Results
Education
Our results showed significant inverse association of genetically determined years of education with overall OC (odds ratio [OR], 0.8 [95% CI, 0.72–0.89]) (Fig. 2A ). Suggestive associations were also observed for the protective effect of education attainment on the subtypes of LMSOC (OR, 0.64 [95% CI, 0.48–0.84]), HGSOC (OR, 0.81 [95% CI, 0.72–0.92]), IMOC (OR, 0.68 [95% CI, 0.51–0.92]), LGSOC (OR, 0.67 [95% CI, 0.46–0.98]), and EOC (OR, 0.8 [95% CI, 0.65–0.99]). The inverse associations were persistent in the weighted median and maximum likelihood mode. Although there was inconsistency in the MR-Egger method, the MR-Egger test for pleiotropy indicated that no significant pleiotropy was present. After removing outliers in the MR-PRESSO analysis, the association of education with all OC and OC subtypes persisted and the P value for the distortion test were above 0.05 (Table S2 ). Most of the associations for education and OC subtypes persisted and attenuated slightly after multivariable adjustment (Table 2 ).
Coffee or tea consumption
There were limited data supporting associations of genetically predicted coffee, or tea consumption with overall OC risk (Fig. 2B-C ). However, higher genetically predicted coffee and tea consumption were both positively associated with the endometrioid subtype. The ORs of EOC were 1.70 [95% CI, 1.18–2.44]) for genetically predicted 50% increase in coffee consumption (Fig. 2B ), and 1.84 [95% CI, 1.10–3.07]) for genetically predicted 50% increase in tea consumption (Fig. 2C ). The positive association between coffee consumption and EOC was also replicated in the sensitivity analyses including the weighted median and maximum likelihood mode, although with wider CI in the MR-Egger method (Table S3 ). For tea consumption, the positive association was persistent in the weighted median method, albeit there was some inconsistency in effect estimates of other sensitivity analyses (Table S4 ). The association between coffee or tea consumption and EOC did not remain after adjusting for BMI (Table 2 ).
Relative fat intake and BMI
The results indicated that a higher relative fat intake causally increases the risk of HGSOC (OR, 1.85 [95% CI, 1.06, 3.22]), but reduces the risk of EOC (OR, 0.3 [95% CI, 0.11, 0.86]) (Fig. 2D ). The associations remained consistent in the maximum likelihood mode, albeit with wider CI in the weighted median and MR-Egger method (Table S5 ). After adjustment for BMI, the association between fat intake and HGSOC attenuated greatly. However, the protective effect of fat intake on EOC was not affected by BMI (Table 2 ).
Genetic predisposition to higher BMI was significantly associated with an increased risk of LMSOC (OR, 1.44 [95% CI 1.18–1.77]) (Fig. 2E ). The positive association was consistent in the further analyses including weighted median, MR-Egger, and Maximum Likelihood, suggesting little evidence for pleiotropy and potential violations of instrumental variable assumptions (Table S6 ). A suggestive association was also found between BMI and the EOC (OR, 1.27 [95% CI 1.07–1.50]). The effect estimate was persistent when performing MR-Egger and maximum likelihood and was slightly attenuated when employing weighted median mode estimators. In addition, a borderline association was observed between BMI and LMMOC in IVW (OR, 1.29 [95% CI 1.01–1.66]) and maximum likelihood mode. Although no statistically significant association was observed between BMI and all OC, sensitive analysis indicated a suggestive association (Table S6 ).
Smoking
Despite null associations between smoking initiation and risk of OC subtypes (Fig. 2I and Table S10 ), genetic liability to lifetime smoking index showed a suggestive association with an increased risk of overall OC (OR, 1.23 [95% CI 1.04–1.46]), HGSOC (OR, 1.25 [95% CI 1.02–1.53]) and EOC (OR, 1.51 [95% CI 1.01–2.26]) histotypes (Fig. 2J) . The positive associations remained consistent in the maximum likelihood mode, albeit with wider CI in the weighted median and MR-Egger method (Table S11 ).
Physical activities, alcohol drinking, sleep duration, and insomnia
Among other lifestyle factors, there were no associations to be found for physical activities (Fig. 2F-H) , alcohol drinking (Fig. 2K) , sleep duration (Fig. 2L) , or insomnia (Fig. 2M) with any type of OC in the primary analysis or the sensitivity analyses (Table S7-9 , S12-14 ).