IEA

International Journal of Epidemiology, 2023, 817–826

https://doi.org/10.1093/ije/dyac238
Advance Access Publication Date: 18 January 2023
International Epidemiological Association
Original article

Genetic Epidemiology

Lifestyle, genetic risk and incidence of cancer:

a prospective cohort study of 13 cancer types
Stephanie Byrne ,1,2,3,* Terry Boyle,1,2,3 Muktar Ahmed,1,3,4,5
Sang Hong Lee,1,2,3 Beben Benyamin1,2,3 and Elina Hyppönen 1,3,4

1
Australian Centre for Precision Health, University of South Australia, Adelaide, SA, Australia, 2UniSA
Allied Health & Human Performance, University of South Australia, Adelaide, SA, Australia, 3South
Australian Health and Medical Research Institute, Adelaide, SA, Australia, 4UniSA Clinical and Health
Sciences, University of South Australia, Adelaide, SA, Australia and 5Department of Epidemiology,
Faculty of Public Health, Jimma University Institute of Health, Jimma, Ethiopia
*Corresponding author. Australian Centre for Precision Health, University of South Australia, South Australia, Australia;
South Australian Health and Medical Research Institute, Level 8, North Terrace, Adelaide, SA 5000, Australia. E-mail:
[email protected]
Received 7 December 2021; Editorial decision 29 November 2022; Accepted 20 December 2022

Abstract
Background: Genetic and lifestyle factors are associated with cancer risk. We investi-
gated the benefits of adhering to lifestyle advice by the World Cancer Research Fund
(WCRF) with the risk of 13 types of cancer and whether these associations differ accord-
ing to genetic risk using data from the UK Biobank.
Methods: In 2006–2010, participants aged 37–73 years had their lifestyle assessed and
were followed up for incident cancers until 2015–2019. Analyses were restricted to those
of White European ancestry with no prior history of malignant cancer (n ¼ 195 822).
Polygenic risk scores (PRSs) were computed for 13 cancer types and these cancers
combined (‘overall cancer’), and a lifestyle index was calculated from WCRF recommen-
dations. Associations with cancer incidence were estimated using Cox regression,
adjusting for relevant confounders. Additive and multiplicative interactions between life-
style index and PRSs were assessed.
Results: There were 15 240 incident cancers during the 1 926 987 person-years of follow-
up (median follow-up ¼ 10.2 years). After adjusting for confounders, the lifestyle index
was associated with a lower risk of overall cancer [hazard ratio per standard deviation
increase (95% CI) ¼ 0.89 (0.87, 0.90)] and of eight specific cancer types. There was no evi-
dence of interactions on the multiplicative scale. There was evidence of additive interac-
tions in risks for colorectal, breast, pancreatic, lung and bladder cancers, such that the
recommended lifestyle was associated with greater change in absolute risk for persons
at higher genetic risk (P < 0.0003 for all).
Conclusions: The recommended lifestyle has beneficial associations with most cancers.
In terms of absolute risk, the protective association is greater for higher genetic risk

C The Author(s) 2023. Published by Oxford University Press on behalf of the International Epidemiological Association.
V 817
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/),
which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
818 International Journal of Epidemiology, 2023, Vol. 52, No. 3

groups for some cancers. These findings have important implications for persons most
genetically predisposed to those cancers and for targeted strategies for cancer
prevention.

Key words: Cancer, lifestyle, genetic risk, diet, obesity, physical activity, smoking, alcohol consumption

Key Messages

• Common genetic variations captured by polygenic risk scores (PRSs) are associated with the risk of multiple cancers,

providing opportunities for the identification of high-risk groups.

• We observed positive additive interactions in the associations between an index of recommended lifestyle and PRSs,

consistent with some lifestyles being more beneficially associated with the risk of colorectal, breast, pancreatic, lung
and bladder cancers in those at higher genetic risk.
• If the recommended lifestyle factors are causally related to risk, our data suggest that screening for genetic

predisposition followed by lifestyle intervention might provide a viable strategy for targeted risk mitigation for some
cancers.

Introduction genetic risk.13,14 However, these associations have not

Cancer is the second leading cause of death worldwide.1 been explored for other cancer types. In this study, we in-
Genetic and lifestyle factors play an important role in the vestigate whether adhering to lifestyle cancer prevention
aetiology of cancer. While the heritability of cancer overall recommendations by the World Cancer Research Fund
has been estimated to be 33%,2 individual genetic variants (WCRF) is associated with the risk of 13 different types of
typically have little impact. However, when assessed col- cancer: bladder, breast, colorectal, kidney, lymphocytic
lectively using a polygenic risk score (PRS), a greater num- leukaemia, lung, melanoma, non-Hodgkin’s lymphoma
ber of these genetic variants can substantially increase the (NHL), oral cavity/pharyngeal, ovarian, pancreatic, pros-
likelihood of developing some cancers. Indeed, a high ge- tate and uterine cancer. Our aim was to further determine
netic risk (top 20%) has been reported to account for up to whether the relationships between lifestyle and cancer dif-
30% of cases, although this varies by cancer type.3,4 fer by genetic risk (multiplicative interaction, i.e. whether
An estimated 30–50% of all cancer cases could be pre- the combined effect of lifestyle and genetic risk is different
vented through lifestyle changes, such as eating more fruit, to the product of their individual effects on the relative risk
vegetables and wholegrains, and less red and processed scale), and whether a particular population group may
meat; being physically active; maintaining a normal body- benefit more from lifestyle intervention programmes (addi-
weight; and avoiding tobacco and alcohol.1 These lifestyle tive interaction, i.e. the combined effect of lifestyle and ge-
factors often coexist, creating issues in estimating indepen- netic risk on the risk of cancer is different to the sum of
dent relationships with disease outcomes. Thus, research their individual effects on the absolute risk scale).15 This
has assessed lifestyle factors collectively in a lifestyle knowledge will inform cancer-prevention intervention
score.5 While there is strong evidence that a lifestyle that strategies.
adheres to these recommendations reduces the risk of colo-
rectal and breast cancer, the evidence for other cancer Methods
types is less clear.6–8
Recent evidence suggests that lifestyle may help reduce Study design and study population
cancer risk, even among those with a higher genetic risk of This prospective cohort study utilizes data from the UK
cancer.9 The strongest evidence has been obtained from Biobank, which is a large population-based cohort of
studies on breast and colorectal cancer, which suggest life- >500 000 adults aged 37–73 years and living in the UK at
style associations persist regardless of genetic risk,10–12 and the time of the initial assessment in 2006–2010.16
adherence to lifestyle recommendations for cancer preven- Participants attended 1 of 22 assessment centres across
tion may even be of greater benefit for those at higher England, Scotland and Wales, and completed an
International Journal of Epidemiology, 2023, Vol. 52, No. 3 819

assessment including a questionnaire, anthropometric ancestries and we included participants with no history of
measures and collection of biological samples. UK Biobank any malignant cancer at baseline and sufficient data avail-
was approved by the National Health Service North West able for the lifestyle factors and other covariates of interest
Multi-centre Research Ethics Committee (11/NW/0382), (Figure 1).
the National Information Governance Board for Health
and Social Care in England and Wales, and the
Community Health Index Advisory Group in Scotland. All PRSs
participants provided written informed consent. We calculated PRSs for each participant for 13 cancer
To reduce the effects of population stratification, our types based on single-nucleotide polymorphisms (SNPs) se-
analyses were restricted to participants of White European lected by Graff et al.3 Briefly, they identified a set of

Figure 1 Ascertainment of the study sample included in the analyses

820 International Journal of Epidemiology, 2023, Vol. 52, No. 3

independent SNPs associated with cancer risk for each can- Cancer outcome ascertainment
cer type from previously published cancer genome-wide as- Malignant cancer diagnoses for each cancer type were
sociation studies. SNP selection was restricted to those identified using the 9th and 10th revisions of the
that are common and available in the UK Biobank International Classification of Diseases (ICD-9 and ICD-
data. SNPs with the strongest associations with the 10) codes, phenotype and tumour behaviour information
broadest phenotype were preferentially selected, with an obtained through linkage to national cancer registries in
overall linkage disequilibrium r2 < 0.3 to ensure indepen- England, Wales and Scotland. Incident cancer events were
dence (see Supplementary Table S1, available as those diagnosed after baseline assessment. Cancer informa-
Supplementary data at IJE online, for the number of SNPs tion was complete up to 31 July 2019 for England and
used in the PRS calculations; see Graff et al. for a full list Wales, and 31 October 2015 for Scotland. Hence, follow-
of genetic variants3). up time was defined as being from the date of baseline as-
We calculated PRSs as the sum of the number of risk sessment to the earliest of the date of diagnosis, date of
alleles multiplied by the log of the odds ratio (OR) for each death (obtained via linkage to death registries) or
SNP, as implemented in PLINK software17 using the fol- 31 July 2019 for England and Wales residents and
lowing formula: 31 October 2015 for Scotland residents.

X
n
PRSi ¼ xil logðORl Þ
Statistical analysis
l¼1
Baseline characteristics were summarized for the total
where n is the number of SNPs in the set of independent study population, those diagnosed with any of the 13
SNPs for that cancer type, xil is the number of risk alleles cancers examined (referred to as ‘overall cancer’) and
(0, 1 or 2) for the i-th individual at the l-th SNP and ORl is individually for each cancer type. Characteristics were
the estimated OR for the l-th SNP using a logistic summarized as percentages for all variables for greater
regression. interpretability.
Each PRS was z-standardized, with higher scores repre- Cox proportional hazard regression models were used
senting a greater genetic risk. A PRS for overall cancer was to investigate associations between PRS, lifestyle index and
also derived as the sum of the z-standardized PRSs for incidence of overall cancer, and separately for each cancer.
the 13 cancer types, each weighted according to the distri- PRSs and lifestyle index were used as continuous variables,
bution of incident cases of these cancers reported by standardized to estimate hazard ratio per SD change.
Cancer Research UK in 2017 (https://www.cancerre Incidences of cancers that are sex-specific were assessed in
searchuk.org/health-professional/cancer-statistics/incidence# the relevant sex only (female cancers—ovarian, uterine,
heading-Four). breast; male cancers—prostate). Breast cancer analyses
were further restricted to post-menopausal women only
(self-reported at baseline). All models were adjusted for
Lifestyle index
age at baseline (continuous); sex (where relevant); assess-
A lifestyle index was constructed based on the WCRF/ ment centre; socio-economic status (Townsend Index—
American Institute for Cancer Research (WCRF/AICR) continuous); education [none listed, General Certificate of
cancer prevention recommendations and standardized Secondary Education (GCSE)/Certificate of Secondary
scoring system (full details of the calculation of the lifestyle Education (CSE) or equivalent, A-levels or equivalent, col-
index are in Supplementary Table S2, available as lege/university or other professional training]; household
Supplementary data at IJE online).5 Baseline anthropomet- income (<18 000, 18 000–30 999, 31 000–51 999, 52 000–
ric and touchscreen questionnaire information was avail- 100 000, >100 000); birth location (north south coordi-
able for five of the eight recommendations included in the nate and east west coordinate—both in deciles) and popu-
2018 WCRF/AICR score (normal weight; physical activity; lation stratification measured by the first 40 principal
wholegrain, fruit and vegetable intake; red and processed components. Models with and without mutual adjustment
meat intake; alcohol consumption). Smoking status (cate- for lifestyle index/PRS were examined. Subsequent analy-
gorized as never, former or current smoking) was also in- ses adjusting for additional covariates specific to each can-
cluded in the lifestyle index, given its relevance as a cer type were also performed (see Supplementary Table S3,
modifiable lifestyle risk factor for multiple cancers. The available as Supplementary data at IJE online, for a list of
lifestyle index ranged from 0 to 6, with higher scores indi- the additional covariates and definitions). Additional anal-
cating greater adherence to the recommendations. yses exploring the risk of cancer in the top 5% of PRSs
International Journal of Epidemiology, 2023, Vol. 52, No. 3 821

were conducted using Cox regression with the bottom ter- After adjusting for confounders, a higher lifestyle index
tile of PRSs as the reference and adjusting for the same was associated with a lower risk of overall cancer [hazard
covariates. ratio (HR) 95% CI per SD increase in lifestyle index ¼ 0.89
For each cancer outcome, we investigated multiplicative (0.87, 0.90)], colorectal cancer [0.85 (0.82, 0.89)], post-
interactions between lifestyle and genetic risk by using a menopausal breast cancer [0.85 (0.81, 0.89)], lung cancer
likelihood ratio test to compare Cox models with and [0.57 (0.54, 0.61)], kidney cancer [0.78 (0.72, 0.85)], uter-
without an interaction term between PRS and lifestyle in- ine cancer [0.82 (0.74, 0.89)], pancreatic cancer [0.79
dex. Meanwhile, additive interactions were assessed by us- (0.72, 0.87)], bladder cancer [0.72 (0.65, 0.79)] and oral
ing the relative excess risk due to interaction (RERI), cavity/pharyngeal cancer [0.78 (0.71, 0.86)] (Figure 2).
which was calculated from Cox models with the interac- There was no association between the lifestyle index and
tion term.15 For these models, the lifestyle index and PRSs melanoma, NHL, ovarian cancer and lymphocytic leukae-
were categorized into tertiles for greater interpretability mia. An association in the opposite direction was observed
[PRS—lower (lowest tertile), intermediate (middle tertile) between the lifestyle index and prostate cancer risk [1.04
and higher (highest tertile) genetic risk; lifestyle index— (1.01, 1.08)] (Figure 2). Estimates remained unchanged
lower (range 0–3), intermediate (range 3.25–3.75) and upon adjustment for PRS (Supplementary Table S4, avail-
higher (range 4–6) lifestyle index, representing lower, in- able as Supplementary data at IJE online). Further adjust-
termediate and higher adherence to lifestyle recommenda- ment for other potential confounders specific to cancer
tions]. Observed interactions were further explored type had little effect on the estimates presented
visually by using a forest plot produced from a multivari- (Supplementary Table S4, available as Supplementary data
able Cox proportional hazards model with a combined ge- at IJE online).
netic and lifestyle risk variable (nine categories with lower A higher PRS was associated with a higher risk of over-
genetic risk and higher lifestyle index as the reference). As all cancer and all cancer types assessed, except oral cavity/
smoking is a strong risk factor for lung cancer, we per- pharyngeal and ovarian cancers (Figure 2). The cancers
formed sensitivity analyses removing smoking from the with the highest risk with increasing PRS were prostate
lifestyle index and including it as a confounder for this out- cancer [1.36 (1.32, 1.40)], colorectal cancer [1.36 (1.31,
come. Proportionality of hazards assumptions were veri- 1.42)], melanoma [1.39 (1.32, 1.47)], pancreatic cancer
fied by using Schoenfeld residuals. Two-sided P-values of [1.51 (1.36, 1.66)] and lymphocytic leukaemia [1.50 (1.33,
0.05 were considered as evidence of an association. A 1.69)]. Again, addition of the lifestyle index as a covariate
P-value threshold of 0.0036 (calculated as 0.05/number of had no effect on the estimates and further adjustment
cancer outcomes) was applied to interaction models to ac- for other cancer-specific potential confounders also had
count for multiple testing. All analyses were performed us- little impact (Supplementary Table S4, available as
ing Stata SE version 16 (StataCorp, College Station, TX). Supplementary data at IJE online). Further exploration of
Results are presented in order of power determined from genetic risk found that those in the top 5% of PRS for pros-
the number of cases for each cancer type. tate, colorectal, melanoma and pancreatic cancers had an
estimated 2.5-, 2.5-, 3- and 4-fold higher risk, respectively
(Table 2).
Results There were no multiplicative interactions observed be-
A summary of the baseline characteristics of the 195 822 tween lifestyle index and PRS for any of the cancer out-
participants included in the analysis and for those diag- comes (P > 0.03 for all outcomes after correction for
nosed with any of the 13 types of cancer during the follow- multiple testing). However, some evidence for an additive
up period are provided in Table 1. Over the 1 926 987 interaction was observed for colorectal, breast, lung, pan-
person-years of follow-up [median (interquartile range) creatic and bladder cancers (Supplementary Table S5,
length of follow-up ¼ 10.2 (9.4–10.9) years], there was a available as Supplementary data at IJE online). Forest plots
total of 15 240 incident cases of the 13 cancer types of in- of risk with a combined lifestyle/genetic risk variable were
terest. The cancers with the highest number of incident constructed to visually depict these interactions. These
cases were prostate cancer, colorectal cancer and post- plots show a greater change in absolute risk across lifestyle
menopausal breast cancer. Generally, risk of cancer was tertiles in individuals with a higher genetic risk of these
higher in men, those of older age and those who have cancers (Figure 3). For lung and bladder cancers, there
lower levels of education and income. Lower adherence to were no differences in cancer risk by PRS among persons
lifestyle recommendations and a higher genetic predisposi- with a higher lifestyle index.
tion were also associated with a higher incidence of most, We further investigated the association between lifestyle
but not all, cancers. and lung cancer by removing smoking status from the
 822
Table 1 Baseline characteristics of the total study population and of incident cases of each cancer type

Characteristic Total Overall Prostate Colorectal Post- Lung Melanoma Non- Kidney Uterine Pancreatic Bladder Oral cavity/ Ovarian Lymphocytic
study cancer cancer cancer menopausal cancer Hodgkin’s cancer cancer cancer cancer pharyngeal cancer leukaemia
population breast lymphoma cancer
cancer

N 195 822 15 240 4476 2150 1990 1256 1200 721 547 494 451 424 415 315 246
Age at baseline
assessment (%)a
50 years 27.0 11.5 4.8 9.8 2.1 5.3 17.5 9.2 8.4 11.1 6.7 3.3 13.7 20.0 5.7
51–60 years 36.7 33.1 29.4 32.5 45.5 30.3 33.5 32.3 33.8 41.9 29.7 23.8 40.5 30.8 31.7
61 years 36.4 55.4 65.8 57.7 52.4 64.4 49.0 58.5 57.8 47.0 63.6 72.9 45.8 49.2 62.6
Female (%) 49.9 43.5 0 38.8 100 44.2 45.7 41.3 33.8 100 41.0 17.5 29.2 100 29.7
Townsend Index (%)a
Quintile 1 22.5 22.6 24.6 22.8 22.5 15.2 26.0 25.4 24.3 17.0 22.4 21.0 16.4 20.6 24.0
Quintile 2 21.6 22.3 23.2 23.7 21.8 19.2 25.0 19.7 21.0 23.9 23.1 24.5 22.7 24.8 17.5
Quintile 3 20.7 20.6 21.6 19.7 20.5 17.8 19.9 21.8 24.5 19.6 22.6 19.6 17.4 19.1 26.8
Quintile 4 19.4 18.9 18.1 18.0 20.9 20.1 17.5 18.9 15.7 21.1 15.5 18.2 21.5 20.6 17.5
Quintile 5 15.8 15.6 12.6 15.8 14.4 27.7 11.6 14.3 14.4 18.4 16.4 16.8 22.2 14.9 14.2
Education (%)a
None listed 12.6 16.3 15.5 15.7 15.1 30.0 13.2 18.5 18.1 13.4 19.3 21.0 19.0 15.6 20.7
GCSE/CSE or equivalent 27.1 25.5 21.6 26.3 29.2 25.1 27.7 22.2 28.0 31.4 22.4 27.1 26.3 31.4 24.8
A-levels or equivalent 12.3 11.3 10.5 11.8 12.2 10.0 10.9 10.8 11.9 10.5 9.5 8.5 12.1 11.4 8.9
College/university or other 48.0 47.0 52.4 46.2 43.6 35.0 48.3 48.5 42.1 44.7 48.8 43.4 42.7 41.6 45.5

International Journal of Epidemiology, 2023, Vol. 52, No. 3

professional training
Average total household
income before tax (%)a
<£18 000 19.4 23.8 20.4 23.8 25.7 41.2 18.3 22.6 24.7 25.3 27.7 29.0 28.9 27.9 30.1
£18 000 –30 999 25.2 28.4 27.7 29.7 31.7 30.1 23.8 32.6 29.8 32.0 27.3 30.7 25.8 31.1 26.4
£31 000–51 999 27.2 25.6 26.8 25.6 24.3 18.4 27.8 26.9 25.2 24.3 25.3 22.6 25.1 24.4 26.0
£52 000–100 000 22.4 17.4 19.6 16.0 15.2 8.1 24.3 13.5 16.3 15.2 14.6 13.9 16.6 12.4 15.0
>£100 000 5.9 4.8 5.5 4.9 3.1 2.2 5.8 4.4 4.0 3.2 5.1 3.8 3.6 4.1 2.4
Lifestyle index (%)a
Lower 36.6 42.0 41.0 44.9 33.6 63.1 34.8 38.0 46.6 38.3 48.3 55.9 50.6 30.5 38.2
Intermediate 32.4 32.0 32.2 30.6 36.9 23.5 35.5 31.9 35.1 34.0 32.2 25.9 28.9 35.9 33.3
Higher 31.0 26.0 26.8 24.5 29.6 13.4 29.8 30.1 18.3 27.7 19.5 18.2 20.5 33.7 28.5
PRS (%)a
Lower genetic risk – 30.5 24.3 22.4 24.9 25.5 23.0 27.7 27.8 22.3 20.6 24.3 32.8 26.7 22.0
Intermediate genetic risk – 33.3 31.2 32.9 33.1 32.9 31.2 33.3 32.2 36.2 33.3 36.6 33.0 34.0 24.8
Higher genetic risk – 36.3 44.5 44.7 42.0 41.6 45.8 39.0 40.0 41.5 46.1 39.2 34.2 39.4 53.3

Overall cancer ¼ overall incident cases of the 13 cancer types assessed in this study.
GCSE, General Certificate of Secondary Education; CSE, Certificate of Secondary Education; PRS, polygenic risk score.
a
Percentages may not add up to 100% due to rounding.
International Journal of Epidemiology, 2023, Vol. 52, No. 3 823

Figure 2 Adjusted associations between the lifestyle index, polygenic risk score and cancer. Cox proportional hazards regression adjusted for age at
baseline, sex (where relevant), assessment centre, 40 principal components of ancestries, Townsend Index, education, birth location and income;
overall cancer ¼ overall incident cases of the 13 cancer types assessed in this study

lifestyle index and including it as a confounder. Although adherence to lifestyle recommendations. Our findings are
strongly attenuated, a higher lifestyle index was associated consistent with the limited comparable research for
with a lower risk of lung cancer also when smoking was breast10,13 and colorectal cancers,11,12,14 and are novel for
excluded [HR (95%CI) per SD increase ¼ 0.91 (0.86, the other cancer types examined here. Communicating
0.97), P ¼ 0.0016; lifestyle index with smoking included these results to groups with a higher genetic risk of cancer
HR ¼ 0.57 (0.54, 0.61)]. The additive interaction between may help alleviate any distress experienced due to aware-
lifestyle index and PRS was attenuated by excluding infor- ness of their increased risk and is likely to be empowering
mation on smoking status [lower lifestyle index/higher ge- and potentially supportive of positive behavioural changes.
netic risk vs higher lifestyle index/lower genetic risk A higher genetic risk of cancer was associated with an
RERI ¼ 0.44 (0.04, 0.85) P ¼ 0.033; Ptrend ¼ 0.021]. increased risk of overall cancer and 11 cancer types: pros-
tate, colorectal, breast (post-menopause), lung, melanoma,
NHL, kidney, uterine, pancreatic, bladder and lympho-
Discussion cytic leukaemia. The highest genetic risk was observed for
In this study, adhering to lifestyle advice by the WCRF was pancreatic cancer, with those in the top 5% of PRS having
associated with a lower risk of several cancers and a higher a 4-fold higher risk compared with those in the bottom ter-
genetic risk was associated with an increased risk. These tile of genetic risk. This increase is similar in magnitude to
associations were independent of each other. There were BRCA1 and BRCA2 gene variants and breast cancer risk,
no multiplicative interactions between lifestyle and genetic which are far less common (prevalence of 0.2–0.3%) and
risk on the relative risk scale, while we observed some evi- trigger more frequent cancer screening.18 Pancreatic cancer
dence for risk differences (additive interaction) suggesting is one of the leading causes of cancer death that is typically
that adherence to lifestyle recommendations for cancer diagnosed at a late stage when the 5-year survival rate is
prevention may be of greater benefit in those with a higher <10%.19 Screening for pancreatic cancer is recommended
genetic susceptibility to colorectal, breast and pancreatic for individuals deemed as high-risk20 and our results sug-
cancers. We found that risks of lung and bladder cancers gest a PRS could be used as an additional tool to assist in
do not differ by genetic risk among individuals with higher the identification of persons at high risk. We also found
824 International Journal of Epidemiology, 2023, Vol. 52, No. 3

Table 2 Risk of cancer for those in the top 5% of polygenic reduction in cancer incidence. Further research is needed to
risk scores establish a causal relationship; however, our findings sup-
port the current public health message that not smoking;
Cancer type Top 5% of polygenic risk scores
avoiding alcohol; consuming a diet rich in wholegrains,
Number HR 95% CIa P-value fruit and vegetables, and low in red and processed meat;
of cases maintaining a normal bodyweight; and engaging in regular
physical activity reduce cancer risk.
Overall cancer 896 1.33 (1.24, 1.43) <0.0001
Prostate cancer 383 2.53 (2.25, 2.84) <0.0001 We found no association between the lifestyle recom-
Colorectal cancer 180 2.52 (2.13, 3.00) <0.0001 mendations and risk of melanoma, NHL, ovarian cancer
Post-menopausal 225 1.92 (1.59, 2.32) <0.0001 and lymphocytic leukaemia. These results were not unex-
breast cancer pected; none of the lifestyle factors included in the lifestyle
Lung cancer 104 2.22 (1.78, 2.77) <0.0001 index has been conclusively linked with the risk of lympho-
Melanoma 121 3.01 (2.43, 3.74) <0.0001 cytic leukaemia, melanoma or NHL and only obesity is
Non-Hodgkin’s 61 2.06 (1.54, 2.74) <0.0001
positively associated with ovarian cancer risk.8 For mela-
lymphoma
noma, although evidence indicates that higher alcohol in-
Kidney cancer 28 1.25 (0.83, 1.87) 0.28
Uterine cancer 35 2.22 (1.52, 3.25) <0.0001 take may increase risk, greater physical activity levels have
Pancreatic cancer 57 4.14 (2.97, 5.76) <0.0001 also been associated with increased risk, which is likely re-
Bladder cancer 31 1.99 (1.33, 2.98) 0.0008 lated to higher sun exposure levels.21,22
Oral cavity/pharyngeal 26 1.31 (0.86, 1.99) 0.21 We found that higher adherence to lifestyle recommen-
cancer dations was associated with a higher risk of prostate can-
Ovarian cancer 20 1.55 (0.95, 2.53) 0.078
cer. This finding is inconsistent with the majority of
Lymphocytic 24 3.00 (1.85, 4.85) <0.0001
previous research suggesting either a negative association
leukaemia
or no association for combined or individual components
a
Reference category ¼ bottom tertile of polygenic risk scores. of the lifestyle index6,23–27 and may be the result of healthy
All models adjusted for lifestyle index category, age at baseline, sex, assess-
volunteer bias.
ment centre, 40 principal components of ancestries, Townsend Index, educa-
tion, birth location and income. This large prospective cohort study has some limita-
tions. The length of the follow-up and age range of the
study population have limited the number of incident can-
that persons with lower adherence to lifestyle recommen- cer cases. Consequently, some statistical models may not
dations had an increased risk of pancreatic cancer and have been adequately powered to observe a modifying ef-
there was a positive additive interaction on the risk differ- fect of genetic risk on lifestyle–cancer associations. There
ence scale, suggesting the combined association of lower are likely to be measurement errors in the components of
adherence to lifestyle recommendations and higher genetic the lifestyle score as they are almost all measured via self-
risk on the risk of pancreatic cancer is more than the sum report; however, this is likely to attenuate the findings to
of the individual associations. the null. We also cannot rule out the possibility of residual
Lower adherence to lifestyle recommendations was as- confounding, although we adjusted for multiple covariates.
sociated with a higher risk of overall cancer and eight can- The measure of genetic risk was limited by the SNPs in-
cer types: colorectal, post-menopausal breast, lung, kidney, cluded in the PRS, which may not be exhaustive. Lastly,
uterine, pancreatic, bladder and oral cavity/pharyngeal this study was conducted in adults of European ancestries,
cancers. These findings are consistent with previously pub- so the relevance of these findings to populations of other
lished research for breast and colorectal cancers, and add ethnicities is unclear.
to the limited and inconclusive research on overall lifestyle In summary, our findings indicate that adhering to the
and other cancers.6,7 In addition to pancreatic cancer, we WCRF/AICR lifestyle recommendations is associated with
also observed additive interactions for colorectal, breast, a lower cancer risk and, for some specific cancer types,
lung and bladder cancers. If these lifestyle recommenda- greater adherence to lifestyle recommendations may be
tions are causally related to risk of these cancers, our find- more beneficial among population groups with a higher ge-
ings suggest that lifestyle changes in line with WCRF/AICR netic susceptibility to that cancer type. Our findings also
lifestyle recommendations may result in greater reductions highlight the potential benefit of using a PRS to identify
in absolute risk of these cancers in high genetic risk popula- those with a higher genetic risk of pancreatic cancer, and
tions compared with low genetic risk groups and, where possibly other cancers, as PRSs are further refined. PRSs
resources are limited, lifestyle interventions should target could be used in conjunction with other tools to increase
these high genetic risk groups to maximize the potential the accuracy of identifying high-risk individuals who may
International Journal of Epidemiology, 2023, Vol. 52, No. 3 825

Figure 3 Risk of colorectal, post-menopausal breast, lung, pancreatic and bladder cancers according to genetic risk and lifestyle. Cox proportional
hazards regression adjusted for age at baseline, sex (where relevant), assessment centre, 40 principal components of ancestries, Townsend Index,
education, birth location and income; LI, lifestyle index; pyar, person-years at risk; HR, hazard ratio

then undergo regular screening and be targeted by preven- Supplementary data

tion intervention strategies. This prospect requires further Supplementary data are available at IJE online.
investigation using cohort studies conducted across various
ethnic populations.
Author contributions
Data acquisition: S.H.L., E.H. Concept and design: S.B., T.B., B.B.,
Ethics approval
E.H. Data analysis: S.B., T.B., S.H.L., B.B. Data management: S.B.,
This study is a secondary analysis of UK Biobank data. The UK M.A., B.B. Drafting of the manuscript: S.B. Obtaining funding:
Biobank cohort study was approved by the National Health Service T.B., B.B., E.H. All authors interpreted the results, revised the paper,
North West Multi-centre Research Ethics Committee (11/NW/
and reviewed and approved the final version. S.B. and T.B. had full
0382), the National Information Governance Board for Health and
access to all the data and take full responsibility for the integrity of
Social Care in England and Wales, and the Community Health
the data and the accuracy of the analysis.
Index Advisory Group in Scotland. All participants provided written
informed consent.

Funding
Data availability This work was supported by Tour de Cure (RSP-013–18/19) and the
This research utilizes data from the UK Biobank resource (applica- Australian Government’s Medical Research Future Fund (MRFF) as
tion number 20175), which is available directly from UK Biobank part of the Rapid Applied Research Translation programme. This
upon submission of a data request proposal. See https://www.ukbio project is part of the work being undertaken by Health Translation
bank.ac.uk. SA (MRF9100005).
826 International Journal of Epidemiology, 2023, Vol. 52, No. 3

Conflict of interest invasive breast cancer among women in the UK biobank. J Natl
Cancer Inst 2020;112:893–901.
None declared.
14. Choi J, Jia G, Wen W, Shu XO, Zheng W. Healthy lifestyles, ge-
netic modifiers, and colorectal cancer risk: a prospective cohort
References
study in the UK Biobank. Am J Clin Nutr 2021;113:810–20.
1. World Health Organisation (WHO). Cancer. Geneva: WHO, 15. VanderWeele TJ, Knol MJ. A tutorial on interaction. Epidemiol
2021. https://www.who.int/health-topics/cancer (18 November Methods 2014;3:33–72.
2021, date last accessed). 16. Sudlow C, Gallacher J, Allen N et al. UK Biobank: an open ac-
2. Mucci LA, Hjelmborg JB, Harris JR et al.; Nordic Twin Study of cess resource for identifying the causes of a wide range of com-
Cancer (NorTwinCan) Collaboration. Familial risk and herita- plex diseases of middle and old age. PLoS Med 2015;12:
bility of cancer among twins in Nordic countries. JAMA 2016; e1001779.
315:68–76. 17. Purcell S, Neale B, Todd-Brown K et al. PLINK: a tool set for
3. Graff R, Cavazos T, Thai K et al. Cross-cancer evaluation of whole-genome association and population-based linkage analy-
polygenic risk scores for 16 cancer types in two large cohorts. ses. Am J Hum Genet 2007;81:559–75.
Nat Commun 2021;12:970. 18. National Cancer Institute (NCI). BRCA Gene Mutations:
4. Kachuri L, Graff RE, Smith-Byrne K et al. Pan-cancer analysis Cancer Risk and Genetic Testing. Maryland: NCI, 2020. https://
demonstrates that integrating polygenic risk scores with modifi- www.cancer.gov/about-cancer/causes-prevention/genetics/brca-
able risk factors improves risk prediction. Nat Commun 2020; fact-sheet (3 June 2021, date last accessed).
11:6084. 19. Klein AP. Pancreatic cancer epidemiology: understanding the
5. Shams-White MM, Brockton NT, Mitrou P et al. role of lifestyle and inherited risk factors. Nat Rev Gastroenterol
Operationalizing the 2018 World Cancer Research Fund/ Hepatol 2021;18:493–502.
American Institute for Cancer Research (WCRF/AICR) cancer 20. Goggins M, Overbeek KA, Brand R et al.; International Cancer
prevention recommendations: a standardized scoring system.
of the Pancreas Screening (CAPS) Consortium. Management of
Nutrients 2019;11:1572.
patients with increased risk for familial pancreatic cancer:
6. Solans M, Chan DSM, Mitrou P, Norat T, Romaguera D. A sys-
updated recommendations from the International Cancer of the
tematic review and meta-analysis of the 2007 WCRF/AICR score
Pancreas Screening (CAPS) Consortium. Gut 2020;69:7–17.
in relation to cancer-related health outcomes. Ann Oncol 2020;
21. Moore SC, Lee I-M, Weiderpass E et al. Association of leisure-
31:352–68.
time physical activity with risk of 26 types of cancer in 1.44 mil-
7. Zhang YB, Pan XF, Chen J et al. Combined lifestyle factors, inci-
lion adults. JAMA Intern Med 2016;176:816–25.
dent cancer, and cancer mortality: a systematic review and meta-
22. Holman DM, Berkowitz Z, Guy GP, Hartman AM, Perna FM.
analysis of prospective cohort studies. Br J Cancer 2020;122:
The association between demographic and behavioral character-
1085–93.
istics and sunburn among U.S. adults—National Health
8. World Cancer Research Fund/American Institute for Cancer
Interview Survey, 2010. Prev Med 2014;63:6–12.
Research. Diet, Nutrition, Physical Activity and Cancer: a
23. Olmedo-Requena R, Lozano-Lorca M, Salcedo-Bellido I et al.
Global Perspective. Continuous Update Project Expert Report,
Compliance with the 2018 World Cancer Research Fund/
2018. https://www.wcrf.org/diet-and-cancer/ (4 March 2021,
date last accessed). American Institute for Cancer Research Cancer Prevention rec-
9. Zhu M, Wang T, Huang Y et al. Genetic risk for overall cancer ommendations and prostate cancer. Nutrients 2020;12:768.
and the benefit of adherence to a healthy lifestyle. Cancer Res 24. Romaguera D, Gracia-Lavedan E, Molinuevo A et al. Adherence
2021;81:4618–27. to nutrition-based cancer prevention guidelines and breast, pros-
10. Al Ajmi K, Lophatananon A, Mekli K, Ollier W, Muir KR. tate and colorectal cancer risk in the MCC-Spain case-control
Association of nongenetic factors with breast cancer risk in ge- study. Int J Cancer 2017;141:83–93.
netically predisposed groups of women in the UK biobank co- 25. Lavalette C, Adjibade M, Srour B et al. Cancer-specific and gen-
hort. JAMA Netw Open 2020;3:e203760. eral nutritional scores and cancer risk: results from the prospec-
11. Cho YA, Lee J, Oh JH et al. Genetic risk score, combined life- tive NutriNet-Sante cohort. Cancer Res 2018;78:4427–35.
style factors and risk of colorectal cancer. Cancer Res Treat 26. Pernar CH, Ebot EM, Wilson KM, Mucci LA. The epidemiology
2019;51:1033–40. of prostate cancer. Cold Spring Harb Perspect Med 2018;8:
12. Carr PR, Weigl K, Jansen L et al. Healthy lifestyle factors associ- a030361.
ated with lower risk of colorectal cancer irrespective of genetic 27. Schulpen M, van den Brandt PA. Adherence to the
risk. Gastroenterology 2018;155:1805–15.e5. Mediterranean diet and risks of prostate and bladder cancer in
13. Arthur RS, Wang T, Xue X, Kamensky V, Rohan TE. Genetic the Netherlands cohort study. Cancer Epidemiol Biomarkers
factors, adherence to healthy lifestyle behavior, and risk of Prev 2019;28:1480–88.