 | 23
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
‘medium’ = 4–6, ‘high’ = 7–9) and tested for a linear trend in cancer risk
across the IQ groups. In addition to the computations of effect estimates
across the IQ categories, we also calculated hazards ratios per standard
deviation (SD) (1.93 IQ units) increase in IQ score. The follow-up period
began at the date of conscription. Men were censored at the time of
death from causes other than cancer, emigration, or 31 December 2001,
whichever came first.
It is plausible that the occurrence of cancer before IQ assessment,
whether clinical or subclinical, may lead to poor test performance, either
because of medical treatment (e.g. radiation therapy of the central nervous
system), the comorbidity itself, or possibly owing to the low mood or
social withdrawal it may precipitate. To address the issue of reverse causality
due to clinically frank cancer, as described, we excluded from our analyses
persons who had a cancer episode before the conscription test (N = 1041).
To examine the issue of reverse causality due to subclinical cancer, we
dropped men with a cancer registration or cancer death within the first
5 years of follow-up after the conscription examination (N = 760) and
repeated the above analyses. In doing so we reasoned that cancer, if hidden
at study induction, would have become clinically apparent within this
5-year period. All analyses were computed using STATA version 8.1
computer software [28].
original article
Annals of Oncology
Table 3. Hazards ratios (95% confidence interval) for the relation between global IQ and cancer sites (N = 959 540)
Cancer outcome
(number of events)
Medium
(N = 524 293)
High
(N = 243 579)
P for
trend
Per one standard
deviation increase
in IQ
1993
1.0
1.0 (ref.)
1.0
1.0
5526
1.03 (0.98,
1.02 (0.96,
1.02 (0.97,
1.00 (0.95,
1.09)
1.07)
1.08)
1.06)
2754
1.08 (1.02,
1.04 (0.98,
1.04 (0.97,
1.01 (0.95,
1.14)
1.10)
1.08)
1.08)
0.01
0.24
0.19
0.69
1.03
1.02
1.02
1.01
(1.01,
(0.99,
(1.00,
(0.98,
1.05)
1.04)
1.04)
1.03)
62
1.0
1.0
1.0
1.0
157
0.95
0.96
1.01
1.00
(0.71,
(0.71,
(0.75,
(0.74,
1.27)
1.29)
1.36)
1.36)
68
0.85
0.87
0.98
0.99
(0.60,
(0.61,
(0.67,
(0.67,
1.19)
1.24)
1.44)
1.46)
0.33
0.44
0.92
0.96
0.92
0.93
0.96
0.97
(0.82,
(0.82,
(0.85,
(0.84,
1.03)
1.04)
1.10)
1.11)
11
1.0
1.0
1.0
1.0
20
0.68
0.61
0.84
0.79
(0.33,
(0.29,
(0.39,
(0.37,
1.42)
1.29)
1.80)
1.71)
7
0.49
0.39
0.75
0.67
(0.19,
(0.14,
(0.26,
(0.23,
1.25)
1.03)
2.16)
1.99)
0.13
0.06
0.58
0.46
0.76
0.70
0.89
0.85
(0.56,
(0.50,
(0.62,
(0.59,
1.04)
0.97)
1.27)
1.23)
31
1.0
1.0
1.0
1.0
79
0.96
0.98
1.10
1.13
(0.63,
(0.65,
(0.72,
(0.73,
1.45)
1.50)
1.68)
1.73)
29
0.72
0.75
0.99
1.04
(0.43,
(0.45,
(0.57,
(0.59,
1.20)
1.28)
1.73)
1.83)
0.20
0.29
0.99
0.86
0.82
0.83
0.91
0.92
(0.70,
(0.70,
(0.75,
(0.76,
0.97)
0.99)
1.09)
1.11)
146
1.0
1.0
1.0
1.0
379
0.97
0.97
0.98
0.97
(0.80,
(0.80,
(0.81,
(0.80,
1.17)
1.17)
1.20)
1.19)
178
0.93
0.93
0.95
0.94
(0.75,
(0.74,
(0.74,
(0.73,
1.16)
1.17)
1.21)
1.21)
0.52
0.54
0.68
0.63
0.97
0.97
0.98
0.98
(0.91,
(0.90,
(0.90,
(0.90,
1.05)
1.05)
1.07)
1.07)
11
1.0
1.0
1.0
1.0
10
0.34
0.32
0.33
0.31
(0.15,
(0.13,
(0.14,
(0.13,
0.81)
0.77)
0.80)
0.75)
11
0.78
0.67
0.84
0.74
(0.34,
(0.28,
(0.32,
(0.28,
1.80)
1.64)
2.16)
1.97)
0.64
0.46
0.68
0.54
0.93
0.88
0.93
0.89
(0.66,
(0.61,
(0.63,
(0.59,
1.31)
1.26)
1.38)
1.32)
41
1.0
1.0
1.0
1.0
103
0.94
0.94
1.06
1.03
(0.66,
(0.65,
(0.73,
(0.71,
1.35)
1.36)
1.54)
1.50)
56
1.05
1.05
1.34
1.27
(0.70,
(0.69,
(0.86,
(0.80,
1.56)
1.59)
1.11)
2.00)
0.78
0.78
0.19
0.30
1.01
1.02
1.11
1.09
(0.88,
(0.88,
(0.95,
(0.93,
1.16)
1.17)
1.30)
1.28)
a
Adjusted for age at testing, parental social class, subject’s educational level, body mass index, height, testing centre, and birth year.
discussion
In the present study of almost one million men who were well
characterised for IQ, socioeconomic position and cancer, we
examined the relation of IQ to 21 separate cancer outcomes
(20 of which were noninclusive). In general, there was limited
evidence of a link between IQ and these malignancies. One
exception was the IQ–skin cancer relationship where increased
rates of this neoplasm were evident in the higher IQ-scoring
men. In Sweden, and elsewhere, elevated rates of this skin
24 | Batty et al.
cancer have also been reported in the socioeconomically
advantaged relative to the less affluent [30], and in the highly
educated relative to persons with basic credentials [31].
Consistent with these results, controlling for markers of
socioeconomic circumstances across the life course in the
present analyses—parental social class and subjects’ own height
and educational attainment—led to some attenuation of the
IQ–skin cancer gradient but a significant effect remained. It
is plausible that adjustment for other indicators of
Volume 18 | No. 1 | January 2007
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
All (10 273)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusteda
Buccal cavity and pharynx (287)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Oesophagus (38)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Stomach (139)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Colorectal (703)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Larynx (32)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Lung (200)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Global IQ
Low
(N = 191 668)
original article
Annals of Oncology
Table 4. Hazards ratios (95% confidence interval) for the relation between global IQ and cancer sites (N = 959 540)
Cancer outcome
(number of events)
Medium
(N = 524 293)
High
(N = 243 579)
P for
trend
Per one standard
deviation increase
in IQ
14
1.0
1.0 (ref.)
1.0
1.0
47
1.22
1.16
1.08
1.02
(0.67,
(0.64,
(0.59,
(0.55,
2.22)
2.13)
2.00)
1.88)
23
1.13
1.01
0.97
0.83
(0.58,
(0.51,
(0.47,
(0.40,
2.20)
2.01)
1.98)
1.74)
0.78
0.93
0.87
0.57
1.03
0.98
0.97
0.92
(0.83,
(0.71,
(0.76,
(0.72,
1.27)
1.37)
1.23)
1.18)
309
1.0
1.0
1.0
1.0
1129
1.36 (1.20,
1.30 (0.15,
1.30 (1.14,
1.24 (1.09,
1.55)
1.48)
1.48)
1.41)
641
1.62
1.47
1.42
1.31
(1.41,
(1.28,
(1.22,
(1.12,
1.85)
1.70)
1.65)
1.53)
<0.01
<0.01
<0.01
<0.01
1.18
1.15
1.14
1.10
(1.13,
(1.10,
(1.08,
(1.05,
1.24)
1.20)
1.19)
1.16)
74
1.0
1.0
1.0
1.0
216
1.09
1.07
1.02
1.02
(0.83,
(0.82,
(0.78,
(0.78,
1.41)
1.40)
1.34)
1.35)
104
1.10
1.08
0.98
1.01
(0.82,
(0.79,
(0.71,
(0.72,
1.49)
1.47)
1.37)
1.42)
0.54
0.65
0.91
0.96
1.02
1.01
0.98
0.99
(0.93,
(0.91,
(0.87,
(0.88,
1.13)
1.13)
1.10)
1.11)
316
1.0
1.0
1.0
1.0
828
0.97
0.96
0.97
0.94
(0.86,
(0.84,
(0.85,
(0.83,
1.11)
1.09)
1.11)
1.08)
413
1.02
0.98
0.97
0.94
(0.88,
(0.84,
(0.82,
(0.79,
1.18)
1.14)
1.15)
1.08)
0.73
0.83
0.75
0.48
1.02
1.00
1.00
0.99
(0.97,
(0.95,
(0.94,
(0.93,
1.07)
1.05)
1.06)
1.05)
21
1.0
1.0
1.0
1.0
34
0.61
0.59
0.69
0.66
(0.35,
(0.34,
(0.39,
(0.37,
1.05)
1.03)
1.21)
1.16)
15
0.55
0.53
0.64
0.59
(0.28,
(0.26,
(0.30,
(0.27,
1.07)
1.05)
1.38)
1.28)
0.08
0.06
0.23
0.16
0.85
0.83
0.91
0.88
(0.67,
(0.65,
(0.70,
(0.67,
1.07)
1.06)
1.19)
1.16)
20
1.0
1.0
1.0
1.0
51
0.96
0.97
1.17
1.14
(0.57,
(0.58,
(0.69,
(0.67,
1.61)
1.64)
1.98)
1.94)
19
0.72
0.75
1.24
1.21
(0.39,
(0.58,
(0.63,
(0.60,
1.36)
1.64)
2.44)
2.42)
0.30
0.39
0.53
0.58
0.86
0.87
1.03
1.02
(0.70,
(0.70,
(0.82,
(0.80,
1.05)
1.07)
1.31)
1.29)
390
1.0
1.0
1.0
1.0
1113
1.06 (0.94,
1.03 (0.92,
1.02 (0.91,
1.00 (0.89,
1.19)
1.16)
1.15)
1.13)
537
1.09
1.03
1.01
0.98
(0.95,
(0.90,
(0.87,
(0.84,
1.24)
1.18)
1.17)
1.13)
0.23
0.73
0.94
0.74
1.05
1.03
1.02
1.01
(1.00,
(0.98,
(0.97,
(0.96,
1.09)
1.07)
1.08)
1.06)
a
Adjusted for age at testing, parental social class, subject’s educational level, body mass index, height, testing centre, and birth year.
socioeconomic position would have attenuated this relation
still further.
comparison with other IQ–cancer studies
As indicated, to our knowledge, the link between early life IQ
and the category of all cancers combined has been explored
in only three studies all of which are at least one order of
magnitude smaller in size than the present cohort. Findings are
mixed, with null [19, 32] and inverse gradients [20] reported.
Volume 18 | No. 1 | January 2007
In the present study, we found evidence of a modest positive
relation between IQ in early adulthood and all malignancies
combined that was eliminated following adjustment for markers
of socioeconomic position. An inverse IQ–cancer gradient
has been reported for neoplasm of the lung in two Scottish
studies [11, 20]. We found no such evidence of an effect herein;
indeed, if anything, the IQ–lung cancer gradient was positive.
We did, however, find some support for lower rates of
stomach cancer amongst higher IQ-scoring persons as
reported elsewhere [11].
doi:10.1093/annonc/mdl473 | 25
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
Prostate (84)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education -adjusted
Multiply adjusteda
Skin (2079)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Bone (394)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Haematopoetic (1557)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Liver (70)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Pancreatic (90)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Testicular (2040)
No. of events
Age adjusted
Age and parental social class adjusted
Age and education adjusted
Multiply adjusted
Global IQ
Low
(N = 191 668)
original article
Annals of Oncology
Table 5. Hazards ratios (95% confidence interval) for the relation between global IQ and cancer sites (N = 959 540)
Cancer outcome
(number of events)
Medium
(N = 524 293)
High
(N = 243 579)
P for
trend
Per one standard
deviation increase
in IQ
adjusted
33
1.0
1.0 (ref.)
1.0
1.0
66
0.75
0.72
0.72
0.71
(0.49,
(0.47,
(0.47,
(0.45,
1.14)
1.10)
1.12)
1.10)
32
0.76
0.70
0.66
0.63
(0.47,
(0.42,
(0.38,
(0.36,
1.24)
1.10)
1.16)
1.12)
0.30
0.18
0.15
0.12
0.96
0.93
0.81
0.90
(0.81,
(0.78,
(0.61,
(0.74,
1.13)
1.11)
1.08)
1.10)
adjusted
39
1.0
1.0
1.0
1.0
93
0.90
0.91
0.99
0.99
(0.62,
(0.62,
(0.67,
(0.67,
1.30)
1.32)
1.45)
1.45)
28
0.55
0.57
0.75
0.76
(0.34,
(0.35,
(0.67,
(0.45,
0.90)
0.94)
1.44)
1.31)
0.02
0.03
0.31
0.37
0.83
0.84
0.92
0.93
(0.70,
(0.71,
(0.77,
(0.78,
0.96)
0.98)
1.10)
1.11)
adjusted
67
1.0
1.0
1.0
1.0
132
0.74
0.73
0.74
0.73
(0.55,
(0.54,
(0.55,
(0.54,
1.00)
0.98)
1.01)
0.99)
71
0.81
0.80
0.82
0.79
(0.58,
(0.57,
(0.56,
(0.54,
1.13)
1.14)
1.20)
1.16)
0.26
0.25
0.32
0.24
0.90
0.90
0.94
0.88
(0.80,
(0.79,
(0.88,
(0.77,
1.02)
1.01)
1.01)
1.01)
adjusted
7
1.0
1.0
1.0
1.0
31
1.66
1.64
1.84
1.79
(0.73,
(0.71,
(0.79,
(0.77,
3.77)
3.75)
4.26)
4.17)
7
0.77
0.75
0.94
0.91
(0.27,
(0.26,
(0.30,
(0.29,
2.21)
2.20)
2.91)
2.87)
0.56
0.54
0.96
0.91
0.96
0.96
1.07
1.05
(0.72,
(0.71,
(0.76,
(0.75,
1.29)
1.30)
1.48)
1.48)
adjusted
252
1.0
1.0
1.0
1.0
618
0.92
0.91
0.89
0.90
(0.79,
(0.79,
(0.77,
(0.77,
1.06)
1.06)
1.04)
1.05)
303
0.95
0.93
0.91
0.92
(0.80,
(0.78,
(0.77,
(0.76,
1.12)
1.11)
1.09)
1.12)
0.54
0.46
0.32
0.42
0.97
0.97
0.96
0.96
(0.92,
(0.91,
(0.89,
(0.89,
1.03)
1.03)
1.02)
1.03)
adjusted
107
1.0
1.0
1.0
1.0
281
0.98
0.99
1.00
0.99
(0.79,
(0.79,
(0.79,
(0.79,
1.22)
1.24)
1.25)
1.25)
145
1.06
1.08
1.09
1.08
(0.83,
(0.83,
(0.82,
(0.81,
1.36)
1.40)
1.44)
1.43)
0.61
0.53
0.53
0.59
1.00
1.01
1.01
1.00
(0.92,
(0.92,
(0.91,
(0.91,
1.09)
1.10)
1.11)
1.11)
adjusted
84
1.0
1.0
1.0
1.0
263
1.17
1.19
1.23
1.22
(0.92,
(0.93,
(0.95,
(0.95,
1.50)
1.53)
1.58)
1.57)
131
1.21
1.27
1.34
1.34
(0.92,
(0.96,
(0.99,
(0.99,
1.59)
1.69)
1.82)
1.83)
0.19
0.11
0.06
0.07
1.07
1.09
1.12
1.12
(0.98,
(0.99,
(1.01,
(1.01,
1.17)
1.20)
1.24)
1.25)
a
Adjusted for age at testing, parental social class, subject’s educational level, body mass index, height, testing centre, and birth year.
validity of the IQ tests
The IQ tests utilised in Swedish conscription examinations are
not part of a recognised battery [21], and we are not aware of
any studies specifically assessing their validity. Findings from
the present analyses, however, provide some insight in this
regard. First, despite the materials used being dissimilar,
intercorrelation coefficients for subtest scores were high—a
common observation in psychometric studies. Secondly, in
keeping with other studies, IQ scores were associated with
educational level [10], socioeconomic background of the family
26 | Batty et al.
[33, 34], BMI [7] and height [35] in the expected directions
(Table 2). Thirdly, relative to their lower IQ-scoring peers, men
in the highest IQ band were also more likely to emigrate (hazard
ratio; 95% CI 4.00; 3.85, 4.17); again, an observation made
elsewhere [36]. Finally, a series of recent studies have shown an
inverse relation between IQ and all-cause mortality [37, 38].
This finding has been replicated in the present cohort [27].
On the basis of these observations, therefore, IQ in the present
study would appear to have a degree of concurrent and
predictive validity.
Volume 18 | No. 1 | January 2007
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
Other genital (131)
No. of events
Age adjusted
Age and parental social class
Age and education adjusted
Multiply adjusteda
Kidney (160)
No. of events
Age adjusted
Age and parental social class
Age and education adjusted
Multiply adjusted
Bladder (270)
No. of events
Age adjusted
Age and parental social class
Age and education adjusted
Multiply adjusted
Eye (45)
No. of events
Age adjusted
Age and parental social class
Age & education adjusted
Multiply adjusted
Brain (1173)
No. of events
Age adjusted
Age and parental social class
Age and education adjusted
Multiply adjusted
Thyroid (533)
No. of events
Age adjusted
Age and parental social class
Age and education adjusted
Multiply adjusted
Other/unspecified (478)
No. of events
Age adjusted
Age and parental social class
Age and education adjusted
Multiply adjusted
Global IQ
Low
(N = 191 668)
original article
Annals of Oncology
problem in the present study, however, the IQ–cancer gradients
we have reported would have to be in opposing directions in
persons omitted from the analyses. There were sufficient cancer
cases in persons excluded from analyses to allow an examination
of the relation between IQ and all cancer combined in this
group. The age-adjusted point estimate (HRper SD increase in IQ;
95% CI 1.02; 0.97, 1.08) was essentially the same as that seen
in the analytical sample (1.02; 1.01, 1.03; Table 3). The CIs,
however, were wider (owing to a lower number of cases) and
crossed unity. These findings therefore provide no strong
support for selection bias.
In conclusion, in this large cohort of Swedish men followed
into middle age, IQ was related to very few of the cancer
outcomes under investigation. This indicates that the recent
observation that low IQ is related to increased mortality rates by
middle age may not be generated by an IQ–cancer gradient.
Given that the present analyses are among the first to examine
these associations, replication is required. In particular, studies
of IQ in relation to cancer in women, which could not be
examined in the present study, are needed.
Table 6. Comparison of selected characteristics of men included in
the analytical sample with those excluded
references
Included
(959 540)
Age at testing (mean, SD)
Body mass index (kg/m2)
(mean, SD)
Height (cm) (mean, SD)
Low total IQ (%)
Low logic IQ (%)
Low synonym IQ (%)
Low spatial IQ (%)
Low technical IQ (%)
Parent in unskilled occupation
1960/70 (%)
Low educational level
(£10 years) (%)
Emigration (%)
Total cancer (HR, 95%
confidence interval)
Excluded
(387 005a)
18.25 (0.53)
21.64 (2.82)
18.44 (0.89)
21.87 (2.96)
179.23 (6.49)
19.98
19.95
20.22
18.94
22.86
21.45
179.18 (6.57)
6.05
7.28
6.99
6.34
6.76
14.48
14.15
3.29
1.0 (ref)
a
14.35
11.86
1.36 (1.29, 1.43)
Owing to missing data, group size varies according to characteristic.
P value for difference is <0.001 throughout.
SD, standard deviation; HR, hazard ratio.
Volume 18 | No. 1 | January 2007
contributors
David Gunnell, Finn Rasmussen and David Batty developed
the research question and the study design. David Batty
conducted the data analyses and wrote the first draft of this
manuscript to which all authors made significant contributions.
acknowledgements
David Batty is a Wellcome Fellow. Ian Deary is the recipient of
a Royal Society-Wolfson Research Merit Award.
1. Deary I, Batty GD. Cognitive epidemiology: a glossary of terms. J Epidemiol
Community Health 2005.
2. Batty GD, Deary IJ, Gottfredson LS. Pre-morbid (early life) IQ and later mortality
risk: systematic review. Ann Epidemiol; In press.
3. Manolio TA, Olson J, Longstreth WT. Hypertension and cognitive function:
pathophysiologic effects of hypertension on the brain. Curr Hypertens Rep 2003;
5: 255–261.
4. Awad N, Gagnon M, Messier C. The relationship between impaired glucose
tolerance, type 2 diabetes, and cognitive function. J Clin Exp Neuropsychol 2004;
26: 1044–1080.
5. Batty GD, Deary IJ. Early life intelligence and adult health. BMJ 2004; 329: 585–586.
6. Taylor MD, Hart CL, Davey Smith G et al. Childhood mental ability and smoking
cessation in adulthood: prospective observational study linking the Scottish
Mental Survey 1932 and the Midspan studies. J Epidemiol Community Health
2003; 57: 464–465.
7. Chandola T, Deary IJ, Blane D, Batty GD. Childhood IQ in relation to adult obesity
and weight gain: evidence from the 1958 birth cohort study. Int J Obes (Lond).
2006; 30: 1422–32.
8. Batty GD, Deary IJ, Macintyre S. Childhood IQ in relation to physiological and
behavioural risk factors for premature mortality in middle-aged persons: the Aberdeen
Children of the 1950s study. J Epidemiol Community Health; In press.
9. Batty GD, Deary IJ, Macintyre S. Childhood IQ and life course socio-economic
position in relation to alcohol-induced hangovers in adulthood: the Aberdeen
Children of the 1950s Study. J Epidemiol Community Health 2006; 60:
872–874.
doi:10.1093/annonc/mdl473 | 27
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
noncausal explanations for the IQ–cancer
relations
Considerations to be made when scrutinising findings from
observational studies include statistical chance, reverse causality,
confounding and selection bias. Given that we related IQ scores
to a total of 21 cancer outcomes—necessarily conducting a high
number of statistical tests in the process—some of the effects we
observed could have arisen by chance alone (e.g., for IQ and skin
cancer). In the present study, we were careful to address the issue
of reverse causality by excluding persons with known cancer
from all our analyses, and dropping men who developed the
condition in the early years of follow-up. Given its relation
with both IQ [33, 34] and mortality risk [38], socioeconomic
position in early life is most frequently posited as an important
confounder in the IQ–mortality link [5]. In the present study,
we took account of differences across IQ scores according to
socioeconomic position by adjusting for the socio-economic
position of both parents of the study participant.
Selection bias would occur in the present study if the IQ–
cancer gradients differed markedly between persons included
in the analyses and those excluded. In the present study,
approximately one quarter of the original study sample was
dropped for a variety of reasons, including pre-existing cancer
and missing covariate data. We compared the characteristics of
these men with those who were retained (Table 6). Owing to the
large sample size, these differences invariably attained statistical
significance although absolute values were not always large.
There was a higher proportion of persons in the lower IQscoring group in the analytical sample although, surprisingly,
this was not reflected in a difference in educational level. The
risk of cancer was lower in the analytical group, possibly because
pre-existing cancer at study induction (conscript examination)
was a criterion for exclusion. For selection bias to be a major
original article
28 | Batty et al.
25. Batty GD, Shipley MJ, Jarrett RJ et al. Obesity and overweight in
relation to organ-specific cancer mortality in London (UK): findings
from the original Whitehall study. Int J Obes (Lond) 2005; 29:
1267–1274.
26. Calle EE, Rodriguez C, Walker-Thurmond K, Thun MJ. Overweight, obesity,
and mortality from cancer in a prospectively studied cohort of U.S. adults.
N Engl J Med 2003; 348: 1625–1638.
27. Gunnell D, Magnusson PK, Rasmussen F. Low intelligence test scores
in 18 year old men and risk of suicide: cohort study. BMJ 2005;
330: 167.
28. Statacorp. Stata Statistical Software (release 8). College Station, TX: Stata
Corporation 2003.
29. Vineis P, Alavanja M, Buffler P et al. Tobacco and cancer: recent epidemiological
evidence. J Natl Cancer Inst 2004; 96: 99–106.
30. Hemminki K, Zhang H, Czene K. Socioeconomic factors in cancer in Sweden.
Int J Cancer 2003; 105: 692–700.
31. Hemminki K, Li X. University and medical education and the risk of cancer in
Sweden. Eur J Cancer Prev 2004; 13: 199–205.
32. Deary IJ, Whiteman MC, Starr JM et al. The impact of childhood intelligence
on later life: following up the Scottish mental surveys of 1932 and 1947.
J Pers Soc Psychol 2004; 86: 130–147.
33. Lawlor DA, Batty GD, Morton SMB et al. Early life predictors of childhood
intelligence: findings from the Aberdeen Children of the 1950s cohort study.
J Epidemiol Community Health 2005; 59: 656–663.
34. Lawlor DA, Najman JM, Batty GD et al. Early life predictors of childhood
intelligence: findings from the Mater University study of pregnancy and its
outcomes. Paediatr Perinat Epidemiol 2006; 20: 148–162.
35. Tuvemo T, Jonsson B, Persson I. Intellectual and physical performance and
morbidity in relation to height in a cohort of 18-year-old Swedish conscripts.
Horm Res 1999; 52: 186–191.
36. Batty GD, Clark H, Morton SMB et al. Intelligence in childhood and mortality,
migration, questionnaire response rate, and self-reported morbidity and risk
factor levels in adulthood—preliminary findings from the Aberdeen ‘Children
of the 1950s’ study (abstract). J Epidemiol Community Health 2002; 56
(Suppl II): A1.
37. O’Toole B, Stankov L. Ultimate validity of psychological tests. Personality and
individual differences 1992; 13: 699–716.
38. Whalley LJ, Deary IJ. Longitudinal cohort study of childhood IQ and survival up
to age 76. BMJ 2001; 322: 819.
39. Galobardes B, Lynch JW, Davey Smith G. Childhood socioeconomic
circumstances and cause-specific mortality in adulthood: systematic review
and interpretation. Epidemiol Rev 2004; 26: 7–21.
Volume 18 | No. 1 | January 2007
Downloaded from http://annonc.oxfordjournals.org/ at Florida State University on December 12, 2011
10. Neisser U, Boodoo G, Bouchard Jnr T et al. Intelligence: knowns and unknowns.
Am Psychol 1996; 51: 77–101.
11. Hart CL, Taylor MD, Davey Smith G et al. Childhood IQ, social class, deprivation,
and their relationships with mortality and morbidity risk in later life: prospective
observational study linking the Scottish Mental Survey 1932 and the Midspan
studies. Psychosom Med 2003; 65: 877–883.
12. Batty GD, Mortensen EL, Nybo Andersen AM, Osler M. Childhood intelligence in
relation to adult coronary heart disease and stroke risk: evidence from a Danish
birth cohort study. Paediatr Perinat Epidemiol 2005; 19: 452–459.
13. Jiang GX, Rasmussen F, Wasserman D. Short stature and poor psychological
performance: risk factors for attempted suicide among Swedish male conscripts.
Acta Psychiatr Scand 1999; 100: 433–440.
14. Gunnell D, Miller LL, Rogers I, Holly JM. Association of insulin-like growth
factor I and insulin-like growth factor-binding protein-3 with intelligence quotient
among 8- to 9-year-old children in the Avon Longitudinal Study of Parents
and Children. Pediatrics 2005; 116: e681–e686.
15. Kalmijn S, Janssen JA, Pols HA et al. A prospective study on circulating insulinlike growth factor I (IGF-I), IGF-binding proteins, and cognitive function in the
elderly. J Clin Endocrinol Metab 2000; 85: 4551–4555.
16. Renehan AG, Zwahlen M, Minder C et al. Insulin-like growth factor (IGF)-I, IGF
binding protein-3, and cancer risk: systematic review and meta-regression
analysis. Lancet 2004; 363: 1346–1353.
17. Gunnell D, Okasha M, Davey Smith G et al. Height, leg length, and cancer risk:
a systematic review. Epidemiol Rev 2001; 23: 313–342.
18. Batty GD, Shipley MJ, Langenberg C et al. Adult height in relation to mortality
from 14 cancer sites in men in London (UK): evidence from the original Whitehall
study. Ann Oncol 2006; 17: 157–166.
19. Hemmingsson T, Melin B, Allebeck P, Lundberg I. The association between
cognitive ability measured at ages 18-20 and mortality during 30 years of followup—a prospective observational study among Swedish males born 1949-51. Int
J Epidemiol 2006; 35: 665–670.
20. Deary IJ, Whalley LJ, Starr J. IQ at age 11 and longevity: results from a follow-up
of the Scottish Mental Survey 1932. In Christen Y (ed): Brain and Longevity:
Perspectives in Longevity. Berlin: Springer 2003.
21. David AS, Malmberg A, Brandt L et al. IQ and risk for schizophrenia:
a population-based cohort study. Psychol Med 1997; 27: 1311–1323.
22. Anon. Manual of the International Statistical Classification of Diseases, Injuries,
and Causes of Death, Vol 2 (seventh revision). Geneva: WHO 1957.
23. Cox DR. Regression models and life-tables. J R Stat Soc [Ser B] 1972; 34:
187–220.
24. Flynn JR. Searching for Justice: the discovery of IQ Gains over time. Am Psychol
1999; 54: 5–20.
Annals of Oncology