Study design and population

We performed a population-based cross-sectional study using linked administrative databases held at the ICES (formerly the Institute for Clinical Evaluative Sciences), a nonprofit research institute authorized to collect and use health data on all residents of Ontario, Canada’s most populous province, for the purposes of health system evaluation and improvement. Because residents of Ontario have universal access to inpatient and outpatient hospital care and physician services, these data are comprehensive. The Research Ethics Board at the University of Toronto provided approval (no. 38212).

We included all adult women (≥20 years) in Ontario, Canada, undergoing elective abdominal hysterectomy for a benign indication from January 1, 2014, to December 31, 2018. We did not include vaginal hysterectomy because of limited anatomic access to the ovaries with this approach. In addition, we excluded (1) non-Ontario residents ineligible for universal health coverage; (2) patients undergoing suspected emergent hysterectomy, because of potential differences in surgical decision making in this setting; (3) patients undergoing hysterectomy for a malignant indication; (4) patients with previous breast or gynecologic cancer; (5) patients for whom the index surgery, or any previous surgery, was performed for a genetic susceptibility to malignancy; and (6) patients who had previously undergone BSO ( Supplemental Table 1 ; Figure 1 ).

Flow diagram of included patients

Cusimano et al. Practice variation in bilateral salpingo-oophorectomy. Am J Obstet Gynecol 2021 .

We identified patients undergoing hysterectomy from the Canadian Institute of Health Information Discharge Abstract Database (DAD) and Same Day Surgery (SDS) Database, which hold records for inpatient and outpatient surgical procedures, respectively, and the Ontario Health Insurance Plan (OHIP) Database, which houses billing claims for virtually all physician services provided in the province. Patients with a procedure code for hysterectomy in the DAD or SDS, and a claim for hysterectomy in OHIP within 6 weeks, were eligible for study inclusion ( Supplemental Table 2 ). Use of both DAD or SDS and OHIP codes enabled the identification of the primary surgeon and their characteristics. Procedure codes for hysterectomy have been validated against medical record abstraction and show sensitivity of 93%, positive predictive value of 100%, and kappa of 97%.

Outcome assessment

The primary outcome was BSO, defined as the 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 salpingo-oophorectomy in women with one remaining ovary because of a previous surgical procedure. We used procedure codes from DAD or SDS to identify salpingo-oophorectomy with sensitivity of 99%, positive predictive value of 98%, and kappa of 99% ( Supplemental Table 1 ).

A secondary outcome was potentially avoidable BSO, defined as BSO without a documented diagnostic indication. We adapted methodology developed by Mahal et al to classify BSO as potentially appropriate or potentially avoidable ( Supplemental Table 3 ) based on a broad list of the International Classification of Diseases (ICD-9, ICD-10) codes capturing common indications for BSO, such as endometrial hyperplasia, nonphysiological ovarian cysts or anomalies, and endometriosis. If pertinent diagnostic codes were not present in either the main diagnostic field or 1 of the 25 secondary diagnostic fields, then these cases of BSO were considered potentially avoidable.

Cohort description and covariates

Patient demographic characteristics were age, year of surgery, rural or urban residence, residential income quintile, immigration status (immigrant, long-term resident), and ethnicity (general population, South Asian, Chinese), evaluated at the time of the index hysterectomy. Residential income quintile is an area-level socioeconomic index derived from Canadian census data on median neighborhood income quintile and is assigned to patients based on their postal code of residence. Immigration status was assigned to patients based on their landing date in Ontario (long-term resident, landing date absent or <1985; immigrant, landing date present). Ethnicity was assigned using validated surname lists that accurately identify South Asian and Chinese individuals, Canada’s two largest visible minority groups.

Patient clinical characteristics included gynecologic conditions (abnormal uterine bleeding [AUB], fibroids, endometriosis, benign ovarian cyst, premalignant disease, prolapse, pelvic pain or inflammation), surgical approach (open, laparoscopic), hysterectomy type (subtotal, total), comorbidities (0–5, 6–9, 10+), morbid obesity (body mass index of ≥40 kg/m ² ), previous ovarian surgery, and previous malignancy. Hysterectomy details and gynecologic conditions were obtained from DAD or SDS using procedure and diagnostic codes, respectively ( Supplemental Tables 3 and 4 ). Comorbidities were ascertained from DAD, SDS, and OHIP using diagnostic codes, which were collapsed into Aggregated Diagnosis Groups (ADGs) based on clinical similarity, chronicity, and disability using the Johns Hopkins ACG System (version 10); patients could have 0 to 32 ADGs, with higher scores indicating more comorbidities. Morbid obesity was defined with OHIP codes claimed by physicians at the time of hysterectomy.

Surgeon characteristics included sex, year of graduation (quartiles), specialty (gynecology or other), country of graduation (Canada, other, unknown), and hysterectomy volume (quartiles). These variables were derived from the OHIP database and ICES Physician Database, which holds demographic data on Ontario physicians. Hysterectomy volume was defined as the number of hysterectomies performed by the patient’s surgeon in the 1 year before the index procedure. Compared with averaging the number of procedures performed over the study years, this definition ensured that surgeon volume could dynamically change in response to year-to-year variation of hysterectomy. ^,

Statistical analysis

We compared baseline characteristics between patients who did and did not undergo concurrent BSO using two-sample 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 groups using standardized differences.

Because we hypothesized a priori that surgeon-level variation in BSO may depend on patient age, our primary analyses were stratified by age group (<45, 45–54, ≥55 years). These categories were chosen because 90% of women experience menopause between the ages of 45–54 years. In addition, we performed sensitivity analyses for women aged 45–49 (menopausal transition) and 50–54 years (early menopause) to determine whether variation differed before and after the median age of natural menopause.

We used hierarchical random intercept multivariable logistic regression models to quantify surgeon-level variation in BSO and to identify surgeon and patient factors associated with BSO. This modeling approach accounts for clustering of patients within the same surgeon and enables the assessment of variation within and across surgeons. We ran separate models for each age strata. These models adjusted for patient age within each stratum as a continuous variable, AS WELL AS all patient and surgeon covariates described. Covariates were chosen on the basis of clinical rationale and literature. The marginal log likelihood was approximated with an adaptive Gauss-Hermite quadrature. The Model fit was confirmed by examining the ratio of the generalized chi-square statistic to its degrees of freedom, which was close to 1. There was no evidence of multicollinearity (variance inflation factor of <2 for all covariates). Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) are reported for each model.

Surgeon-level practice variation was described using (1) risk-adjusted rates of BSO for each surgeon, (2) the variance partition coefficient (VPC), and (3) the median odds ratio (MOR). The risk-adjusted rate of BSO denotes the rate of BSO that would be expected for each surgeon if patient case mix and surgeon characteristics were held constant between them. ^, The VPC reflects the proportion of residual variation in BSO DUE TO between-surgeon differences after accounting for all covariates, and can range from 0% (no variation because of surgeon) to 100% (all variations because of surgeon). The MOR compares the odds of BSO between two patients whose only difference is their treating surgeon. If one were to repeatedly draw two otherwise identical patients from available surgeons and create an OR between the patient from the surgeon with a greater tendency to perform BSO and the patient from the surgeon with a lesser tendency to perform BSO, then the MOR would reflect the median of these ORs when all possible comparisons of greater-tendency to lower-tendency surgeons are made. The MOR can be interpreted on the same scale as other ORs from the multivariable model. Additional details are provided in the Supplemental Methods .

Datasets were linked using unique encoded identifiers and analyzed at the ICES. Tests were two-sided with P <.05 deemed significant. Standardized differences reflect differences between group means and proportions relative to the pooled standard deviation; standardized differences of ≥0.1 were deemed indicative of meaningful differences between groups. Complete case analyses were performed as missing data were exceedingly rare for all variables (<0.1%), except for surgeon country of graduation (7%), which was coded with a level for unknown. Analyses were performed in SAS.

Study population

A total of 61,890 women (≥20 years) underwent elective abdominal hysterectomy from January 1, 2014, to December 31, 2018, in Ontario, Canada. After exclusions, our cohort included 44,549 women operated on by 706 unique surgeons; 39.9% of patients (n=17,797) underwent concurrent BSO, and 60.1% (n=26,752) did not ( Figure 1 ).

Median age was 44 years (interquartile range [IQR], 39–47) in the ovarian conservation group and 51 years (IQR, 47–59) in the BSO group ( Table 1 ). Patients who did undergo BSO were more likely to have open surgery (51.5% vs 46.5%; P <.001) and more likely to have benign ovarian cysts (22.6% vs 8.2%; P <.001) or premalignant disease (14.1% vs 5.2%; P <.001) than patients who did not undergo BSO ( Table 1 ). Rates of BSO varied by age group ( Supplemental Table 5 ): 17%, 44%, and 84% of patients aged <45, 45–54, and ≥55 years, respectively, underwent BSO ( P <.001). In the 45–54 year group, 33% of those aged 45–49 years and 64% of those aged 50–54 years underwent BSO. The proportion of patients undergoing BSO annually ranged from 38.5% to 41.2%.

Potentially avoidable bilateral salpingo-oophorectomy

Among the patients who underwent BSO (n=17,798), 41% (n=7373) did not have a documented indication and were classified as having a potentially avoidable procedure. Patients aged <45 years were less likely to have a potentially avoidable BSO than patients aged 45–54 or ≥55 years (31.1%, 44.0%, and 43.2%, respectively; P <.001). The proportion of BSO that were potentially avoidable each calendar year ranged from 40.1% to 42.7%.

Surgeon variation in bilateral salpingo-oophorectomy

Model-based risk-adjusted rates of BSO with each surgeon varied widely ( Figure 2 ). Surgeons predominantly preserved ovaries in women aged <45 years (IQR for BSO, 13.2%–22.8%) and removed ovaries in women aged ≥55 years (IQR, 81.5%–86.2%), but differed substantially in their practice for women aged 45–54 years (IQR, 33.9%–54.1%). In sensitivity analyses, surgeon practice was similarly variable in both women aged 45–49 (IQR, 23.9%–43.4%) and 50–54 years (IQR, 52.7%–72.6%) ( Supplemental Figure 2 ). Unadjusted rates of BSO with each surgeon showed the same pattern ( Supplemental Figure 1 ).

Risk-adjusted proportion of women (<45, 45–54, ≥55 years) undergoing BSO with individual surgeons

Each unique surgeon is represented by a single point along the x-axis. Three surgeons with risk-adjusted proportions ranging between 102% and 113% were plotted at a maximum of 100%.

BSO , bilateral salpingo-oophorectomy.

Cusimano et al. Practice variation in bilateral salpingo-oophorectomy. Am J Obstet Gynecol 2021 .

Significant variation between surgeons remained even after adjustment for patient and surgeon factors and was particularly marked in women 45–54 years ( Table 2 ). Surgeons accounted for 22% of the residual observed variation in BSO in women aged 45–54 years, compared with 16% and 14% of the residual observed variation in BSO in women <45 and ≥55 years, respectively. In each age strata, the magnitude of the MOR for the surgeon was comparable with or greater than the strongest patient-level predictors of BSO (<45 years, 2.10; 45–54 years, 2.53; ≥55 years, 2.00). In sensitivity analyses, surgeon variation in BSO was similarly high for both women aged 45–49 years (VPC, 21.8%; MOR, 2.30) and 50 to 54 years (VPC, 26.7%; MOR, 2.84) ( Table 2 ).

Surgeon and patient factors associated with bilateral salpingo-oophorectomy

Age-stratified multivariable logistic regression models are presented in Table 3 (<45, 45–54, ≥55 years) and Supplemental Table 6 (45–49 and 50–54 years). No specific surgeon factors were significantly associated with BSO except for specialty in women aged ≥55 years (other vs gynecology [reference]: OR, 0.28; 95% CI, 0.12–0.65; P =.003).

Although many patient factors, including age, gynecologic diagnoses, and route of hysterectomy, were associated with BSO, few had adjusted ORs of comparable magnitude to the MOR for the individual surgeon (2.00–2.53) ( Table 2 ). Women with benign ovarian cysts or premalignant disease had approximately twice the odds of BSO compared with women who did not have these diagnoses, whereas women with AUB or prolapse had approximately half the odds of BSO compared with women who did not have these diagnoses. Subtotal hysterectomy was also associated with approximately half the odds of BSO compared with total hysterectomy.

Several non-gynecologic patient factors were associated with BSO. Living in low-income (1 vs 5 [reference]: OR, 1.34; 95% CI, 1.16–1.55; P <.001) or urban areas (urban vs rural [reference]: OR, 1.18; 95% CI, 1.03–1.35; P =.017) was associated with BSO in women aged <45 years. Multiple comorbidities were associated with increased odds of BSO in women aged <45 years (10+ vs 0–5 [reference]: OR, 1.25; 95% CI, 1.09–1.43; P =.001) and 45–54 years (OR, 1.40; 95% CI, 1.25–1.56; P <.001) but not in women aged ≥55 years (OR, 1.04; 95% CI, 0.84–1.29; P =.37). Morbid obesity was associated with more than 30% significantly increased odds of BSO in all age strata.

Principal findings

This population-based cross-sectional study, including more than 44,500 patients and 700 unique surgeons, suggests that BSO at benign abdominal hysterectomy remains common overall, varies markedly between surgeons even after controlling for differences in patient case mix, and often lacks a clear diagnostic indication in hospital discharge records. Between-surgeon variation was particularly prominent among perimenopausal women; surgeons accounted for 22% of the residual observed variation in BSO in this age group.

Results and implications

Our study identified marked between-surgeon variability in BSO, indicating that the individual surgeon providing care is one of the strongest factors influencing whether a patient undergoes BSO (MOR, 2.0–2.5). Studies from other jurisdictions have noted a substantial variation in the use of BSO at the hospital level but have not evaluated variation at the physician level despite surgeons’ direct involvement in preoperative counseling and intraoperative decision making. We found that between-surgeon variability was significant in all age strata but the highest in women aged 45–54 years, regardless of whether they were <50 or ≥50 years. This result likely reflects persistent uncertainty among surgeons in how to allocate BSO, particularly as women transition to menopause. ^, Observational studies to date have not established a definitive age threshold for consideration of ovarian conservation vs removal. Practice guidelines offer little detail on whether BSO should be performed or withheld for women in menopausal transition or menopause and what valid indications are. Other potential contributors to the between-surgeon variation observed include differences in patient preference or case mix not captured by our data sources. Future research that examines the root causes of between-surgeon differences in BSO and that generates stronger evidence on the risk-to-benefit ratio of BSO as women age may help optimize strategies to reduce unwarranted variation.

In addition, we demonstrated that BSO remains relatively common in contemporary practice despite a large, well-known body of literature highlighting the potential consequences of surgical menopause since as early as the 1980s. ^, In our study, BSO was performed on 17% of patients aged <45 years and 44% of patients aged 45–54 years, and approximately 40% of all cases were considered potentially avoidable based on the diagnoses documented in patients’ discharge records. We also found that numerous non-gynecologic factors, such as low-income residence, obesity, and comorbidity burden, were associated with BSO, suggesting possible disparities in the provision of surgical care. Similar findings have been reported previously. Mahal et al showed that 38% of premenopausal women undergoing BSO at benign hysterectomy in California from 2005 to 2011 had no clear indication for the procedure. Several American studies have shown that non-gynecologic factors, such as race and insurance provider, were significantly associated with BSO in comparable cohorts cohorts from 1979 to 2010. It is concerning that these findings persist in our population-based study that provides a comprehensive picture of contemporary practice. The selection of patients for BSO must improve, and both knowledge translation and quality improvement initiatives directed at this issue are needed.

Strengths and limitations

Our study examined a contemporary, population-based cohort of women undergoing both inpatient and outpatient hysterectomy that enabled accurate assessment of practice patterns, virtually eliminated the risk of selection bias, and ensured that our results will be highly relevant to other jurisdictions. Studies to date have used databases limited to inpatient settings, known to offer an incomplete picture of overall trends. By using a multilevel modeling approach and provider billing claims uniquely available in Ontario, we were able to directly quantify surgeon-level practice variation in BSO. In addition, our data sources enabled identification of unilateral vs bilateral salpingo-oophorectomy, which is not often possible in American health service data.

Several limitations merit consideration. First, administrative data sources used in this study did not have information on menopausal status, which may have influenced decisions with respect to BSO and contributed to differences in case mix between surgeons. Because the onset of menopause is highly correlated with age, we adjusted for age within each stratum as a surrogate and performed sensitivity analyses using age 50 years as an additional cut point, which showed similar results. Second, our data sources lacked patient-level details on intraoperative findings, gynecologic disease severity (eg, endometriosis), family history, and genetic testing that would be needed to completely evaluate the appropriateness of every BSO performed. Therefore, we classified BSO as potentially avoidable only, by adapting the methodology developed by Mahal et al for discharge abstracts that was based on an exhaustive review of diagnostic codes that would support BSO. Third, we lacked detailed data on patient race, which has been associated with BSO in other settings. To mitigate this, we adjusted for South Asian and Chinese ethnicity using a validated algorithm available in Ontario. Finally, our study was done in a Canadian province where all patients have universal access to health services and may not be generalizable to practice in the United States. However, gynecologists in both countries rely on guidelines from the same specialty societies, and we do not anticipate that between-surgeon variation would be lower in the United States, where there may be additional insurance-related drivers of surgical care. In addition, it is important to note that none of these limitations are likely to explain the differences in between-surgeon variation observed.

Conclusions

Despite decades of published literature highlighting the potential harms of BSO and recent clinical practice guidelines recommending ovarian preservation at the time of benign hysterectomy, rates of BSO remain high and vary substantially between surgeons even after controlling for differences in patient case mix. These findings indicate a need for quality improvement and continued uptake of ovarian preservation. Future research should focus on generating high-quality data on the risk-to-benefit ratio of BSO in perimenopausal women; such evidence may help reduce uncertainty in practice for this population.