Ovarian cancer incidence and death in average-risk women undergoing bilateral salpingo-oophorectomy at benign hysterectomy

BACKGROUND: Opportunistic bilateral salpingo-oophorectomy is often offered to patients undergoing benign hysterectomy to prevent ovarian cancer, but the magnitude of risk reduction obtained with bilateral salpingo-oophorectomy in this population remains unclear and must be weighed against potential risks of ovarian hormone deﬁciency. OBJECTIVE: This study aimed to quantify the relative and absolute risk reduction in ovarian cancer incidence and death associated with bilateral salpingo-oophorectomy at the time of benign hysterectomy. STUDY DESIGN: We performed a population-based cohort study of all adult women ( (cid:1) 20 years) undergoing benign hysterectomy from 1996 to 2010 in Ontario, Canada. Patients with ovarian pathology, previous breast or gynecologic cancer, or evidence of genetic susceptibility to malignancy were excluded. Inverse probability of treatment e weighted Fine-Gray subdistribution hazard models were used to quantify the effect of bilateral salpingo-oophorectomy on ovarian cancer incidence and death while accounting for competing risks and adjusting for demographic characteristics, gynecologic conditions, and comorbidities. Analyses were performed in all women and speciﬁcally in women of postmenopausal age ( (cid:1) 50 years) at the time of hysterectomy. RESULTS: We identiﬁed 195,282 patients (bilateral salpingo-oophorectomy, 24%; ovarian conservation, 76%) with a median age of 45 years (interquartile range, 40 e 51 years). Over a median follow-up of 16 years (interquartile range, 12 e 20 years), 548 patients developed ovarian cancer (0.3%), and 16,170 patients (8.3%) died from any cause. Bilateral salpingo-oophorectomy was associated with decreased ovarian cancer incidence (hazard ratio, 0.23; 95% conﬁdence interval, 0.14 e 0.38; P < .001) and decreased ovarian cancer death (hazard ratio, 0.30; 95% conﬁdence interval, 0.16 e 0.57; P < .001). At 20 years follow-up, the weighted cumulative incidences of ovarian cancer were 0.08% and 0.46% with bilateral salpingo-oophorectomy and ovarian conservation, respectively, yielding an absolute risk reduction of 0.38% (95% conﬁdence interval, 0.32 e 0.45; number needed to treat, 260). After restricting to women aged (cid:1) 50 years at hysterectomy, the absolute risk reduction was 0.62% (95% conﬁdence interval, 0.47 e 0.77; number needed to treat, 161). CONCLUSION: Bilateral salpingo-oophorectomy resulted in a signiﬁ-cant absolute reduction in ovarian cancer among women undergoing benign hysterectomy. Population-average risk estimates derived in this study should be balanced against other potential implications of bilateral salpingo-oophorectomy to inform practice guidelines, patient decision-making, and surgical management.


Introduction
Ovarian cancer (including epithelial carcinoma of the ovary, fallopian tube, and peritoneum) is typically diagnosed at an advanced stage with limited prospects for cure. 1,2 Opportunistic bilateral salpingo-oophorectomy (BSO) is therefore considered at the time of clinically indicated hysterectomy, the most common major surgery performed in nonpregnant women worldwide, to prevent ovarian cancer later in life. 3,4 Although BSO substantially reduces the rates of both ovarian cancer and allcause death in women who carry pathogenic mutations in BRCA1 and BRCA2, 5e8 its risk-to-benefit ratio in the general population undergoing hysterectomy remains unclear. 4,9 BSO results in the complete cessation of ovarian estrogen production and therefore may have harmful effects that outweigh the benefit of ovarian cancer prevention in nonehigh-risk women. Studies to date suggest that BSO at benign hysterectomy may be associated with increased all-cause death compared with ovarian conservation when performed in women aged <50 years. 10e14 It is also important to note that the residual risk of peritoneal carcinoma remains even after BSO.
Ovarian cancer risk reduction may remain a reasonable goal for women undergoing hysterectomy, particularly those aged !50 years. 4 However, published guidelines provide no recommendation on whether BSO should be performed in this population, 15e20 and the use of BSO varies substantially among surgeons, indicating ongoing uncertainty in practice. 9 Estimates of the magnitude of ovarian cancer risk reduction obtained with BSO are needed to guide treatment standards and patient decision-making. Current data are derived from studies limited by selection bias, 13,14,21e23 too few events for multivariable adjustment, 13,14,21,22 and inclusion of patients with ovarian pathology at baseline. 12 We therefore examined the effect of BSO on the incidence of ovarian cancer and ovarian cancer death in a large population-based cohort of Accrual dates were selected to (1) ensure sufficient follow-up for the development of ovarian cancer in all patients and (2) ensure clinical practice was consistent throughout the study so that the risk reduction associated with BSO could be accurately estimated. Surgeons in Canada routinely performed hysterectomy either with or without BSO before 2010 but began to adopt bilateral salpingectomy (BS) alone after publication of the tubal hypothesis in 2010, 25,26 which postulated that high-grade serous cancers may originate in the fallopian tube and thus BS alone may reduce the risk of ovarian cancer. 27 Although the potential risk reduction associated with BS is also unclear, 28,29 we did not consider patients undergoing BS alone as a separate exposure in this study because of insufficient power and follow-up for that group.
We excluded (1) non-Ontario residents ineligible for universal health coverage, (2) patients undergoing emergent hysterectomy, (3) patients undergoing hysterectomy for a malignant indication, (4) patients with previous breast or gynecologic cancer, (5) patients whose index surgery or any previous surgery was performed for genetic susceptibility to malignancy, (6) patients who had previously undergone BSO, and (7) patients who had evidence of ovarian pathology or cysts at the index hysterectomy. These criteria were chosen to ensure our cohort represented a population that was not at high risk of ovarian cancer (Appendix 2; Figure 1).

Exposure assessment
The primary exposure was BSO, defined as removal of all ovarian tissue and corresponding fallopian tubes on the date of hysterectomy (index date). This included BSO in women with both ovaries and unilateral salpingooophorectomy in women with 1 remaining ovary because of a previous surgical procedure. We used procedure codes from DAD and SDS to identify salpingo-oophorectomy with a sensitivity of 99%, positive predictive value of 98%, and kappa of 99% (Appendix 1). 24 We compared women undergoing BSO with women undergoing conservation of one or both ovaries, as this represents standard practice if surgery is done for ovarian cancer prophylaxis and reflects loss or retention of ovarian endocrine function, respectively. We considered BSO as a static covariate, as the overall rate of adnexal surgery following hysterectomy was only 1.6%.

Outcome assessment
The primary outcome was ovarian cancer, obtained from the Ontario Cancer Registry (OCR), which holds records of all incident cancers in the province from 1964 to 2019 and is over 95% complete (Appendix 3). 30,31 The secondary outcome was ovarian cancer death. Date and cause of death were ascertained from the ICES Registered Persons Database and OCR, respectively, and were available from 1990 to 2017. Patients were therefore followed from the date of hysterectomy (time 0) to December 31, 2019, for ovarian cancer incidence and December 31, 2017, for ovarian cancer death.

Covariates
Covariates were ascertained at the time of the index hysterectomy. Demographic characteristics were age, rural or urban residence, era of surgery (1996e2000, 2001e2005, 2006e2010), residential income quintile, ethnicity (general AJOG at a Glance Why was this study conducted? This study aimed to describe the extent to which opportunistic bilateral salpingooophorectomy (BSO) would reduce ovarian cancer incidence and death in nonehigh-risk women undergoing benign hysterectomy.
What does this add to what is known? BSO effectively prevents ovarian cancer. This study uniquely quantified the absolute risk reduction and number needed to treat associated with BSO, which should be balanced against other potential implications of the procedure to inform practice guidelines and prevention strategies.
ajog.org GYNECOLOGY Original Research population, South Asian, Chinese), and immigration status (immigrant, longterm resident). Residential income quintile is an area-level socioeconomic index derived from Canadian census data on median neighborhood income and is assigned to patients based on their postal code of residence. 32 Immigration status was assigned to patients based on their landing date in Ontario 33 (longterm resident: landing date absent or <1985). Ethnicity was assigned using validated surname lists that accurately identify South Asian and Chinese individuals, Canada's 2 largest visible minority groups. 34 Clinical characteristics were gynecologic conditions documented at the index hysterectomy (abnormal uterine bleeding, fibroids, endometriosis, premalignant conditions [endometrial hyperplasia, cervical dysplasia], pelvic pain or inflammation, prolapse), 35e37 surgical approach (abdominal, vaginal), hysterectomy type (total, subtotal), overall comorbidities (0e5, 6e9, !10), 38 specific comorbidities (hypertension, diabetes mellitus, chronic obstructive pulmonary disease, previous malignancy, cardiovascular disease), previous ovarian surgery, and previous tubal ligation. Because our exposure was meant to reflect the loss or retention of ovarian endocrine function, we did not exclude patients based on previous adnexal procedures unless they had already undergone BSO and instead adjusted for previous ovarian surgery and previous tubal ligation as covariates in our analysis. Gynecologic conditions and surgical history were ascertained with DAD and SDS and OHIP codes; of note, abdominal surgery included open, laparoscopic, roboticassisted, and laparoscopic-assisted vaginal hysterectomies (Appendix 4). 35,37,39 Comorbidities were categorized into Aggregated Diagnosis Groups using the Johns Hopkins Adjusted Clinical Group System (version 10). 38,40 Specific comorbidities were ascertained from validated registries of Ontarians affected by these conditions (Appendix 5). 41e44

Statistical analysis
Datasets were linked using unique encoded identifiers and analyzed at ICES. We compared baseline characteristics between patients undergoing BSO and ovarian conservation using 2sample t tests or Mann-Whitney U tests for continuous variables and chi-square tests for categorical variables. We quantified the difference in means and proportions of baseline characteristics between the 2 groups using standardized differences. 45 We used inverse probability of treatment (IPT) weighting to adjust for differences between patients undergoing BSO and ovarian conservation; IPT weighting uses weights based on the propensity score to create a synthetic sample in which the distribution of baseline covariates is independent of treatment status. 46,47 Propensity scores were obtained using logistic regression, modeling BSO as the outcome and all demographic and clinical characteristics previously described as covariates. 48 To ensure that systematic differences in exposed and unexposed subjects were eliminated, we modeled age as a restricted cubic spline with 5 knots 49 and added individual interaction terms between age and both surgical approach and prolapse. 48,50 Furthermore, we generated stabilized IPT weights for each patient, equal to the inverse probability of undergoing the surgery received, multiplied by the probability of undergoing that surgery in the overall sample. 46 We again used standardized differences to assess for balance in baseline characteristics after applying stabilized IPT weights.
We used IPT-weighted Fine-Gray subdistribution hazard models to estimate the relative effect of BSO on each outcome; Fine-Gray models are time-toevent models that account for competing risks. 51e53 The model for incident ovarian cancer treated death as a competing risk, and the model for ovarian cancer death treated death owing to other causes as a competing risk. Patients were censored at loss to follow-up (defined as loss of eligibility for provincial health insurance) or end of followup (December 31, 2019, for ovarian cancer or December 31, 2017, for ovarian cancer death). Models used robust variance estimators to account for IPT weighting and were presented with subdistribution hazard ratios (HRs) and corresponding 95% confidence intervals (CIs).   ajog.org

GYNECOLOGY Original Research
We generated IPT-weighted cumulative incidence functions to estimate the absolute effect of BSO on ovarian cancer and ovarian cancer death. 52, 53 We plotted weighted cumulative incidence curves for patients undergoing BSO and ovarian conservation and used P values from respective Fine-Gray models to test the equality of curves across  53 We computed the risk difference in weighted cumulative incidence functions between the 2 groups, which in this setting corresponded to the absolute risk reduction (ARR), at 10, 15, and 20 years of follow-up. We took the inverse of the ARR to compute the number needed to treat (NNT) to prevent 1 case or 1 death by each time point. 54 We generated 95% CIs for all estimates using the 2.5th and 97.5th percentile of 1000 bootstrapped estimates.
To estimate the ARR associated with BSO in postmenopausal women, we repeated the analyses in a subcohort of women aged !50 years at the time of hysterectomy. To ensure that our estimates accurately represented the effect of BSO in patients who were not at high risk of ovarian cancer, we repeated the analyses (1) after excluding patients with any previous malignancy and (2) after excluding a small proportion of patients in the ovarian conservation group who underwent BS alone. To determine whether our findings were robust to multiple approaches for confounder control, we repeated all analyses using conventional multivariable Fine-Gray subdistribution hazard models rather than IPT weighting.
The proportionality assumption was confirmed by testing for an interaction between BSO status and time, which was not significant for either outcome. Tests were 2-sided, with P<.05 and standardized differences of !0.1 considered significant. Complete case    Figure 2; Table 3).

Ovarian cancer death
Median follow-up for ovarian cancer death was 14 years (IQR, 10e18); during that time, 240 women (0.1%) died from ovarian cancer, and 13,315 women (6.8%) experienced the competing event of death from other causes ( Table 2).

Principal findings
In this population-based cohort study of >195,000 nonehigh-risk women undergoing benign hysterectomy, BSO was associated with significant decrease in the incidence of ovarian cancer and ovarian cancer death. At 20 years followup, the absolute reduction in ovarian cancer incidence was 0.38% (95% CI, 0.32e0.45). Although the 20-year ARR was greater in a subset of women aged !50 years at hysterectomy (0.62%; 95% CI, 0.47e0.77), young women would likely be similarly protected on reaching a comparable age. In comparison, existing modalities for screening have not been shown to reduce ovarian cancer deaths by any margin. 55

Results and implications
Our study provided population-average risk reduction estimates for ovarian cancer and ovarian cancer death after BSO at benign hysterectomy that in combination with other data can help inform practice guidelines, patient counseling, and surgical management. Previous studies have reported absolute incidences of ovarian cancer in a similar range (Table 4) but enrolled selected cohorts of women that may not represent typical women undergoing benign hysterectomy (ie, Nurses' Health Study, 13 Women's Health Initiative, 21 Cancer Prevention Study II 23 ) and did not adjust risk estimates for differences in patients undergoing BSO and ovarian conservation. 13,14,21 A cohort study by Chan et al 22 is most comparable with ours in its use of administrative data and design focused specifically on ovarian cancer outcomes. Among 56,692 women undergoing benign hysterectomy from the Kaiser Permanente of Northern California database, BSO was associated with a decrease in the rate of ovarian cancer (HR, 0.12; 95% CI, 0.05e0.28) based on 40 cases identified over a median follow-up of 5 years. Our estimates are precise, based on 584 ovarian cancer cases identified over a median 16 year follow-up; better reflect outcomes of the general population undergoing benign hysterectomy; and account for demographic factors, gynecologic conditions, and comorbidities.
Decisions on whether to perform BSO must weigh the magnitude of the benefit illustrated here against other possible risks of the surgery. The addition of BSO to hysterectomy does not increase the complexity of the procedure and is not thought to appreciably influence shortterm surgical risk; studies that adjust for underlying gynecologic disease have not shown an increase in complications or length of stay with BSO, except  Original Research GYNECOLOGY ajog.org possibly when performed via a vaginal approach. 57e59 In contrast, the addition of BSO may influence long-term surgical risk because of the cessation of ovarian hormone production. BSO has been associated with increased all-cause mortality in women aged <50 years, 10e14 who are likely premenopausal at the time of surgery. Similar findings have not been consistently identified in women aged !50 years, 10,11,21 and only 1 decision analysis based on estimates derived from limited data has reported a protective effect of ovarian conservation to women aged 65 years. 60 Other theoretical effects of BSO include reduced quality of life or sexual function, 61,62 as the ovaries continue to produce androgens even after menopause, 19,63 but most studies have not shown significant differences in these measures postoperatively. 64e68 Although our estimates for ovarian cancer reduction were supportive of BSO in women aged !50 years, additional studies confirming that BSO is not associated with all-cause mortality in this age group are needed before the procedure can be routinely recommended after menopause.

Strengths and limitations
This study addressed the main limitations of previous works. We included a large cohort with many incident cases and deaths throughout prolonged follow-up, excluded patients with ovarian pathology at the index hysterectomy, explicitly accounted for competing risks, and adjusted for differences in patients undergoing BSO and ovarian conservation. The databases and registries employed were of high quality, allowing accurate and complete ascertainment of BSO status and cancer outcomes. With IPTweighting, we were able to derive population-average risk estimates that can be directly quoted in consent discussions and easily interpreted by patients. Furthermore, our study was population based, capturing all patients undergoing benign hysterectomy in Ontario; this ensured that our findings were highly generalizable to populations in other jurisdictions. Several limitations have merited consideration.
First, we lacked ajog.org GYNECOLOGY Original Research information on family history and genetic testing and could not definitively confirm whether patients were not at high risk. To mitigate this, we applied strict exclusion criteria, and the results were similar in sensitivity analyses restricted to patients without a previous malignancy. Second, we were unable to obtain information on factors that protect against ovarian cancer at baseline, including combined contraceptive use and parity for most women in our study.
As women receiving ovarian conservation are more often parous and past contraceptive users, 11,13 our results may be conservative estimates of the effect of BSO. Third, there were few incident ovarian cancer cases and ovarian cancer deaths throughout the study period. We addressed this by generating absolute and relative risk estimates, controlling for covariates with a propensity-based approach, and using bootstrapping to generate accurate 95% CI; moreover, it is also important to note that we identified many more events than previous studies (Table 4). Fourth, despite the high quality of our data sources overall, some covariates had limited sensitivity (eg, ethnicity) or relied on diagnostic or procedural codes with unknown accuracy (eg, gynecologic conditions). Any misclassification of these covariates is likely non-differential with respect to BSO status; therefore, adjustment will still likely reduce confounding bias. 69 Finally, because of limited numbers and follow-up for patients undergoing surgery after 2010, we could not assess ovarian cancer outcomes following BS alone. 28,29 Our results did not change after excluding 2% of patients in the ovarian conservation group who also underwent BS, but additional studies are required to determine the magnitude of benefit offered by BSO compared with BS alone; we intended to complete this analysis as soon as sufficient follow-up has accrued for our cohort.

Conclusions
In the United States, >400,000 women undergo hysterectomy annually. 70 The population-average risk estimates derived from this study were generalizable. Such estimates can be used to counsel patients on the value of concurrent BSO preoperatively, so they can make informed choices that weigh the magnitude of ovarian cancer prevention against other potential risks of the procedure. n