Practice-Level Severity Case Mix and Treatment Patterns for Premenopausal Noncancerous Hysterectomy.

Below, we describe how a case series of premenopausal and non-cancerous hysterectomy patients was formed by linking: 1) extracted structured electronic health record (EHR) data, including procedure and diagnostic codes, lab values, and demographic data; 2) manually abstracted EHR data, allowing detailed assessment of patient symptoms; and 3) physician licensing data, identifying the practice of the surgeon leading each patient’s hysterectomy (see Figure 1 for an overview). The Carolina Hysterectomy Cohort (CHC) consists of a case series of premenopausal patients aged 18 to 44 years who underwent hysterectomy for non-cancerous conditions in one of 10 hospitals affiliated with a large healthcare system in a southeastern US state between October 2nd, 2014, and December 31st, 2017. 19 – 23 The study and eligibility criteria are described in detail elsewhere. 19 CHC leverages structured and unstructured electronic health records (EHR) data to provide data on patient demographics, diagnosis and procedure codes associated with hysterectomy, relevant treatments and procedures used to treat gynecologic conditions in the years before surgery, and previous gynecologic healthcare encounters. Manual abstraction of unstructured data (including patient progress notes, operative notes, and imaging and pathology reports) enabled rich measurement of current and previous symptoms, imaging and pathology findings, and primary reasons for surgery. 24 Initially, 1857 hysterectomies were identified, and patients were excluded if they identified as belonging to a race and ethnicity group that was too small for meaningful analysis (n=92) or had missing covariates (n=50). Some further exclusions because of practice-level information are detailed below. For each hysterectomy, logs from the operating room recorded the names of surgeons involved, including the “billing” surgeon and, potentially, additional “primary” surgeons , residents, or fellows. We designated the billing surgeon to be the surgeon of interest (“lead” surgeon) for our analysis, as we focus on decision-making about whether to conduct surgery. In cases with multiple billing surgeons, the “primary” surgeon was considered the designated “lead” surgeon. Once a lead surgeon was identified for the hysterectomy, p ractices were identified by the address given for their “primary practice” location. Distinct OBGYN practices at these addresses were identified with internet searches. Since surgeons change practices over time, surgeries were linked to the closest submission of the lead surgeon’s licensing record ( Supplemental Appendix 1 ). Most addresses were uniquely associated with a practice, but a large academic medical center had multiple practice groups at the same address. Here, practice specialty noted by the lead surgeon (e.g., Urogynecology, Minimally Invasive Gynecologic Surgery, General Obstetrics & Gynecology), was used with the address to identify practices corresponding to OBGYN departmental divisions. Our team included OBGYN MD collaborators with long-standing clinical practices in the state who reviewed the lead surgeon and practice identification results, adjudicated, and confirmed the final list of practices included. Practices (and their patients) were excluded if <6 CHC patients underwent hysterectomy there during the study period (to reduce patient identifiability; n=115 patients), or if they specialized in gynecological oncology (since surgical decision-making in these practices may include broader considerations than the severity of gynecological symptoms, such as cancer risk assessment; n=10 patients). Thus, our final sample included 1590 individual patients linked to 100 surgeons at 20 distinct practices. We capitalized on detailed, validated pre-surgical symptom severity data to create a novel measure of patient symptom severity case mix. 24 Each patient was scored for severity of symptoms in three domains: uterine bulk (pressure), vaginal bleeding, and pelvic pain. Development and validation of these severity scores are explained elsewhere. 24 We classified patients as lower-severity if their symptom scores were below the median for all three domains. We calculated the proportion of lower-severity patients treated at each practice (range: 0–24%). We examined the distribution of lower-severity patient proportions across 21 practices (in this initial stage, we included a gynecology-oncology practice) for a natural cut-point that would distinguish between practices operating at higher and lower severity (see results ). Patient race and ethnicity was derived from two EHR variables that were self-reported: 6-level race (White, Black, Asian, American Indian/Alaska Native /Pacific Islander, Other, Refused/Unknown) and the dichotomous Hispanic ethnicity (Yes/No). Patients whose ethnicity was Hispanic were coded as Hispanic regardless of race. Patients who identified as White and another race were classified as the other race (i.e., not White). Analyses were restricted to non-Hispanic White, non-Hispanic Black, and Hispanic patients (hereafter: White, Black, and Hispanic) since there were not enough patients for meaningful analysis from any other single racial group. Patients were categorized into three insurance groups: public, private, and uninsured. The public insurance category comprised patients covered by Medicare (N = 61), Medicaid (N = 198), or those receiving care in prison facilities (N = 10). Private coverage included patients with commercial insurance plans (N = 1157 ) and Tricare beneficiaries (N = 47), the latter covering military service members and their dependents. The uninsured category consisted of patients designated as “self-pay” in medical records, including those responsible for partial (n < 10) or complete payment (n = 117) of their care. We included two classes of covariates. The first class included patient characteristics that could influence decisions about performing surgery. Possible contraindications to hysterectomy include patient characteristics such as younger age, being nulliparous, and having tried no or few uterus-sparing alternative treatments . We counted previous uterus-sparing treatments prior to hysterectomy, including combined estrogen-progestin contraceptives, oral progestins, medroxyprogesterone acetate, contraceptive implants, oral tranexamic acid, gonadotropin-releasing hormone agonists, hormonal intrauterine devices, uterine artery embolization, endometrial ablation, and myomectomy. We also included hysteroscopy and laparoscopy for gynecologic indications with alternative treatments. The second class of covariates were factors that may influence whether a practice accepts a patient versus recommending referral to a different practice with more specialized experience. Most were indicators of surgical complexity (which is distinct from symptom severity) and included: Body Mass Index at the time of surgery (BMI: weight in kilograms divided by height in meters squared; categorized as 25–30, >30–35, >35–40, and >40 kg/m 2 ); uterine size (measured in grams from pathology reports, analyzed as a continuous variable); certain gynecological indications for hysterectomy captured from abstracted EHR notes with a lookback period ranging from six months to one year (such as the presence of endometriosis, chronic pelvic pain/dysmenorrhea, or abnormal uterine bleeding); and having had previous abdominal surgery (laparotomy), excluding cesarean sections (also from abstracted EHR notes with the same lookback period). We also included a binary indicator for hysterectomy hospital site location (academic vs. non-academic). We first descriptively explored the relationship between our novel measure of severity case mix, our two primary exposures (race and ethnicity and insurance status), and our covariates. Differences between hysterectomies performed at lower and higher severity practices were tested with chi-squared/fisher’s exact for categorical variables and the Wilcoxon test for continuous variables. As validation of our measure, we repeated this descriptive comparison after excluding lower-severity patients (with severity scores below the median for bulk, bleeding, and pain) to assess whether practice differences in patient characteristics were still evident when not considering the patients who determined severity case mix . For our primary objective, investigating differences by race and ethnicity and insurance in the severity case mix of practices where patients get treated, we used log-binomial regression models to estimate prevalence ratios (PRs) for the outcome/ dependent variable (lower vs. higher-severity case mix practices) in relation to patient race and ethnicity (reference group: White) and insurance (reference group: Private). We started with separate unadjusted models for each exposure (i.e. one model for race and ethnicity and one for insurance), then included both exposures in the same model, then adjusted sequentially for patient age at hysterectomy, gynecological indications for surgery, previous abdominal surgery, and the number of prior treatments attempted. There was some sparseness of data for insurance status in our analytical sample: practices with lower symptom severity case mix almost exclusively treated privately insured patients. As this could potentially cause estimation problems, we repeated analyses with race and ethnicity as the primary independent variable and restricted the sample to privately insured patients (n=1,393). Similarly, no academic practices had a low severity case mix so, rather than adjusting for hospital site, we repeated models with the sample restricted to patients treated at non-academic centers. Additional supplemental analyses were performed to better understand contributors to racial and ethnic differences in practice severity case mix . First, due to the high number of missing values (n =376, 23.5%) for the parity variable, we re-ran our fully adjusted models on the subset of respondents who had parity data and included parity as a covariate. Finally, we conducted a sensitivity analysis using generalized estimating equations with a robust variance estimator to account for potential correlations among surgeons within the same practice . For this sensitivity analysis examining clustering effects , we specified an independent working correlation matrix , as attempts to use compound symmetry resulted in correlation estimates of 1. All findings were interpreted based on data compatibility and in accordance with literature that discourages interpreting findings based solely on statistical significance 25 . This study was performed in line with the principles of the Declaration of Helsinki. This study was approved by the University of North Carolina (17–2728) and Duke IRBs (Pro00109220).

The analytic sample consisted of 1590 patients who self-reported as White (58.1%), Black (33.1%), or Hispanic (8.8%). Of these patients, 16.9% were publicly insured, 75.7% were privately insured, and 7.4% were uninsured. The distribution of lower severity patients across 21 practices, including the gynecological oncology practice at this preliminary stage (see reasons for excluding gynecology-oncology practice above), revealed a natural cut-point at 18% to distinguish practices operating at lower and higher severity (approximately 75 th percentile; Figure 2 ). The exclusion of gynecology-oncology practice did not alter the natural cut-point. The six practices identified as lower severity practices had treated 528 patients, while the 14 higher severity practices had treated 1062 patients (total of 1590 patients from 20 practices). The range in the number of patients treated by each practice was similar for lower and higher severity practices ( Supplementary Tables 1 and 2 ) ; 29 to 177 compared to 9 to 213). As expected, given how practices were categorized, patients at lower severity practices also had lower median scores for bleeding (4 vs.7) and pain (1.5 vs. 6) than patients treated at higher severity practices, although median bulk scores (1 vs. 1) were similar between lower and higher severity practices. The proportion of Hispanic patients was smaller in the lower than in the higher-severity practices (3.4% vs. 11.5%; Table 1 ). Patients at lower severity practices were primarily private insurance beneficiaries (96.2%), while almost a third of those in higher severity practices were either on public insurance (24.2%) or uninsured (10.3%). The median age at the time of hysterectomy for those at lower severity practices was 40.6 years, compared to 39.8 years at higher severity practices (i.e. a difference of <1 year and not clinically meaningful). Patients at lower compared to higher severity practices had received fewer alternative treatments prior to hysterectomy (32.2% vs 43.8% had undergone more than one prior alternative treatment). There was evidence that main indications for surgery for patients treated at lower compared to higher severity practices is less likely to include endometriosis (5.9% vs. 8.3%; P =0.083), chronic pelvic pain or dysmenorrhea (29.0% vs. 40.1%) and more likely to reference leiomyomas (47.7% vs. 33.5%) and abnormal uterine bleeding or menorrhagia (64.5% vs. 58.3%). BMI, parity, rates of prior abdominal surgery, and median uterine weight were comparable between lower and higher severity practices (though parity was recorded less often at lower severity practices). Notably, there were no lower severity practices treating patients at an academic medical center, while 59.9% of patients treated at higher severity practices were treated at an academic medical center. All of the above differences were still evident after excluding the lower severity patients who were used to define practice severity case mix ( Supplemental Table 3 ). After excluding these patients, there was actually stronger evidence that lower severity practices treated fewer patients with endometriosis as an indication (4.5% vs 8.3%; P =0.010). Descriptive statistics for patients by individual practices included in this study, showed the median number of uterine-sparing therapies was 1 in all lower severity practices, whereas it varied between 0 and 2 in higher severity practices ( Supplemental Tables 1 and 2 ). Hispanic compared to White patients were 63% less likely to be treated at a lower severity practice [PR: 0.37 (0.20 – 0.68)] ( Table 2 ). Further adjustment for insurance and surgically relevant indicators including BMI, gynecological indications, previous abdominal surgeries, and prior treatments attempted showed that Hispanic patients were still 48% less likely to be treated at lower severity practices than White patients [model 4: PR: 0.52 (0.33 – 0.82)]. There was no strong evidence of differences between Black and White patients in chances of being treated at lower severity practices in any of the models. Differences by insurance status were also substantial. In unadjusted analyses and compared to privately insured patients, publicly insured patients were 89% less likely to receive care at lower-severity practices [PR: 0.11 (95% CI: 0.03–0.35)] and uninsured patients were 84% less likely to be treated at lower-severity practices [PR: 0.16 (95% CI: 0.05–0.53)]. These differences persisted across all adjustment models, with publicly insured patients remaining 87% less likely [Model 4: PR: 0.13 (95% CI: 0.05–0.36)] and uninsured patients 72% less likely [Model 4: PR: 0.28 (95% CI: 0.12–0.68)] to receive care at lower-severity practices compared to privately insured patients. We did not find meaningful differences from the above models when including clustering effects for surgeons ( Supplemental Table 4 ). We repeated our final adjusted model (model 4 from Table 2 ) among the sub-sample of respondents who had parity data, including parity as an additional characteristic, and this yielded similar findings ( Supplemental Table 5 ). When restricting to patients not treated at academic centers, there was not strong evidence for Hispanic patients being less likely than White patients to be treated at lower severity practices, nor for uninsured patients being less likely than privately insured patients to be treated at lower severity practices. Restricting to privately insured patients, 24.3% of Hispanic patients were treated at lower severity practices, compared to 42.6% of White and 44.6% of Black privately insured patients ( Figure 3 ). The difference between Hispanic and White privately-insured patients was still evident in models with adjustment for clinically relevant factors [PR: 0.55 (0.34 – 0.89)] and with additional adjustment for parity [PR: 0.48 (0.25–0.93)] ( Supplemental Table 6 ). However, with additional restriction to non-academic centers, there was not strong evidence to support differences between Hispanic and White patients, with or without adjustment for parity.

With detailed information on pre-surgical severity of bleeding, bulk and pain symptoms for pre-menopausal hysterectomy patients with non-cancerous conditions, this study created a novel, practice-level measure of severity case mix distinguishing medical practices performing hysterectomy with higher proportions of cases (>=18%) with low symptom severity (i.e. severity scores below the median for all domains) . Practice severity case mix was associated with segregation of patients by race and ethnicity and insurance status. Specifically, Hispanic patients were less likely than White patients to be treated at lower severity practices, and, independent of this, publicly insured and uninsured patients were also considerably less likely than privately insured patients to receive treatment at such practices. Indeed, almost all (96.2%) of patients treated at lower severity practices were privately insured. Both differences were robust to adjustment for clinical and patient characteristics. Few studies examine healthcare practice-level determinants of receiving hysterectomy. Some prior studies have investigated geographic variation in hysterectomy rates among reproductive-aged women with non-cancerous gynecologic conditions, e.g., at the county or zip code level . 3 , 9 , 26 Such studies have found that hysterectomies are more common in economically disadvantaged areas 9 , and that differences between Black and White hysterectomy rates are less pronounced in such areas. 26 These previous studies tended to rely on claims databases, which exclude the uninsured, and provide limited information on symptom severity . We included all hysterectomies performed from 2014–2017 in a large, not-for-profit healthcare system (excluding outpatient practices) in the US South, serving a diverse racial and ethnic population, including many uninsured patients. Combining structured data (i.e., diagnostic and procedures codes) with abstraction of provider notes over the 12 months preceding hysterectomy enabled detailed measurement of pre-surgical symptom severity. 24 Novel linkage of individual-level EHR with state medical professional licensure data enabled aggregation of this rich symptom severity information into a practice-level measure of severity case mix . Severity case mix was associated with surgically relevant patient characteristics after excluding the lower-severity patients that the measure was based on , which validates the measure , supporting the idea that it identifies institutional segregation (interrelated issues of patient race and ethnicity, insurance, and possibly language preferences) of patients into distinct groups of healthcare practices. Our findings suggest financial incentives play an important role in segregation of care. P ractices with a lower severity case mix operated on few uninsured and publicly-insured hysterectomy patients. While all US healthcare systems are motivated to maximize revenues, some practices may tend to perform hysterectomies at lower symptom thresholds because profit incentives for hysterectomies result in higher revenue under the current reimbursement structure compared to continued outpatient medical management of symptoms . 27 , 28 The primarily privately-insured patients at lower-severity practices had also received fewer alternative treatments before hysterectomy than other patients, potentially indicating over-treatment of these patients . Leiomyomas and abnormal uterine bleeding were more prevalent as indications for surgery at these lower severity practices. Available uterine-sparing alternative therapies for these indications at the time may not necessarily have been curative, 29 offering symptom control that may decrease in efficacy over time. Thus, curative hysterectomy may have been a preferred choice for patients and, rather than aiming to maximize revenue, lower severity practices may have been fitting treatment to needs and preferences of their patients and emphasizing patient autonomy over recommendations to attempt uterine-sparring alternative treatments before hysterectomy. 1 , 17 Understanding the extent to which these novel findings represent practices emphasizing revenue, patient autonomy, or other mechanisms needs further study. Treatment site may be another key factor associated with segregation of gynecologic care. No practices at academic medical centers were classified as having a low severity case mix. Moreover, while Hispanic patients were less likely than White patients to be treated at lower severity practices , this difference was not present when restricting to non-academic locations. Academic center practices were integrated into the public healthcare system, with policies designed for the region’s heavily immigrant and uninsured Hispanic population, including language interpreter services and financial assistance programs. 30 – 32 Hispanic patients may have been more likely to get treated there than at lower severity practices because they lacked qualifying insurance, faced language or other social barriers, or because their high symptom burden 33 led to their being referred to or choosing to seek care at locations with more resources. Even among Hispanic patients who were privately insured, less than a quarter were treated at lower severity practices. If lower severity practices are prioritizing patient autonomy, then Hispanic patients may be less likely to receive such patient-led care. Future research should include patient interviews and examine the decision-making process to understand key factors leading to racial and ethnic differences in gynecological care prior to hysterectomy. We note the following limitations. Our case series design does not include patients who never received hysterectomy because uterine-sparing treatments were successful, limiting inference about differential access to care. Similarly, we had no reliable data on where patients initially sought treatment (only the hysterectomy location) or on whether patients were offered, covered for, or decided to forego uterine-sparing treatments. Furthermore, only documented symptoms were measured, so practice differences in symptom recording may have biased our measure of severity case mix . In regions with multiple practices, patient self-selection may shape practice case mix , potentially impacting generalizability of findings to settings with fewer or different options for treatment . With severity case mix measured at practice level and patient demographics and other characteristics captured at the individual level, outcomes for all patients within the same practice were identical. Adjusting for within-practice correlation in outcomes could yield nearly identical point estimates and introduce bias due to this lack of outcome variation at the practice level. 34 Instead, we accounted for correlations in outcomes between surgeons in a sensitivity analysis with similar findings. Critically, results of this study should not be interpreted causally. We aimed to descriptively examine how practice severity case mix , was associated with segregation of patient care by race and ethnicity and insurance status, without ascribing practice severity case mix as a cause of that segregation . In conclusion, we identified segregation of pre-menopausal patients with non-cancerous indications into practices with hysterectomy case mixes of higher and lower pre-surgical symptom severity. In this study, we did not find strong evidence of differences in practice severity case mix between Black and White patients, but Hispanic patients, the majority of whom were uninsured, were rarely treated at practices with a lower severity case mix . Almost all uninsured or publicly-insured patients were operated on at higher severity practices as they were most frequently treated at the academic center. Overall, the presence of practices performing hysterectomy at a lower severity case mix could reflect a mix of economic incentives, greater latitude for patient autonomy, and clinical differences in the patient mix, among other potential explanations. Practice-level factors may play an important role in socially stratified gynecologic care pathways in the US healthcare system.

Hysterectomy (surgical removal of the uterus) is the second most common surgical procedure among individuals with a uterus in the US (second only to cesarean sections) with over half a million individuals undergoing hysterectomy annually. 1 , 2 Hysterectomy incidence varies widely by race and ethnicity , age group (premenopausal vs. menopausal), geographic region, and clinical setting (inpatient/outpatient). Causes of these variations are poorly understood. 1 , 3 – 9 Most research on unexplained variation in hysterectomy focus on individual-level patient factors, like race and ethnicity 4 , 7 , 8 , education and income 10 , or adiposity 11 . It remains unclear how healthcare settings and institutions structure access to gynecologic healthcare in the US . 12 Hysterectomy for non-malignant conditions in premenopausal women is sensitive to patient preferences. There are no definitive, objective markers establishing clear standards of care for these hysterectomies . 13 , 14 Current guidelines on indications for hysterectomy focus on: symptom relief (i.e., pelvic pain, excessive or unpredictable uterine bleeding, or “uterine bulk,” pressure caused by large leiomyomas); 13 patient age relative to anticipated onset of menopause, when symptoms may subside; 15 , 16 and individual patient preferences, 13 particularly regarding preservation of fertility. While clinical recommendations generally advocate attempting uterine-sparing medications and procedures before hysterectomy, 1 , 15 , 17 this stepped approach may conflict with patient preferences for immediate and permanent symptom resolution. The absence of objective clinical criteria , combined with potential misalignment of recommended treatment sequencing and patient priorities , heightens the scope for institutional and structural factors, rather than purely clinical considerations, to contribute to determining which patients ultimately receive hysterectomy. With hysterectomy care influenced by institutional factors, variations in hysterectomy rates likely stem from two related phenomena: (1) patients with different symptom severities sorting to different practices (case mix), and (2) practices applying different thresholds at which providers recommend hysterectomy. 18 We use ‘severity case mix’ to describe the observable distribution of patient symptom levels at practices, while recognizing this reflects both patient sorting and provider decision thresholds. We speculate that practices may have a lower severity case mix because they emphasize patient autonomy, which in turn maximizes revenue, or have a tendency to refer more complex patients to practices with more resources. We propose that severity case mix could be used as an indicator to assess segregation of gynecologic surgery in the US. Our primary objective is to assess whether practice-level hysterectomy severity case mix is associated with patient race and ethnicity and patient insurance status among premenopausal women with non-cancer-related conditions, as this may represent differential access to surgical settings for hysterectomy . We additionally explore how practice-level case mix severity relates to other indicators relevant to surgical decision-making in this population.