Integrating surveillance data to estimate race/ethnicity-specific hysterectomy inequalities among reproductive-aged women: who’s at risk?

Abstract

Background:

Inequalities by race and ethnicity in hysterectomy for non-cancerous conditions suggest that some subgroups may be shouldering an unfair burden of procedure-associated negative health impacts. We aimed to estimate race- and ethnicity-specific rates in contemporary hysterectomy incidence that address three challenges in the literature: exclusion of outpatient procedures; no hysterectomy prevalence adjustment; and paucity of non-White and non-Black estimates.

Methods:

We used surveillance data capturing all inpatient and outpatient hysterectomy procedures performed in North Carolina from 2011-2014 (N=30,429). Integrating data from the Behavior Risk Factor Surveillance System and United States Census population estimates, we calculated prevalence-corrected hysterectomy incidence rates and differences, by race and ethnicity.

Results:

Prevalence-corrected estimates show that Non-Hispanic (nH) Blacks (62, 95% confidence interval [CI]: 61-63) and nH American Indians (85, 95%CI: 79-93) per 10,000 person-years [PY] had higher rates, compared to nH Whites (45 [95%CI: 45-46] per 10,000 PY), while Hispanic (20, 95%CI: 20-21) and nH Asian/Pacific Islander rates (8, 95%CI: 8.0-8.2) per 10,000 PY were lower than nH Whites .

Conclusion:

Through strategic surveillance data use and application of bias correction methods, we demonstrate wide differences in hysterectomy incidence by race and ethnicity.

INTRODUCTION

Hysterectomy, or uterus removal, is a common surgical procedure ¹ with major consequences for health and well-being, including fertility cessation and increased risk of emotional strife ^2,3, cardiovascular events, and mortality ^4–6. Emerging evidence suggests that hysterectomy at early ages may be associated with particularly detrimental impacts on health ^7–11. Racial differences in hysterectomy incidence are widely acknowledged, particularly that Black women are more commonly treated with hysterectomy than White women ^12–14. However, many hysterectomy rates are based on incomplete rate numerators because they rely on inpatient data for a procedure that is increasingly performed in an outpatient setting. Further, estimated rates may also have incorrect denominators by not accounting for those who have already had a hysterectomy and are therefore no longer eligible for the procedure ¹⁵. Both procedure setting and prior hysterectomy differ by race and ethnicity, biasing comparisons across subgroups. Given this, current hysterectomy incidence rates and inequality measures may be inaccurate and worth updating.

Three issues need to be addressed to create more accurate hysterectomy rates. The data commonly used to estimate hysterectomy rate numerators are not population-representative because they often draw only from inpatient records. However, between 2000 and 2014 the proportion of hysterectomies performed in outpatient settings increased from 14% to 70% ¹⁶. Because Black women are more likely to receive surgeries in inpatient settings ^17,18, studies drawing only from inpatient samples may be underestimating overall and race- and ethnicity-specific rates and simultaneously overestimating Black–White differences. Second, no studies have accounted for the differential prevalence of hysterectomy by race and ethnicity. Incidence rate denominators are generally defined from age- and race-specific census counts of women and thus include person–time for women that have already had a hysterectomy, and who are no longer at risk for the procedure. The extent of denominator overestimation differs by race and ethnicity due to differential hysterectomy prevalence by race and ethnicity. Lastly, there is need for accurate hysterectomy rate estimates across a more diverse set of racial and ethnic subgroups. The majority of information about racial and ethnic differences come from Black–White comparisons. This is problematic because the experiences of other racial and ethnic subgroups may differ. Indeed, a small body of literature suggests American Indians have a higher prevalence of hysterectomy ¹⁹, are treated with hysterectomy at younger ages ²⁰, and are more likely to regret their decision to be sterilized than women of other racial and ethnic subgroups ²¹. Hispanic and Asian American women, on the other hand, appear to be less likely than White women to be treated with hysterectomy, with Asian American women having very low prevalence ^12,13,22.

Socially disadvantaged racial and ethnic subgroups generally have fewer health-relevant resources and therefore are likely to be under-users of health care systems when compared with Whites ²³. While reduced access to resources might be expected to lead to lower hysterectomy rates, disempowerment in medical decision-making, the complex history of forced sterilization in the US south, suboptimal care at earlier disease stages, and limited access to less invasive, uterus-sparing alternative treatments might be expected to have the opposite effect. Regardless of the reason behind elevated rates among some subgroups, potential over-treatment with certain types of medical care is an understudied topic that deserves more attention, ²⁴ particularly because in this case higher procedure rates are associated with higher incidence of adverse procedure-related events. As such, treatment with hysterectomy makes for an interesting and potentially illuminating study. Through the strategic combination of existing surveillance data sources, we provide timely population-level estimates of treatment with hysterectomy. Specifically, this paper aims to refine race- and ethnicity-specific hysterectomy incidence rate estimates among reproductive aged women by: 1) improving the rate numerators by combining comprehensive in- and outpatient surveillance datasets, 2) improving the rate denominators by using surveillance data sources that help to account for those not at risk of hysterectomy, and 3) estimating rates for a wider array of race and ethnicity subgroups than is typically the case.

METHODS

Data sources & associated measures

We drew the primary data for this study from two administrative databases containing medical billing records mandated for collection by the State of North Carolina (NC): the North Carolina Hospital Discharge database ²⁵ and the North Carolina Ambulatory Surgery database ²⁶ record procedures performed in inpatient and outpatient settings. Outpatient surgeries are those that occur in free-standing ambulatory surgery centers or those instances where patients are discharged within 24 hours, regardless of surgery location. The Cherokee Regional Hospital reports to the Indian Health Service, not to the State; however, since this hospital does not perform hysterectomy, its exclusion does not affect our estimates. Once consolidated, the database included the universe of hysterectomies performed in NC between 2011 and 2014, and provided the incidence rate numerators.

We identified hysterectomy procedures using Current Procedural Terminology (CPT) codes for outpatient settings and International Classification of Disease, 9^th revision (ICD-9) procedure codes for inpatient settings (eTable A); both are known to have high sensitivity and specificity ²⁷. Women’s age at the time of hysterectomy was reported in years since birth. State of residence at time of hysterectomy was indicated using Federal Information Processing Standards (FIPS) codes.

Numerator hysterectomy counts were categorized by race and ethnicity. We used the racial and ethnic categorization employed by the US Census and the Office of Management and Budget: Black non-Hispanic (nH), White (nH), American Indian/Alaskan Native (nH), Asian/Pacific Islander (nH) and Hispanic. We used these categories not as indicators of biologic differences, but as indicators of the sociopolitical realities and histories that accompany various racial identities ^28,29. Starting on 1 January 2010, a NC General Statute mandated collection of self-reported race and ethnicity with its medical data. This mandate reduced missingness of race and ethnicity, though it is unclear the extent to which race and ethnicity were administrator-reported rather than self-reported.

The starting point for the naive rate denominators (i.e., prior to prevalence correction) were population estimates provided by the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program, which, in partnership with the US Census Bureau’s Population Estimates Program and the National Center for Health Statistics, publishes population statistics that provided population estimates by year, 1-year age groups, and race and ethnicity for NC women.

To estimate NC’s hysterectomy prevalence for denominator adjustment, we used NC-specific data from every other year (2010-2014) of the Centers for Disease Control and Prevention’s (CDC) Behavioral Risk Factor Surveillance System (BRFSS), which included the question, “Have you had a hysterectomy?”. The BRFSS is an annual telephone survey that provides state-level prevalence estimates for the leading causes of mortality and morbidity among non-institutionalized adults (ages 18+). BRFSS survey weights were assigned to account for the probability of being selected to participate and demographic factors to make sure that the weighted estimates are representative of the target population: NC female adult residents under the age of 45.

Context & study population

NC provides an important context for this analysis because it: 1) is a racially diverse state that, in addition to a large population of Blacks, includes a large population of American Indians and, more recently, Hispanics and Asians; 2) has a state mandate to systematically collect medical billing data with race and ethnicity information; and, 3) has high historical and contemporary hysterectomy rates. NC has one of the largest populations of American Indians east of the Mississippi and is home to one federally recognized and seven state-recognized tribes. Because most of NC’s American Indian population does not live on federal reservation land, mirroring the experience of the broader US American Indian population, this research provided an opportunity to explore the contemporary experiences of American Indians. Moreover, since 1990, NC has had a substantial flow of immigrants and thus provides insight into hysterectomy rates among Hispanics and Asian Americans.

The source population included all women between the ages of 18 and 44 receiving a hysterectomy in NC between 2011 and 2014. We used age 18 as the lower limit to ensure there were corresponding hysterectomy prevalence estimates to use for denominator adjustment, while age 44 was used as a conservative proxy for pre-menopausal status ³¹. The study population was restricted to pre-menopausal women because, for providers and patients, the decision-making processes around hysterectomy differs from post-menopausal women and the detrimental impacts of hysterectomy are particularly felt when hysterectomy occurs at younger ages ^7,8,10. Previous research has also observed that racial and ethnic differences are most pronounced for pre-menopausal women and that patterns for post-menopausal women are qualitatively different ¹⁷. Women were excluded if their hysterectomy resulted from trauma (e.g., as a result of an automobile accident) (n = 8, 0.0%) or if not a NC resident at the time of the procedure (n = 4033, 11.0%). We excluded cases with missing race or ethnicity (n = 355, 1.0%) or listed as “other” race (n = 451, 1.2%) because the denominator data did not include an “other” category.

We also excluded women with presence of a malignancy (n = 1539, 4.2%) (eTable A), leaving only women with non-cancerous diagnoses. Hysterectomy occupies a complicated clinical context. As such, hysterectomy may not always be the most ideal treatment option, but can be the best option in some cases (e.g., life-threatening post-partum hemorrhage or intractable chronic pain). In other cases, optimal care may be harder to identify. There is likely a spectrum of clinical need and patient preference that influence appropriateness for treatment with hysterectomy, with some cases being very clear that hysterectomy is a reasonable option, others where it is not, and many that are more ambiguous. When clinical guidance is not clear, decision making becomes based upon a mix of patient preference, severity of symptoms, and fertility intentions, for example. There are many relevant values and perspectives, some of them competing, that should be considered within this treatment process. We exclude emergent and cancer-related hysterectomy, for which clinical need and appropriateness are much less ambiguous and racial and ethnic inequalities determined by a different set of processes. The final analytic dataset contained 30,429 hysterectomy procedures.

Analytic approach

To estimate numerators for crude hysterectomy incidence rates, we first stratified by race and ethnicity and combined data across four years (2011-2014). We then divided the race- and ethnicity-specific aggregate count by four to get an average one-year incidence estimate of hysterectomy procedures, by race and ethnicity, for the 2011 to 2014 period. We took the same approach to calculate the naïve rate denominators, which we estimated by averaging the counts of NC women between ages 18 and 44 years between 2011 and 2014 from SEER data. Race- and ethnicity-specific numerator counts were divided by their complementary race- and ethnicity-specific denominator counts to produce crude, state-level race- and ethnicity-specific rate estimates. Race-aggregated hysterectomy incidence has been decreasing in recent years ¹⁶, though our analyses of year and race-specific rates suggested the decrease in the NC-specific race-aggregated hysterectomy rate was driven by the decrease in the White-specific rate. Ultimately, average 1-year rates were estimates because rates among the non-White subgroups were similar over the four years ¹⁷. We multiplied estimates by 10,000 for interpretation as rates per 10,000 person-years (PY) of exposure time.

To more accurately calculate the at-risk population for prevalence corrected rates, we removed the proportion of NC women estimated to no longer have a uterus from the naïve rate denominators. To do this, we supplemented race- and ethnicity-specific US Census counts with race- and ethnicity-specific hysterectomy prevalence estimates from the BRFSS. We calculated prevalence estimates for each racial and ethnic subgroup by stratifying by race and ethnicity, then pooling sample-weighted counts of NC women in 2010, 2012, and 2014, between ages 18 and 44 years. This process resulted in one hysterectomy prevalence estimate for each racial and ethnic subgroup. These prevalence estimates were applied to naïve rate denominators once rates were age adjusted. We used direct standardization to adjust rates to the statewide 2010 NC population of women 18 to 44 years, by 1-year age groups.

Using the three types of rate estimates (crude, age-adjusted, and prevalence-corrected), we then calculated race and ethnicity-specific rate differences using nH Whites as the referent. Our language to describe racial and ethnic differences in rates is intentional. We followed previous literature in defining disparities as “systematic, plausibly avoidable health differences adversely affecting socially disadvantaged groups” ³². Because it is unclear whether high or low hysterectomy rates are ideal, we preferred the term inequality to label mathematical dissimilarity in rates (e.g., a rate of 4 per 10,000 PY is not the same as a rate of 10 per 10,000 PY). We used differences to refer to the actual measures, or estimates, (e.g., the rate difference is 6 per 10,000 PY, [10/10,000 PY minus 4/10,000 PY]). Last, to explore the impact of racial and ethnic inequalities in health status as reason for rate inequalities, we calculated Charlson Comorbidity Indices (CCI) using the 2011 Quan algorithm update for each procedure, and then calculated the mean score for each racial and ethnic group ^33,34. The CCI provides, in a single measure, an individual’s comorbidity burden. The higher the score, the higher the burden.

We used bootstrapping (percentile interval method) to estimate 95% confidence intervals for our prevalence corrected rates and rate differences that reflect uncertainty associated with the BRFSS hysterectomy prevalence estimates ³⁵. For each race and ethnicity, we simulated a dataset with the same mean and variance as the race- and ethnicity-specific BRFSS hysterectomy prevalence estimates. We assumed a normal distribution, given the large BRFSS sample size. We then took 10,000 random samples from the simulated dataset and used them to calculate 10,000 new rate denominators. Once 10,000 new rates were calculated with these denominators, we ordered the rates from smallest to largest and took the 5^th and 95^th percentile values as the bootstrapped 95% confidence intervals. The same process was used to calculate 95% confidence intervals for the rate differences. We did not include confidence intervals for crude or age-adjusted rate estimates because the hysterectomy count data capture the universe of hysterectomy procedures occurring within NC. All analyses were conducted using R (R Studio, Boston, Massachusetts) ³⁶. This research was approved by the University of North Carolina at Chapel Hill Institutional Review Board (# 17-2166).

Sensitivity Analysis

To explore the impact of cancer exclusions on the estimated rates of hysterectomy, we adjusted numerators by including incident hysterectomy for cancer-related diagnoses. We also explored the impact of 1) misclassification of race and ethnicity within numerators, 2) underreporting in SEER data by race and ethnicity within denominators, and 3) BRFSS underestimation of hysterectomy prevalence, assuming plausible levels of bias. For these sensitivity analyses, we started with the age and prevalence corrected rates and then re-classified proportions of procedures and person–time by race and ethnicity using estimates found in the literature. Last, because 11% of procedures in NC were among non-residents, and a similar percent of NC residents may be leaving NC for treatment, we also compared rates among NC residents living in border and interior counties.

RESULTS

Population & context

Table 1 includes data sources and their respective roles in hysterectomy rate estimation. Using hospital discharge databases, we observed the greatest number of hysterectomy procedures was among nH White women (4,682, 62%) and the smallest among nH Asian/PI women (50, 1%). Supplemental eTable B provides year-by-year numerator and denominator counts. Mean CCIs, by race and ethnicity, ranged from 0.66 (nH American Indian/Alaska Native) to 1.41 (nH Black). Supplemental eFigure A provides the score distribution, by race and ethnicity.

Table 1:

Data sources and associated information used to build hysterectomy rate estimates among reproductive-aged women in North Carolina by race and ethnicity, 2011-2014

	Total	Black, nH	Hispanic	AI/AN, nH	Asian/PI, nH	White, nH
Hysterectomy procedures completed (ages 18-44 years) [purpose: rate numerators]
Data Sources: 2011-2014 North Carolina Hospital Discharge & Ambulatory Surgery Visit Databases
Outpatient	20,864	5,545	716	375	138	14,090
Inpatient	9,565	4,027	512	324	63	4,639
Total	30,429	9,572	1,228	699	201	18,729
Yearly mean, 2011-2014	7,607	2,393	307	175	50	4,682
Population size (women ages 18-44 years) [purpose: rate denominators]
Data Sources: 2011-2014 SEER Population Estimates
Yearly mean, 2011-2014	1,785,292	434,652	176,151	23,211	63,496	1,087,783
Hysterectomy prevalence (women ages 18-44 years) (95% CI) [purpose: prevalence estimates for denominator adjustment]
Data Sources: 2010, 2012, 2014 Behavioral Risk Factor Surveillance System
Pooled estimate, 2010, 2012, 2014	6% (4.9-7.1)	6% (4.5-7.9)	5% (2.5-7.7)	10% (2.2-17)	1% (0.043-2.1)	6% (5.1-7.1)

Abbreviations: nH = non-Hispanic; sd = standard deviation; AI/AN = American Indian/Alaskan Native; PI = Pacific Islander; CI = Confidence Interval

Hysterectomy prevalence estimates were weighted using BRFSS sampling weights, 95% CI were calculated after pooling data across three time points (2010, 2012, 2014) and using the approach outlined in Lewis, 2017 ⁴⁷.

Population-relevant race- and ethnicity-specific hysterectomy rates and rate differences

NC’s overall hysterectomy rate decreased over time from 47/10,000 PY in 2011 to 40/10,000 in 2014. Figure 1 shows crude, age-adjusted, and prevalence-corrected hysterectomy incidence rates. Non-Hispanic AI/AN consistently had the highest rates, while nH Asian/PI had the lowest. Regardless of the adjustment strategy, nH Blacks and nH AI/AN had the greatest number of hysterectomy procedures per person–time at risk (i.e., positive rate differences), while Hispanic and nH Asian/PI rates were lower (i.e., negative rate differences) (Figure 2).

Figure 1:

Comparison of race- and ethnicity-specific hysterectomy incidence rate estimates, calculated using three approaches. Crude rates (naïve denominators) are shown with orange circles, age-adjusted rates (naïve denominators) are shown with dark blue bars, and age-adjusted rates (prevalence corrected denominators) with light blue triangles.

Abbreviations: nH = non Hispanic; AI/AN = American Indian/Alaskan Native; PI = Pacific Islander; PY = Person Years

Data Sources: 2011-2014 North Carolina Hospital Discharge & Ambulatory Surgery Visit Databases; 2010-2014

SEER Population Estimates; 2010, 2012, 2014 Behavioral Risk Factor Surveillance System (BRFSS)

95% CI calculated using percentile interval method of bootstrapping

Incidence rates (per 10,000 person years) of hysterectomy among reproductive aged women in North Carolina by race and ethnicity, 2011–2014

Figure 2:

Comparison of race- and ethnicity-specific hysterectomy incidence rate differences (referent group: non-Hispanic White), calculated using three approaches. Crude rate differences (naïve denominators) are shown with orange circles, age-adjusted rate differences (naïve denominators) are shown with dark blue bars, and age-adjusted rate differences (prevalence corrected denominators) with light blue triangles.

Hysterectomy rate differences among reproductive aged women in North Carolina by race and ethnicity (referent group: nH White), 2011-2014

Abbreviations: nH = non Hispanic; AI/AN = American Indian/Alaskan Native; PI = Pacific Islander; PY = Person Years

Data Sources: 2011-2014 North Carolina Hospital Discharge & Ambulatory Surgery Visit Databases; 2010-2014

SEER Population Estimates; 2010, 2012, 2014 Behavioral Risk Factor Surveillance System (BRFSS)

95% CI calculated using percentile interval method of bootstrapping

Impact of adjustment for baseline prevalence of hysterectomy

Using the BRFSS, we estimated the prevalence of hysterectomy in NC to be 7% (95% CI: 5.6-8.7) in 2010 and 5% (95% CI: 3.2-6.0) in 2014. We suspect the decrease is likely due to a true reduction in hysterectomy incidence and prevalence, as well as changes in BRFSS sampling within NC. Once pooled and stratified by race and ethnicity, nH AI/AN women had the highest prevalence, while nH Asian/PI women had the lowest. For within-group comparisons, prevalence correction resulted in rate increases for nH Black, Hispanic, nH AI/AN, and nH White women when compared to crude rates with naïve denominators (Figure 1: blue triangles vs. orange circles). Prevalence correction had the largest effect among nH AI/AN, nH Black, and Hispanic women: the crude nH Black rate was 11% lower than the corrected rate (55 vs. 62 [95% Confidence Interval (CI): 61-63)] per 10,000 PY), the crude nH AI/AN rate was 12% lower (75 vs 85 [95% CI: 79-93] per 10,000 PY), and the crude Hispanic rate was 13% lower (17 vs 20 [95% CI: 20-21] per 10,000 PY). Prevalence correction did not, nor did age adjustment, meaningfully change nH Asian/PI rate estimates (Table 2).

Table 2:

Hysterectomy incidence rates (per 10,000 person years), rate differences, and bias estimates among reproductive aged women in North Carolina by race and ethnicity, 2011-2014

	Race- and Ethnicity-Specific Rates				Racial and Ethnic Rate Differences (ref: nH White)
	Crude (naïve denominator)		Prevalence Corrected (95% CI) a	Percent & direction of bias in crude estimate b	Crude (naïve denominator)	Prevalence Corrected (95% CI) a		Percent & direction of bias in crude estimateb
Race and ethnicity
Asian/PI, nH	8	8	(8.0, 8.2)	−2%	−35	−37	(−38, −37)	−5%
Hispanic	17	20	(20, 21)	−13%	−26	−25	(−26, −24)	+2%
White, nH	43	45	(45, 46)	−5%	(ref)	(ref)	--	--
Black, nH	55	62	(61, 63)	−11%	12	17	(16, 18)	−28%
AI/AN, nH	75	85	(79, 93)	−12%	32	40	(34, 48)	−20%

Abbreviations: nH = non-Hispanic; AI/AN = American Indian/Alaskan Native; PI = Pacific Islander; CI = confidence interval; ref = referent

Data Sources: 2011-2014 North Carolina Hospital Discharge & Ambulatory Surgery Visit Databases;

2010-2014 Surveillance, Epidemiology & End Results (SEER) Population Estimates; 2010, 2012, 2014 Behavioral Risk Factor Surveillance System (BRFSS) Rates are per 10,000 person years

^aRates have been age-adjusted; 95% confidence intervals estimated using percentile interval method of bootstrapping

^bRounding to the nearest whole number may result in slight inconsistencies between rates and calculated bias estimates

Racial and ethnic inequalities (referent: nH Whites) calculated with crude rates underestimated the inequalities for all subgroups except Hispanics (Figure 2: orange circles vs. blue triangles). The crude nH AI/AN-White rate difference underestimated the inequality by 8 cases per 10,000 PY (20%), the crude nH Black–White rate difference by 5 cases per 10,000 PY (28%), and the crude nH Asian/PI-White by 2 cases per 10,000 PY (5%) (Table 2).

DISCUSSION

The accurate documentation of inequalities in surgical procedures – particularly invasive ones with long-lasting consequences, such as hysterectomy – is needed. This is particularly true when there are wide differences across racial and ethnic subgroups. In the case of hysterectomy, overall lack of healthcare access may not explain higher rates among young Black and AI/AN women, though structural disparities in gynecologic or reproductive-specific healthcare may amplify the inequalities. Often proposed explanations for Black–White rate differences include differences in clinical indications, age of symptom onset, and symptom severity ^12,13,18. However, these factors are difficult to quantify because symptom onset is often subclinical and symptom severity is not well-captured in many studies. In addition to these explanations, we believe that the rates among Black and AI/AN women – two historically, socially, and economically marginalized subgroups ^37,38 – may point to limited access to less invasive, uterus-sparing treatments ^39,40; or, racial and ethnic variation in patient preferences against the backdrop of historically high hysterectomy rates in these populations; or other considerations such as limited paid time off work and restricted access to contraception. Regardless of the source of inequality, accurate estimates of the magnitude of inequalities are imperative to support further investigations.

Despite the high incidence of hysterectomy, reliable race- and ethnicity-specific rate estimates are difficult to find. Prior to the transition to outpatient settings in the late 1990s, inpatient clinical samples provided a reasonable account of hysterectomy incidence. While hospital-based studies may still seem a promising data source, they neglect outpatient procedures where the proportion of hysterectomy procedures differs from those done inpatient. Indeed, age-standardized inpatient hysterectomy rates in 2013 NC were 10/10,000 PY while outpatient hysterectomy rates were 28/10,000 PY. Inpatient and outpatient rates also differed by race, ethnicity and age, with Black women experiencing higher inpatient rates at earlier ages ¹⁷. The most accurate contemporary estimates are those that consider both inpatient and outpatient procedures.

While hysterectomy incidence is decreasing over time ^16,18, it is still the second most commonly performed surgical procedure among women ages 18-65 years ¹. We estimated that contemporary rates among pre-menopausal Black women in NC are only slightly lower than national-level estimates among women of all ages from 1988-1990 ¹⁴. Conversely, we estimated rates among pre-menopausal White women in NC to be much lower than national-level estimates among women of all ages from 1988-1990 ¹⁴. Unfortunately, older estimates for other subgroups are not available for comparison. Given that hysterectomy incidence has remained fairly steady, at least among Black women, prevalence adjustment is necessary to produce rates that are not underestimated ¹⁵. We found baseline prevalence adjustment to have the greatest impact on nH Black and nH AI/AN rate estimates, which is expected given these subgroups are known to have higher hysterectomy prevalence. Prevalence adjustment also mattered for rate difference estimates, as those were also underestimated without adjustment.

Consistent with previous literature, we observed a great deal of racial and ethnic variation in hysterectomy rates. This variation did not waver when including incident hysterectomy for cancer diagnoses. Racial and ethnic inequalities in rates are compatible with many scenarios and previous literature has not adequately addressed their causes ^12,13,41,42. Studies tend to focus on differences resulting from higher or lower prevalence, than Whites, of the different non-cancerous conditions treated by hysterectomy. Unfortunately, this focus on biologic explanations has come at the detriment of building a deeper understanding of the roles of institutionalized and interpersonal forms of racism and classism or historical context in contributing to the racial patterning of contemporary gynecologic health. For example, there may be historical and continued discrimination and devaluing of bodies and fertility of Indigenous and women of color within the medical system. North Carolina’s state-sanctioned eugenics program of the 20^th century disproportionately sterilized the same subgroups that now show contemporary excess hysterectomy rates. Women may also experience stigma when seeking reproductive health care and some studies suggest that interpersonal interaction and communication with providers may steer women towards or away from hysterectomy ^{12,13,41–43}. It is also possible that these factors operate simultaneously or that their significance matters differently for different racial and ethnic subgroups.

Conducting a de novo study to estimate population-based race- and ethnicity-specific rates of hysterectomy would be costly and inefficient. Rather, we strategically combined several existing surveillance data sources to create accurate rates. Through this work we addressed three primary challenges in the existing hysterectomy rate literature – exclusion of outpatient procedures, no hysterectomy prevalence adjustment, and paucity of non-White and non-Black estimates – to reveal inequalities and underestimates of hysterectomy in five racial and ethnic subgroups.

Limitations

Beginning in 2011, NC mandated self-reported race and ethnicity with medical billing data. However, these data may include racial and ethnic misclassification since administrators, rather than patients, may be reporting race and ethnicity. Unfortunately, no linkages with these data are allowed, limiting our ability to investigate the validity of race and ethnicity in these data, as have been done in mortality studies ⁴⁴. Previous reports suggest that subgroups most likely to be misclassified in medical records are AI/ANs and Hispanics ⁴⁵. Focusing on these two subgroups for sensitivity analysis, the Hispanic and AI/AN-specific rates we report in this manuscript may be underestimates (eTable C). Relatedly, the denominator estimates may be inaccurate, particularly for highly mobile or undercounted race and ethnicity groups. After adjusting denominators based on differential Census omissions by race and ethnicity, our sensitivity analyses indicate that our reported Black and AI/AN-specific rates may be overestimates (eTable C). Last, if there are differences in the methods used to record race and ethnicity in the different datasets, the numerator and denominator data may be imperfectly combined.

BRFSS data are the best source for nationally representative hysterectomy prevalence estimates, but present some challenges. We were not able to calculate stable race- and ethnicity-specific prevalence estimates for each year, particularly among nH Asian/PI and nH AI/AN subgroups, due to small cell sizes. Because of this, we had to pool 3 years of data. Despite pooling, if prevalence is still underestimated, our adjusted rates also remain underestimated due to enduring inflated denominators. Additionally, BRFSS-specific sampling patterns may result in underestimation of NC-specific hysterectomy prevalence, particularly among non-White subgroups. Potential hysterectomy prevalence underestimation in BRFSS leads to slight underestimation of Black and AI/AN rates (eTable C), so some inequalities may be even larger than we estimated them to be. We were also not able to capture hysterectomy procedures among NC residents that occurred outside of NC borders, though rates among those residing in counties bordering other states are not lower than those residing in interior counties (eTable D). Last, given our data, we were not able to estimate Institute of Medicine concordant disparity estimates ⁴⁶. However, to explore the impact of baseline health status on race- and ethnicity-specific rate estimates, we used Charlson Comorbidity Indices (CCI) and observed minimal differences in score by race and ethnicity (eFigure A). Similarities in score are not surprising since women treated with hysterectomy are deemed healthy enough for surgery and, further, the CCI was designed to capture comorbidities within mortality studies, rather than in gynecologic contexts.

Conclusion

Our research provides timely population-level estimates of treatment with hysterectomy for non-cancerous gynecologic conditions. While rates have decreased over time, inequalities by race and ethnicity endure and may be particularly problematic because hysterectomy at younger ages carries later-life risks. Hysterectomy rates remain disproportionately high among subgroups who previously had high rates, including Black women. Our research also demonstrates that nH AI/AN women experience very high rates, while nH Asian/PI and Hispanic women exhibit low rates relative to nH White women. These inequalities are important to note. Further research is necessary to understand the drivers of these differences to ensure both that women are treated when they need hysterectomy and that unnecessary hysterectomy is avoided.