Ovarian cancer is the fifth leading cause of cancer deaths in US women, due to its typically advanced stage at presentation.1,2 Furthermore, unlike breast or colorectal cancer, there is no proven screening method for ovarian cancer to identify early disease and initiate treatment to improve survival.3,4 Family history, oral contraceptive use, parity, body mass index (BMI), and genetic variants are potentially useful in estimating lifetime risk.1 In particular, inherited BRCA1 and BRCA2 mutations are associated with increased lifetime risk of ovarian cancer and account for ~10–15% of overall disease incidence.5,6,7 However, among mutation carriers, age at diagnosis is variable. Penetrance of BRCA1/2 mutations is likely modified by other genetic variants and lifestyle or reproductive factors.8,9 Investigation of these factors could aid in implementation of strategies to reduce ovarian cancer risk among mutation carriers.

Both height and BMI are quantitative traits with substantial genetic bases. In recent genome-wide association studies (GWAS), numerous genetic variants were found to be associated with these traits.10,11 In the general population, both height and BMI appear to be positively but inconsistently associated with risk of ovarian cancer.12,13,14 Previous studies also showed that the association between BMI and ovarian cancer was stronger in premenopausal women.12,15,16 Because of differences in age at onset and tumour histology/grade, risk factors for ovarian cancer might be different for BRCA1/2 mutation carriers than women in the general population.17 Only one case–control study, with 469 ovarian cancer cases, has examined anthropometric measurements in BRCA1/2 mutation carriers and found that neither height nor BMI were related to ovarian cancer risk.18 Larger, adequately powered studies are needed to assess whether a relationship between either height or BMI and ovarian cancer risk exists for BRCA1/2 mutation carriers, and whether the direction of association is concordant with that in the general population.

Mendelian randomisation (MR) methods use genetic markers associated with a trait as an instrumental variable (IV) to assess their potential relationship with a disease outcome.19,20,21 Compared to traditional epidemiologic approaches, MR can reduce biases such as reverse causation and residual confounding, which can interfere with causal interpretations. However, the MR approach requires that the genetic variants are associated with the exposure, the variants are not or only weakly associated with confounding factors in the causal pathway, and the variants only affect disease risk through the exposure (i.e. absence of pleiotropic effects).20,21 To the degree that these assumptions are met, the MR approach can strengthen the evidence for a causal relationship between exposure and disease.

Herein, using traditional epidemiologic and MR methods, we conducted analyses of height and BMI and their association with ovarian cancer risk in the Consortium of Investigators for the Modifiers of BRCA1/2 (CIMBA) with 22,588 participants. We examined heterogeneity of these associations with respect to the mutation carried (BRCA1 vs BRCA2), menopausal status, tumour histology, and tumour grade.


Characteristics of the CIMBA consortium and information on specific genotyping protocols are provided in Supplementary Methods and were described previously.22,23,24

Selection of genetic variants

From the latest publications of the Genetic Investigation of Anthropometric Traits, we identified single-nucleotide polymorphisms (SNPs) associated with height or BMI at genome-wide significance level (P < 5 × 10−8).11,25 SNPs with low imputation quality (<0.5) were excluded, leaving 586 SNPs for height and 93 for BMI. Supplementary Tables 1 and 2 provide additional details on these SNPs.

Statistical analysis

Calculation of the height and BMI genetic scores (GS) was described in detail previously.24 Briefly, we calculated the weighted sums of all of the height- and BMI-associated variants under additive models, which do not include interactions between variants. Namely, we used the formulas: \({\mathrm{Height - GS}} = \mathop {\sum}\nolimits_{i = 1}^{586} {\beta _{XGi}{\mathrm{SNP}}_i}\) and \({\mathrm{BMI - GS}} = \mathop {\sum}\nolimits_{i = 1}^{93} {\beta _{XGi}{\mathrm{SNP}}}\), where βXGi is the literature-reported per-allele magnitude of association of the ith SNP for height and BMI, respectively. A scaling factor was calculated by regressing each GS against its respective trait among non-case carriers. The corresponding regression coefficients were β0 (intercept = 165.455) and β1 (slope = 5.217) for height and β0 (22.607) and β1 (5.523) for BMI. In the present study, BMI-GS was scaled to BMI at the date of questionnaire, rather than BMI at age 18 years, as previous GWAS were based on BMI measurements in middle-aged adults.

We subsequently modelled each scaled GS against ovarian cancer risk using weighted Cox models. Our primary outcome of interest was ovarian cancer diagnosis, with individuals censored for breast cancer diagnosis, risk-reducing bilateral salpingo-oophorectomy, death, or end of follow-up, whichever occurred first. Owing to the study design of CIMBA, weights in the model were applied for cases and non-cases based on previously observed incidence of ovarian cancer in BRCA1/2 carriers.26,27 We applied a robust sandwich variance-estimation approach to the risk estimates to account for non-independence among multiple carriers per family. In addition, we performed subgroup analyses by BRCA1/2 mutations and menopausal status. Menopausal status was defined as a time-varying covariate, coded as premenopausal from birth until age at natural menopause or bilateral salpingo-oophorectomy. For individuals with missing age at menopause, we imputed the age as 50 years. Imputing missing age at menopause as 46 years did not materially change the results. The mean and median ages at natural menopause in this population were 46 and 48 years, respectively. All analyses were adjusted for the first eight principal components (to account for ethnicity and population stratification), birth cohort, and country of enrolment. Additional analyses assessed the associations of height and BMI with ovarian cancer subgroups by histological type (serous vs. non-serous) and by tumour grade (well or moderately differentiated tumours vs. poorly or undifferentiated).

In addition, phenotype associations with each individual height and BMI variant were assessed and pooled using inverse variance-weighted meta-analysis. The individual associations were obtained by first extracting βXGi for each SNP i, which represents the per-allele magnitude of association with height or BMI from previous GWAS. Next, we calculated βYGi and \(SE\left( {\beta _{YGi}} \right)\) using multivariate-adjusted weighted Cox models for each SNP using the CIMBA data, where ovarian cancer risk is predicted by genotype G (with G = 0, 1, 2 for the allele corresponding to greater height or BMI), principal components, birth cohort, BRCA mutation, and country of enrolment. The overall causal association (βYX) is calculated using inverse-variance weighted estimate of each variant’s effect: \(\beta _{YX} = \frac{{\mathop {\sum }\nolimits_i \beta _{XGi}\beta _{YGi}SE(\beta _{YGi})^{ - 2}}}{{\mathop {\sum }\nolimits_i \beta _{XGi}^2SE(\beta _{YGi})^{ - 2}}}\). Standard error was estimated as \(SE_{YX} = \sqrt {\frac{1}{{\mathop {\sum }\nolimits_i \beta _{XGi}^2SE(\beta _{YGi})^{ - 2}}}}\) using the Burgess’s method.19,28 Egger’s test was used to assess for possible pleiotropic effects of the variants (i.e. whether variants influence the outcome through other pathways), to ensure that this assumption held.29

Finally, in participants with available data on height and BMI, we conducted a formal IV analysis using the method of two-stage residual inclusion regression.30 In stage one, observed height or BMI was regressed against the corresponding GS, principal components, birth cohort, country, and mutation status. In the second stage, we used a Cox model to fit ovarian cancer risk against height or BMI, birth cohort, country, mutation status, and residuals from stage one. Variance estimates were obtained through 10,000 boot-straps (see details in Supplementary Methods). In these individuals, we also analysed the association between observed measurements of height or BMI and ovarian cancer risk using weighted Cox models, adjusted for established ovarian cancer risk factors, including birth cohort, menopausal status, age at menarche (years), and parity (continuous). The BMI values used were obtained at the date of questionnaire, usually close to the date of genetic testing and recalled for BMI at age 18 years.

In models with menopausal status as time-varying variable, the test for heterogeneity by menopausal status was essentially a test of the proportional hazards assumption. All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC) and Stata 14.0 (StataCorp, College Station, TX). A two-sided P-value < 0.05 was considered statistically significant unless stated otherwise.


Demographic and clinical characteristics

Characteristics for the 22,588 individuals in the CIMBA consortium, comprising 14,676 BRCA1 and 7912 BRCA2 mutation carriers, are shown in Table 1. We documented 2923 women with ovarian cancer (BRCA1: 2319; BRCA2: 604). Compared with non-cases, participants who developed ovarian cancer were more often parous women, were younger at first live birth, and were from earlier birth cohorts. At the date of questionnaire/interview, height measurement was available for 7657 participants and BMI measurement for 7516 participants. Most tumours for BRCA1/2 mutation carriers were invasive, of serous, poorly, or undifferentiated grade, and stages 3 or 4 at diagnosis, characteristics which are consistent with prior reports.31

Table 1 Baseline characteristics of participants in the CIMBA consortium with genotype information

Observed and predicted height on risk of ovarian cancer

In the survival modelling of ovarian cancer risk, age was used as the underlying timescale and the numbers of individuals retained in the analysis were 20535, 14647, 7375, and 2832 at ages 30, 40, 50, and 60 years, respectively, suggesting that statistical power for the late age is limited. After adjustment for birth cohort, country of enrolment, mutation, menopausal status, and principal components, a nonsignificant association was found for observed height and ovarian cancer risk (hazard ratio (HR) = 1.07 per 10-cm increase, 95% confidence interval (CI): 0.94–1.23, P = 0.31) (Table 2). We found broadly consistent associations of height in both BRCA1 and BRCA2 mutation carriers by menopausal status and by tumour histological type and grade.

Table 2 Association of height and ovarian cancer risk using observed height among 7657 participants

The height GS was significantly associated with height in all participants, in ovarian cancer cases, and in non-case participants (all P < 10−24) (Supplementary Table 3). Overall, approximately 13.4% of the variation in height was explained by the height GS. Besides height, we found weaker associations between the height GS and body weight and age at menarche.

In MR analysis, height GS had a nonsignificant positive association with ovarian cancer risk, HR = 1.02 per 10-cm increase in genetically predicted height, 95% CI: 0.85–1.23, P = 0.82 (Table 3). We found similar associations by subgroups of mutation, menopausal status, and tumour grade.

Table 3 Association of height and ovarian cancer risk among 22,588 participants in CIMBA per 10-cm increase in genetically predicted height

Combining the effects of all 586 height-associated variants using inverse-variance weighted meta-analysis, we obtained similar findings (HR = 1.02, 95% CI: 0.83–1.26, P = 0.83) (λ). Among the SNPs that were combined, there was a low degree of heterogeneity (I2 = 0%). Examining small-study effects using Egger’s test did not suggest likely pleiotropic effects. In the two-stage residual inclusion analysis, the estimated relative risk was larger though with wide CIs, which overlapped with those derived using other methods (HR = 1.20, 95% CI: 0.86–1.69, P = 0.29).

Observed and predicted BMI on risk of ovarian cancer

After multivariable adjustment, we found a nonsignificant positive association between BMI at the date of questionnaire completion and ovarian cancer risk, HR = 1.04 per 5-kg/m2 increase in BMI, 95% CI: 0.95–1.14, P = 0.42 (Table 4). In a pre-specified analysis, the association between BMI and ovarian cancer risk was stronger in premenopausal women (HR = 1.25, 95% CI: 1.06–1.48; P = 0.009), whereas no association was found in postmenopausal women (HR = 0.98, 95% CI: 0.88–1.10), with significant interaction (P = 0.02). We found that BMI was a significant predictor of non-serous ovarian cancer risk (HR = 1.25, 95% CI: 1.06–1.49) but not for serous ovarian cancer (HR = 0.98, 95% CI: 0.84–1.15).

Table 4 Association of body mass index (BMI) and ovarian cancer risk using observed BMI

Similar to BMI at the date of questionnaire completion, we detected a significant interaction of BMI in young adulthood and menopausal status (P = 0.01), with a stronger association for premenopausal women (HR = 1.34, 95% CI: 0.97–1.84) compared with postmenopausal women (HR = 0.82, 95% CI: 0.65–1.04).

BMI-GS was strongly associated with BMI at both the date of questionnaire completion and young adulthood (Supplementary Table 4). Overall, the BMI-GS explained 2.6% of the variation in BMI at the date of questionnaire completion and 1.7% of the variation in young adulthood BMI. We found associations between the BMI-GS and height and age at menarche, though the strength of the association was weaker than the association with BMI.

In the entire consortium, the BMI-GS had a nonsignificant positive association with ovarian cancer risk with a HR = 1.10 per 5-kg/m2 of genetically predicted BMI, 95% CI: 0.86–1.42, P = 0.44 (Table 5). We found heterogeneity by menopausal status (P = 0.006). BMI-GS was positively associated with ovarian cancer risk in premenopausal women (HR = 1.59, 95% CI: 1.08–2.33) but not in postmenopausal women (HR = 0.80, 95% CI: 0.58–1.11). BMI-GS also tended to be more associated with non-serous (HR = 1.60, 95% CI: 0.83–3.08) than serous tumours (HR = 0.92, 95% CI: 0.59–1.43).

Table 5 Association of body mass index genetic score (BMI-GS) and ovarian cancer risk among 22,588 participants in CIMBA, per 5 kg/m2 increase in genetically predicted BMI

We found similar results when we statistically combined the associations of the 93 BMI-associated variants, with an overall HR = 1.12, 95% CI: 0.86–1.46. Heterogeneity was low (I2 = 15.9%), indicating a low likelihood of pleiotropic associations. Using the two-stage residual inclusion approach, we found a generally similar association (HR = 1.37, 95% CI: 0.84–2.24, P = 0.21).

Individual SNPs and ovarian cancer risk

We found 22 height-associated and 4 BMI-associated SNPs that were nominally associated with ovarian cancer risk (P < 0.05; Table 6). None of these SNPs were significantly associated with ovarian cancer risk after correcting for multiple testing. We cross-checked these identified SNPs with the most up-to-date list of ovarian cancer susceptibility SNPs and did not find any overlaps.32

Table 6 Height or body mass index (BMI) single-nucleotide polymorphisms (SNPs) statistically significantly associated (P < 0.05) with ovarian cancer risk in CIMBA


Using data from a large international consortium of BRCA1/2 mutation carriers, we found no statistically significant association between height and ovarian cancer risk. Interestingly, we observed interactions between BMI (both observed and genetically predicted) and menopausal status on ovarian cancer risk, with increasing BMI associated with increased risk in premenopausal but not in postmenopausal women.

Our finding of a positive association between BMI and overall ovarian cancer risk in BRCA1 and BRCA2 mutation carriers is corroborated by several prior studies in the general population.12,14,15,33 One MR analysis using 77 BMI-associated SNPs, conducted in the general population, found that each 1-standard deviation (SD) increment in genetically-predicted adult BMI corresponded to an odds ratio (OR) of 1.35 (95% CI: 1.05–1.72).34 We found that 5-kg/m2 (about 1 SD) increment in genetically predicted BMI was associated with an HR = 1.10 (95% CI: 0.86–1.42) in mutation carriers. However, the association of BMI with ovarian cancer risk is likely to vary by menopausal status. In the general population, significant differential association of BMI with ovarian cancer risk by menopausal status has been found in some studies15,16,35,36 but not in others.12,37 A pooled analysis of 47 epidemiologic studies with 25,157 ovarian cancer cases showed that the relative risk per 5-kg/m2 increase in BMI was 1.12 (95% CI: 1.07–1.17) in premenopausal women and 1.08 (95% CI: 1.04–1.12) in postmenopausal women.12 The largest single cohort study, with 3686 ovarian cancer cases, found that the HR per 5-kg/m2 increase in BMI was 1.21 (99% CI: 1.09–1.33) in premenopausal and 1.07 (99% CI: 1.02–1.12) in postmenopausal women.15 An MR analysis conducted in the general population also observed stronger associations for non-high-grade serous carcinomas in premenopausal women (OR = 1.62, 95% CI: 0.88–3.01) compared with postmenopausal hormone replacement therapy (HRT) users (OR = 1.26, 95% CI: 0.57–2.82) and postmenopausal HRT non-users (OR = 1.17, 95% CI: 0.61–2.24), though no formal statistical tests examining heterogeneity were performed.14 Similarly, we found in BRCA1/2 mutation carriers that 5-kg/m2 increment in genetically predicted BMI was associated with an HR = 1.59 (95% CI: 1.08–2.33) for premenopausal ovarian cancer and an HR = 0.80 (95% CI: 0.58-1.11) for postmenopausal ovarian cancer. Studies that have not demonstrated significant variation by menopausal status tended to show that the positive association between BMI and ovarian cancer risk was primarily among those who had never used HRT.12 Taken together, our results and previous literature are suggestive that higher BMI may increase ovarian cancer risk in premenopausal women but not in postmenopausal women.

In addition, several studies that had sufficient numbers of cases to evaluate the relationship between BMI and ovarian cancer risk by histologic subtype have shown significant heterogeneity. Observational studies in the Ovarian Cancer Cohort Consortium found stronger associations between BMI and endometrioid (OR = 1.17 per 5-kg/m2, 95% CI: 1.11–1.23) or mucinous ovarian cancer (OR = 1.19, 95% CI: 1.06–1.32) but no association with serous ovarian cancer (OR = 0.98, 95% CI: 0.94–1.02).16 A more recent MR analysis in the same consortium using a genetic score comprised of 87 SNPs showed that a genetically predicted BMI had a stronger association with endometrioid (OR = 1.17, 95% CI: 0.87–1.59) or mucinous ovarian cancer (OR = 1.18, 95% CI: 0.84–1.67) than high-grade serous cancer (OR = 1.06, 95% CI: 0.89–1.27), though the 95% CIs for these estimates were largely overlapping.14 Consistent with findings in the general population, our study in BRCA1/2 mutation carriers showed that BMI was positively associated with non-serous ovarian cancer (HR = 1.25 per 5-kg/m2, 95% CI: 1.06–1.49 in observed BMI and HR = 1.60, 95% CI: 0.83–3.08, per 5-kg/m2 in genetically predicted BMI), of which endometrioid is a major subtype. Of note, obesity is an established risk factor for endometrial cancer.38 However, subsequent studies with greater number of cases of different ovarian cancer subtypes are needed to assess whether the effect of obesity truly differs by tumour subtype.

Our finding of a nonsignificant positive association between height and ovarian cancer risk is also consistent with prior epidemiological studies in the general population.12,37,39 In the general population, 5-cm increment in height was associated with a 7% increase (95% CI: 5–9%) in ovarian cancer risk,12 and 5-cm increment in genetically predicted height was associated with a 6% (95% CI: 1–11%) increase in ovarian cancer risk.39 The associations for observed height did not differ significantly between ovarian histological types,2,12 while genetically predicted height had a stronger association with clear cell (OR = 1.20, 95% CI: 1.04–1.38) or low-grade/borderline serous ovarian cancers (OR = 1.15, 95% CI: 1.01–1.30) compared to high-grade serous (OR = 1.05, 95% CI: 0.99–1.11).39 We did not find statistically significant heterogeneity by histology in our study of mutation carriers, though point estimates varied across histology.

Several biological mechanisms potentially explain the associations observed in our study. Overweight/obese women are more likely to have anovulatory cycles and fertility issues, particularly when caused by polycystic ovarian syndrome (PCOS), and thus have an increased risk of ovarian cancer.40,41 The association of PCOS with ovarian cancer risk was mainly confined to premenopausal women.42 Some studies have suggested that BRCA1/2 mutation carriers may have subclinical ovarian insufficiency, which could mediate the relationship between obesity-related infertility and increased ovarian cancer risk.43 Obesity itself also creates a proinflammatory state and adipocyte-secreted inflammatory markers have been implicated in ovarian cancer development.44 Circulating levels of oestradiol, androgen, and progesterone have also been implicated in the risk of ovarian cancer.45,46 One study in BRCA1/2 mutation carriers showed higher oestradiol levels during each menstrual cycle compared with non-carriers, supporting the potential role of sex hormones in ovarian tumorigenesis in this population.47 Obese premenopausal women tend to have lower circulating levels of progesterone compared with normal weight women.48 Higher progesterone levels may reduce ovarian cancer risk, through upregulation of p53, leading to tumour cell apoptosis.46,49,50,51 Taken together, these pathways may explain the association of higher BMI with premenopausal ovarian cancer risk. In addition, height has been associated with higher levels of circulating insulin-like growth factor-1 (IGF-1),52,53 a pathway that has been implicated in tumour transformation and may exert antiapoptotic and mitogenic effects.54,55 Moreover, BRCA1 may directly interact with the IGF-1 pathway to mediate cancer risk.56

Our study has several strengths, including large sample size, genetic scores utilising most identified height and BMI variants, several MR methods, and consistent findings between observed and genetically predicted phenotypes. Several limitations of our study should be considered. First, even with a large sample size, the CIs for most risk estimates were wide, which limits inferences about causation. While both the height- and BMI-GS were clearly associated with their respective traits, they were only able to explain 13.4% and 2.6% of the variation, respectively. This reduced the statistical precision of our risk estimates. During the preparation of our manuscript, a new genome-wide meta-analysis57 found a substantial number of new genetic loci related to height and BMI, increasing the amount of variation that could be explained for these two traits to 24.6% and 6.0%, respectively, although the variation that could be explained when examining these SNPs in a validation cohort was 14.0% and 2.3%. This is comparable to the amount of variation that could be explained using the set of genetic variants in our study. Including these additional SNPs may be able to improve the precision of our estimates for both height and BMI. Moreover, the inclusion of rare variants to strengthen the height and BMI genetic instruments should also be considered in future studies.58 Our study did not explicitly examine whether adding height or BMI (either observed or genetically predicted) to existing polygenic risk scores for ovarian cancer could further refine risk prediction. Histology was only available in a subset of ovarian cancer patients, which limits our capacity to understand subtype-specific effects of BMI and height. Our study only included women of European ancestry, which may preclude generalisation to women of other racial/ethnic groups.

In summary, our study suggests that higher BMI may be causally associated with ovarian cancer risk in BRCA1/2 carriers, possibly more so for premenopausal women. BMI could be used to identify premenopausal women at elevated risk of ovarian cancer. Our finding of a stronger association between BMI and non-serous ovarian cancer warrants confirmation in future studies.