Page 1
THE UK CHILDHOOD CANCER STUDY: MATERNAL
OCCUPATIONAL EXPOSURES AND CHILDHOOD
LEUKAEMIA AND LYMPHOMA
Patricia A. McKinney1,*, Olaide Y. Raji1, Martie van Tongeren2 and Richard G. Feltbower1
1Paediatric Epidemiology Group, Centre for Epidemiology and Biostatistics, Room 8.49J, Level 8, Worsley
Building, University of Leeds, Clarendon Way, Leeds LS2 9JT, UK
2Institute of Occupational Medicine, Research Avenue North, Riccarton, Edinburgh EH14 4AP, UK
Risks of childhood leukaemia and lymphoma were investigated for specific work-related exposures of mothers in the UK
Childhood Cancer Study. Interviews with parents of 1881 leukaemia and lymphoma cases (0–14 years) and 3742 controls col-
lected job histories recording exposure to eight specific agents. Exposure was (1) self-reported and (2) reviewed, based mainly
on exposure probability and exposure level. Completeness, consistency and sufficiency evaluated data quality. Of all job
exposures which were self-reported as exposed, 33% cases and 34% controls remained classified as exposed after review, with
the remainder designated as partially exposed or unexposed. No review of underreporting of exposure was made. Data quality
was ‘good’ for 26% of cases and 24% of controls. For self-reported exposure, significant risks of acute lymphoblastic leukae-
mia (ALL) were observed for solvents and petrol in all time windows. For reviewed exposure, solvents remained significant for
ALL during pregnancy and postnatally. Restricting analyses to good-quality information removed all significant results.
Refinement of exposure assessment revealed misclassification of self-reported exposures and data quality influenced risk
assessment. Maternal exposure to solvents should further be investigated. These findings must invoke caution in the inter-
pretation of risks reliant on self-reported occupational data.
BACKGROUND
The volume of epidemiological literature on the role
of parental occupational exposures in the develop-
ment of childhood cancer has grown over many
years, suggesting that parental workplace exposure
to hazardous substances remains a potentially
important aetiological factor. Published studies have
generally focussed on paternal exposures and have
reported inconclusive results but there has been some
consistency for links between acute lymphoblastic
leukaemia (ALL) and maternal preconception and
perinatal exposures to solvents, paints, thinners and
lacquers(1 – 4). The lack of consistent evidence in the
literature may be explained by the wide ranging
methodologies employed, inherent flaws in study
design, lack of power to detect associations and
variation in exposure assessment methods.
Case–control studies, the most commonly employed
design for determining risks of childhood cancer and
parental occupation, are subject to biases and con-
founding(5,6). They frequently rely on information
recorded at interview involving retrospective recall of
job histories and associated exposure information. The
validity of self-reported information can be question-
able(7,8). Studies which have attempted to avoid pro-
blems of recall bias by employing routine data, such as
birth certificates(9), are hampered by inadequate infor-
mation. Investigations of parental exposures require
clear definition of the time windows of exposure,
with those occurring prenatally, during pregnancy
and postnatally associated with different routes of
exposure and separate mechanisms(10,11). Timing of
exposure is often ill-defined in publications.
Furthermore, statistical methods applied for risk esti-
mation frequently vary.
Inconsistent findings in the field of occupational
epidemiology can potentially be explained by varia-
tion in the methods of exposure assessment. Coding
of job titles or industrial groupings using different
schema making comparisons difficult, and different
job exposure matrices (JEMs) are applied to estimate
exposures. At a greater level of detail, self-reported
exposure to specific agents (e.g. specific chemicals) is
difficult to systematically elicit from interviewees
across studies.
The UK Childhood Cancer Study (UKCCS) is a
large nationwide case–control study established to
test five main hypotheses with respect to risk of
childhood cancer and leukaemia. One of the hypo-
theses addressed was ‘whether chemical or radiation
exposure of parents affects the risk of leukaemia or
cancer in their offspring’(12). Under this hypothesis,
findings of the analysis of parental occupation relat-
ing to coded job titles and industries have already
been published(13). Significant positive associations
for mothers’ occupational groups and childhood leu-
kaemia/ALL included periconception exposure
to dermal hydrocarbons and metal. The results of
analyses that have been extended to incorporate
*Corresponding author: p.a.mckinney@leeds.ac.uk
# The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org
Radiation Protection Dosimetry (2008), pp. 1–9 doi:10.1093/rpd/ncn265
Radiation Protection Dosimetry Advance Access published October 15, 2008
Page 2
additional information on specific work-related
exposures investigating possible associations for
childhood leukaemia and lymphoma are reported.
A novel method refining exposure assessment has
been applied with an exploration of the effects of
exposure misclassification and data quality on risk
estimates.
METHODS
Details of the UKCCS design and methods are pub-
lished elsewhere(12). In summary, the project was a
population-based case–control study of childhood
cancer, including histopathologically confirmed
cases of childhood leukaemia and lymphoma diag-
nosed 0–14 years between 1991 and 1996. Two ran-
domly selected controls were age (month and year of
birth), sex and regionally (n ¼ 8) matched to each
case. Case and control participation rates in the
UKCCS were 87 and 64%, respectively.
Personal interviews with parents of case and
control children collected data on a range of topics
including a full occupational history of jobs held for
over 6 months from leaving full-time education until
the date of diagnosis or pseudodiagnoses for cases
and controls, respectively. The gathering of occu-
pational details was facilitated by the use of a ‘pre-
interview questionnaire’ recording job title, industry,
job start and end dates and for each job a yes/no
response to exposure to the following specific agents:
(1) solvents, degreasers and cleaning agents;
(2) paints, thinners, turpentine, paint removers
and lacquers; (3) dyes and pigments; (4) petrol/pet-
roleum products; (5) lead and lead compounds; (6)
fertilisers, herbicides, pesticides and fungicides; (7)
radiation, radioactive materials, other forms of
ionising radiation; and (8) wood dust and sawdust.
If the respondent reported exposure to any of these
agents, further details were recorded, during the face-
to-face interview, including the name of the agent,
contact circumstances, specific tasks, direct handling
or not and duration of contact. Occupational exposure
information was available for 5623 mothers and 5349
fathers of children diagnosed with leukaemia and
lymphoma and their matched controls.
Exposure assessment
A novel, systematic exposure assessment procedure
was designed for characterising parental occu-
pational exposure from the detailed information on
exposure to the eight specific agent groupings. This
was undertaken blinded to parental or case–control
status. The key processes of exposure review are illus-
trated in detail in Figure 1. The review of
six exposure determinants provided an assessment of
exposure probability, exposure level, frequency of
exposure and degree of protection, deriving a final
‘reviewed’ exposure status (exposed, partially
exposed unexposed). For frequency of exposure and
degree of protection, data were too sparse for formal
evaluation, so the final reviewed exposure status was
based on exposure probability and exposure level for
the majority of reported job exposures.
A blind external validation of the review was
undertaken on a random sample of 250 job exposures
by an expert occupational hygienist (M.v.T.) for
exposure probability, exposure level and reviewed
exposure status. Comparing the designed and expert
assessments with respect to the three exposure indices
showed strong agreement for exposure status and
moderate agreement for exposure probability and
exposure level. The highest odds of agreement were
obtained for the exposed and unexposed categories of
the reviewed exposure status demonstrating good dis-
crimination between these two categories.
Data quality
A separate blind assessment of completeness, con-
sistency and sufficiency of the detailed information
on the eight agents was documented during the
review procedure. Overall data quality was then
classified as good, satisfactory or poor.
Statistical analysis
The statistical observational unit of analysis for the
estimation of risks was ‘job exposures’ to the eight
specified agents, that is, each exposure (or not) to a
specific agent, making the analysis ‘job-based’ not
‘person-based’. Thus, individual mothers were
included in each analysis dependent on their job
exposures in a particular time window. Unconditional
multivariable logistic regression produced odds ratios
(ORs) and 95% confidence intervals (CIs) for each
Figure 1. Procedure for retrospective review of occupational
exposures to specific agents.
P. A. MCKINNEY ET AL.
Page 2 of 9
Page 3
separate agent for both self-reported (yes/no) and
‘reviewed’ (exposed vs. unexposed, excluding partial
exposure). All ORs were adjusted for age (month and
year of birth), sex, study region (n ¼ 8) and an area-
based measure of deprivation using the Townsend
score(14,15). Analyses were conducted by diagnostic
subgroups, ALL, other leukaemia (OTH) [including
acute myeloid leukaemia (89%), chronic myeloid leu-
kaemia (10%) and other and unspecified leukaemia
(1%)], Hodgkin lymphoma (HL), non-Hodgkin lym-
phoma (NHL) and time window of exposure (precon-
ception, pregnancy, postnatal). The standard errors of
the ORs were estimated using the clustered sandwich
estimator(16) to correct for dependency among jobs
reported by the same mother. Dose–response ana-
lyses were investigated for statistically significant
exposure–cancer associations using the incremental
OR approach, which compares adjacent exposure
levels(17,18). Dose–response analyses were investigated
for statistically significant exposure–cancer associa-
tions using the incremental OR approach, which com-
pares adjacent exposure levels.
Results are presented for maternal exposures
and additionally for fathers when the context is
appropriate.
RESULTS
Occupational history data from personal interviews
with parents of 1881 leukaemia and lymphoma
cases (aged 0–14 years) and 3742 age-, sex- and
region-matched controls were analysed (Table 1).
This represented 49.1% of cases and controls in
the entire UKCCS data set. A total of 24 316
self-reported (yes/no) job exposures were reviewed
and re-assessed: 4833 (19.9%) for mothers and
19 326 (79.5%) for fathers and 157 (0.6%) for
non-biological parents. There was a lower overall
prevalence of exposure for mothers (cases 20.4%;
controls 17.3%) compared with fathers (cases 48.1%;
controls 42.1%). Broken down by agents, diagnostic
subgroups and time windows, the overall pattern for
mothers did not show any consistent pattern of case
excesses in reporting of exposures.
Comparing the components of the reviewed
exposure assessment by case–control status gave an
exposure probability of ‘certain’ for 33% of cases
and 36% of controls, inferring potential misclassifi-
cation of reported yes/no exposure for 67% of case
and 64% control mothers’ jobs. The low proportion
of ‘certain’ exposure probability was also observed
for each individual agent except for dyes, which
had higher proportions of case (61%) and control
(67%) mothers’ jobs assigned ‘certain’ exposure
probability. The minimal case–control differences
appeared general to all agents with control excesses
mostly.
For levels of exposure (high, moderate, low or
background), 78% of case and 74% of control jobs
were assigned as having low/background exposure.
These proportions were high compared with fathers
(66% cases and 64% controls) and showed case–
control variation by agent.
Table 2 shows the exposure distribution after
review of those job exposures which mother’s had
originally self-reported as exposed. The review
retained the original self-report of exposures for 33
and 34% of case and control mothers’ job exposures,
respectively. This was lower than for fathers (48%
cases, 51% controls). Sixty-seven and sixty-six per
cent of case and control mothers’ self-reported jobTable 1. Frequency (numbers and percentages) of
interviewed mothers by case–control status by diagnostic
subgroup.
Diagnostic subgroup Mothers
Case Control
n % n %
All haematological malignancies 1881 49.1 3742 49.1
Leukaemia 1580 91.0 3141 91.0
ALL 1324 90.7 2633 90.7
OTH 256 92.8 508 92.7
Lymphoma 301 86.5 601 86.6
Hodgkin’s lymphoma (HL) 92 78.6 184 79.0
NHL 209 90.5 417 90.5
Bold percentages calculated using all participating parents
in the whole UKCCS centres and for all cancers as
denominators. Unbold percentages calculated using parents
interviewed in centres across England and Wales only
(eight out of nine) and for each diagnostic subgroup as
denominators.
Table 2. Proportion of mothers self-reported job exposures
(case/control) by final ‘reviewed’ exposure status, agent,
case–control status and parent.
Exposure
agents
Number of
job exposures
case/control
Reviewed exposure
status (%) case/control
Exposed Partial Unexposed
All
agents
1797/3036 32.6/33.7 39.0/37.1 28.4/29.3
Solvents 582/853 25.3/23.8 55.1/56.5 19.6/19.7
Paints 254/440 41.3/40.0 33.5/32.7 25.2/27.5
Dyes 277/499 65.0/66.7 23.1/22.0 11.9/11.2
Petrol 154/179 21.4/35.8 44.8/41.3 33.8/22.9
Lead 55/136 54.6/54.4 25.5/24.3 20.0/21.3
Fertilisers 72/112 27.8/25.0 38.9/44.6 33.3/30.4
Radiation 300/630 18.7/20.0 21.0/26.2 60.3/53.8
Wood
dust
103/187 14.6/10.2 54.3/36.3 31.1/53.5
MATERNAL OCCUPATION AND CHILDHOOD LEUKAEMIA
Page 3 of 9
Page 4
exposures were, respectively, reclassified as unexposed
or partially exposed and were therefore potentially
false-positives. The proportion of misclassified job
exposures varied considerably by agent, and cases
and controls had variable levels of re-assignment
with marked disparities for exposure to wood dust
and petrol.
Overall data quality was assessed as good for 26%
of cases and 24% of controls with the majority of
the data deemed satisfactory (54% cases, 56% con-
trols). The quality of the data varied by agent with
high proportions of ‘good’ data available for asses-
sing dyes and radiation exposures.
The estimation of risks associated with job
exposures, rather than individual mothers, is first
reported for the self-reported exposure and secondly
for the reviewed exposure status and then the effects
of reclassification are presented. Results for risk of
ALL by time window of exposure are given in
Figure 2 for the dichotomous self-reported exposure
to the eight agent groupings. Mothers’ exposure to
solvents during the pregnancy and the postnatal
periods, and exposure to petrol in the three time
windows were significantly associated with elevated
risks of ALL. Statistically significant risks of OTH
were observed for mothers’ preconception exposure
to solvents (OR ¼ 1.6, 95% CI ¼ 1.1–2.3) and expo-
sure to petrol during pregnancy (OR ¼ 3.7, 95%
CI ¼ 1.6–8.6). A nearly 5-fold statistically significant
increased risk of OTH was also observed for exposure
to fertilisers during the pregnancy period (OR ¼ 4.8,
95% CI ¼ 1.7–13.4). Risk estimates for different diag-
nostic subgroups of childhood lymphoma showed
statistically significant relationships between the risk
of NHL and maternal preconception exposure to sol-
vents (OR ¼ 2.0, 95% CI ¼ 1.3–3.0) and petrol
(OR ¼ 3.7, 95% CI ¼ 1.8–7.5), exposure to petrol
(OR ¼ 4.4, 95% CI ¼ 1.9–10.4) and fertilisers
(OR ¼ 6.3, 95% CI ¼ 2.3–17.1) during the pregnancy
as well as postnatal exposure to fertilisers (OR ¼ 4.2,
95% CI ¼ 1.5–12.0). The risks of HL could not be
estimated for many of the mothers’ exposures because
of lack of exposed cases; no significant associations
were seen for the few estimated odds ratios.
Table 3 shows the OR for reviewed exposure
status (exposed vs. unexposed) by diagnostic group
and time window. Statistically significant positive
associations were seen between the risk of NHL
Figure 2. OR and 95% CI for mothers’ self-reported occupational exposure in relation to childhood ALL.
P. A. MCKINNEY ET AL.
Page 4 of 9
Page 5
and maternal preconception exposure to solvents
(OR ¼ 2.2, 95% CI ¼ 1.2–4.0), and ALL and
exposure to solvents during pregnancy (OR ¼ 2.7,
95% CI ¼ 1.6–4.6) and postnatally (OR ¼ 1.9, 95%
CI ¼ 1.1–3.3). For the pregnancy period, there was a
significant dose–response effect. No other maternal
exposures had statistically significant relationships.
To clarify the effect of the reviewed exposure on
the risk estimates, a comparison of ORs for self-
reported (yes/no) and reviewed exposure (exposed
vs. unexposed) are summarised for ALL in Figure 3.
Except for solvents, all statistically significant
mothers’ self-reported exposures for all diagnostic
subgroups were non-significant for the reviewed
exposure; no new statistically significant exposure
was found.
Restricting analyses to data assessed as ‘good’ did
not show any significant associations for any agent
by any time period or diagnostic group.
DISCUSSION AND CONCLUSION
This paper addresses the relationship between
maternal occupational exposure to eight different
groups of agents and the risk of childhood leukaemia
and lymphoma using interview data from the
UKCCS. It has extended previous analyses by using
all available occupational and exposure details to
assess parental exposure status using a novel approach.
Risks of two different diagnostic subgroups of child-
hood leukaemia (ALL and OTH) and lymphoma
(HL and NHL) have been investigated for three rel-
evant time windows (preconception, pregnancy and
postnatal). The findings are reported in line with the
recent Strengthening the Reporting of Observational
Studies in Epidemiology (STROBE) guidelines(19).
The UKCCS is one of the largest case–control
studies of childhood cancer ever conducted providing
a sample size adequate for investigating environmental
Table 3. OR and 95% CI for ‘reviewed’ exposed vs. unexposed maternal jobs in the three time windows by diagnostic group.
Exposure agent Diagnostic subgroup ORa (95% CI)
Preconception Pregnancy Postnatal
Solvents ALL 1.4 (0.9–1.9) 2.7 (1.6–4.6) 1.9 (1.1–3.3)
OTH 0.9 (0.4–2.1) 2.0 (0.7–5.8) 1.7 (0.7–4.3)
HL 0.5 (0.1–2.5) — 2.2 (0.7–7.3)
NHL 2.2 (1.2–4.0) 1.5 (0.5–4.5) 0.8 (0.2–3.0)
Paints ALL 1.1 (0.7–1.6) 0.8 (0.4–1.7) 1.2 (0.6–2.7)
OTH 1.5 (0.5–4.1) 2.1 (0.3–12.7) 0.6 (0.1–3.7)
HL 0.6 (0.1–3.8) 2.2 (0.3–15.4) 2.7 (0.4–8.9)
NHL 1.4 (0.4–5.0) 1.6 (0.4–7.4) 1.2 (0.2–6.6)
Dyes ALL 1.0 (0.7–1.5) 0.9 (0.6–1.7) 0.9 (0.5–1.8)
OTH 1.1 (0.5–2.2) 0.9 (0.2–4.1) 1.4 (0.4–4.6)
HL 1.4 (0.4–4.8) 0.9 (0.1–6.2) —
NHL 0.8 (0.3–1.9) 0.8 (0.2–2.9) 0.8 (0.1–4.7)
Petrol ALL 0.8 (0.4–1.4) 0.9 (0.2–3.3) 2.4 (0.8–6.9)
OTH 0.7 (0.2–3.4) 1.8 (0.2–14.6) 3.4 (0.5–25.6)
HL 0.6 (0.1–4.9) — —
NHL 1.1 (0.4–3.2) — 3.0 (0.6–14.5)
Lead ALL 0.9 (0.6–1.8) 0.5 (0.2–1.5) 0.3 (0.1–1.3)
OTH — — —
HL 1.2 (0.2–5.6) — —
NHL 0.7 (0.2–2.7) 0.7 (0.1–7.2) 1.3 (0.2–8.1)
Fertilisers ALL 0.6 (0.2–1.6) 1.9 (0.5–7.4) 2.0 (0.7–5.8)
OTH 1.8 (0.5–6.5) 0.9 (0.1–8.5) 2.5 (0.7–8.7)
HL 3.5 (0.6–21.2) — —
NHL 2.1 (0.6–7.7) 0.9 (0.1–10.5) 5.1 (0.9–27.8)
Radiation ALL 0.9 (0.5–1.5) 0.8 (0.4–1.8) 1.3 (0.6–2.9)
OTH 1.0 (0.5–2.2) 1.8 (0.5–6.4) 2.3 (0.5–9.5)
HL — — 2.3 (0.3–15.6)
NHL 0.3 (0.1–1.3) 0.5 (0.1–3.8) 0.8 (0.1–5.9)
Wood dust ALL 1.5 (0.6–3.8) 1.3 (0.2–7.9) 1.6 (0.3–7.5)
OTH — — 2.6 (0.5–12.9)
HL — — —
NHL 1.4 (0.3–7.8) — —
aORs computed using unconditional logistic regression adjusting for age, sex, region, socioeconomic status and other
occupational exposure, and all controls as comparison group for each diagnostic subgroup.
MATERNAL OCCUPATION AND CHILDHOOD LEUKAEMIA
Page 5 of 9
End of preview.
