Discussion
In the present MR study, we systematically assessed the causal
associations of a broad range of lifestyle factors with the risk of OC
by histologic subtypes. We found genetically determined years of
education to be inversely associated with overall OC, and subtypes of
LMSOC, HGSOC, IMOC, LGSOC, and EOC. The associations of education with
risk of all OC, HGSOC, IMOC, and LMSOC were independent of BMI. Genetic
liability to higher coffee or tea consumption was positively associated
with the risk of EOC, which may partly mediated by BMI. Our results
further showed that increased dietary fat intake was positively
associated with HGSOC, but inversely and independently associated with
EOC. Genetic predisposition to a higher BMI was associated with an
increased risk of LMSOC, LMMOC, and EOC. Although little clear evidence
supports the role of smoking initiation in ovarian carcinogenesis,
elevated lifetime smoking index was positively associated with overall
OC, HGSOC, and EOC. In contrast, there was limited evidence for causal
associations of physical activities, alcohol drinking, sleep duration,
and insomnia with any type of OC.
Education level is a concerned social determinant and has been proposed
as a modifiable risk factor for a number of diseases. An understanding
of the causal links from education level to health outcomes may help
disease prevention. A recent observational study reported that the
incidence of borderline ovarian tumors decreased in women with a high
educational level [4]. Our study found that genetically predicted
higher education level was causally and independently associated with
lower risks of many OC subtypes, further suggesting the protective
effect of educational attainment on OC development. Possible pathways
may link education level to OC prevention. Firstly, modifiable risk
factors have been reported largely mediates the educational effects on
diseases. Moreover, high level of education is more likely to drive
positive health-related behaviors and thus reduce the risk of OC.
[38]. The associations between education and OC subtypes were
attenuated in the multivariable MR analysis with adjustment for
genetically predicted BMI and lifetime smoking index liability, which
may suggest that BMI and lifetime smoking index partly mediate the
associations.
Studies have not demonstrated either an overall protective or
detrimental effect on OC development in coffee or tea drinkers. In
previous observational analyses, the RR estimates of coffee or tea
consumption were significantly heterogeneous, with individual studies
suggesting positive [5], [6], [39]–[41], null [8],
[42]–[45], and inverse [7], [45]–[48] associations
with OC risk. Inconsistencies observed in the literature may be due to
the interference of confounders, lack of compatibility of categorical
definitions, and differences in definitions for baseline groups. Our
study is the first to use MR analyses to explore the relationship
between coffee or tea consumption with OC histotypes. In the subtype
analyses, we found evidence for a positive association between
endometrioid OC risk and genetically predicted coffee intake levels.
Similarly, a suggestive increased risk was only observed in the
endometrioid subtype in the tea consumption analyses. The etiology of
endometrioid OC resulting from coffee or tea consumption warrants to be
established. The major endometriosis mechanism in OC is
estrogen-dependent, endometriosis acts as a precursor for OC, which is
easily developed in the low-progesterone and high-estrogen conditions
[49]. The intake of caffeine and caffeine-containing beverages has
been positively associated with estrone and sex hormone-binding globulin
concentrations and inversely with bioavailable testosterone [50],
[51]. These hormonal changes may influence hormone-dependent
diseases. Therefore, caffeine intake has been associated with the risk
of endometriosis and has been hypothesized to exert its effects through
the alteration of endogenous hormone levels [52]. Our findings of an
increased risk for endometrioid carcinoma in coffee or tea drinkers are
consistent with the hormonally mediated hypothesis.
The current knowledge on the association between dietary fat intake and
OC was inconclusive. Pooled analyses of observational studies have
reported positive [9] or null [53] association between fat
intake and overall OC risk, however, there were limited data for the OC
subtypes. Although we found no indication of a causal effect of relative
fat intake on overall OC risk, a persistent borderline association was
observed between fat intake and HGSC, which was corroborated with the
findings that serous ovarian cancer was more susceptible to dietary fat
consumption than other subtypes [53]. Due to the pathological
specificity of endometrioid OC and the composition of dietary fat, our
result of an inverse association between relative fat intake and this
subtype may get support from a prospective study that higher intake of
omega-3 may be protective for OC overall and endometrioid tumors in
particular [54]. Further MR analyses for the effects of dietary fat
components on OC subtypes are needed.
BMI has been associated with OC in observational studies [14],
[55]. However, the effect of BMI increase on the etiology of
histological subtypes of OC remains controversial [56] [57]. In
our analyses with updated GWAS studies including 583 BMI SNPs and 25,509
OC cases, which may afford the analysis greater instrument strength and
greater statistical power to detect effects, we observed positive
associations of genetically predicted BMI with LMSOC, LMMOC, and EOC,
which complemented the findings from previous observational studies
[14]. A high BMI has been linked to benign ovarian tumors, the
evidence from epidemiological, histopathological, and molecular studies
suggests that borderline tumors may develop from benign tumors [58],
[59]. This theory of progression for borderline tumors is supported
by our finding that low malignant potential tumors were associated with
BMI, but not high-grade cancers.
The association between physical activity and OC risk remains less
clear. It has been hypothesized from previous meta-analysis of
observational studies that physical activity may protect against the
development of OC [12], [60], [61], whereas positive or null
associations were found from time to time [11], [13], [62].
Across studies, the findings are inconclusive especially when
considering histologic type [13], [63], indicating that this
issue needs to be resolved. We identified no causal relationship for
overall acceleration average, self-reported MVPA, and VPA, with overall
OC by MR analyses, which supported the findings from three large
prospective studies including up to 96,216 participants showing that
MVPA, VPA, or overall physical activity were not independently
associated with reduced risk of OC [10], [11], [62]. In this
study, we further showed that physical activities do not affect the risk
of OC subtypes, which contributed to the knowledge regarding the
evaluation of lifestyle behaviors on OC histotypes incidence.
Studies of the association of smoking with the risk of OC have not been
entirely consistent. [17], [64], [65]. One reason for the
variation in these results may be the different methods employed to
assess and analyze these measures of smoking exposure. Here we improve
our study by using two sets of instruments for smoking [28]. Our
analyses by OC histological subtypes demonstrated limited associations
between smoking initiation and OC subtypes. However, we found that the
lifetime smoking index is suggestively associated with overall OC,
HGSOC, and EOC subtypes. Although smoking has been strongly linked to
mucinous OC in previously pooled analyses, we didn’t observe a
consistent result. However, we still emphasize that independent GWAS and
large prospective studies by OC subtypes are warranted to further
validate our findings in other cohorts and ethnicities.
Results from epidemiological studies on the association between alcohol
drinking and OC risk are inconsistent, reporting either a null
association [66]–[69], a positive association [70], or a
negative association [71], [72]. A recent study focused on
overall OC and alcohol consumption, including single instrument MR using
rs1229984 and multiple instruments MR using 34 SNPs, showed no causal
evidence of association [15]. In our subtype analyses with 83 SNPs,
we further found no causal effect of alcohol drinking on the development
of OC histotypes, which complement findings of overall OC from previous
studies.
Sleeping disorders are associated with the poor quality of life of women
suffering from OC. However, there are limited data about the effect of
sleeping disorders and lack of a proper amount of sleep on the
occurrence of OC. A previous prospective study reported a null
association between sleep duration and OC risk [18]. A recent
observational study also shows no association between sleep duration,
sleep quality, or insomnia with the risk of overall OC among
postmenopausal women. Nevertheless, in subtype analyses, restful sleep
quality was associated with a lower risk of invasive serous OC, and
insomnia was associated with a higher one [19]. In our MR analyses
by using up to 74 and 702 SNPs, we didn’t observe a causal effect of
sleep duration or insomnia on the development of any type of OC.
The major strengths of the present study are the MR design, the
systematic assessment of multiple lifestyle factors, and the inclusion
of well-classified OC subtypes. MR design strengthened the causal
inference by diminishing residual confounding and other biases. In
addition, consistent results from several sensitivity analyses
guaranteed the robustness of our findings. Our analysis was confined to
individuals of European ancestry, which largely diminished population
stratification bias. However, the population confinement might limit the
generalization of our findings to other populations. Limitations in our
study need consideration. Although indicative pleiotropy was observed in
a few analyses of BMI by the MR-Egger regression, several aspects of our
results suggested a minimal probability that pleiotropy would bias our
results, including a few outliers detected in the MR-PRESSO analysis and
consistent results from multiple sensitivity models. In addition, we
were unable to examine possible pre- and post-menopausal differences in
associations of studied exposures with OC due to a lack of menopause
data. However, given that most OC cases occur after menopause and that
age-matched controls were used, the inclusion of some pre- or
perimenopausal women in these analyses would less likely have biased
results. Finally, further experimental studies to reveal the possible
mechanisms, through which factors that appear to influence OC in these
analyses, could help to expand the scope for prevention opportunities
across the life course.