The Health of Transmasculine Patients Compared With Cisgender Patients Treated With Hysterectomy in the U.S. South, 2014-2017.

Transmasculine hysterectomy patients were younger, had higher pelvic pain, and treated in academic medical centers more than cisgender hysterectomy patients. Even if a self-reported gender identity variable is available, computational phenotypes may still be needed because some gender variables may have relatively low sensitivity for identifying transgender patients.

Our study population included 1857 patients ages 19 to 44 years old at surgery who underwent hysterectomies for non-emergency reasons in a large not-for-profit health system between April 14, 2014, and December 31, 2017. Pregnant patients and those with malignancy at the time of surgery were excluded because these surgeries are often performed in emergency or life-threatening situations. The parent study restricted to those under the age of 45 years at surgery to get a presumptively pre-menopausal sample. Additionally, our study age distribution closely reflects the age distribution of transmasculine patients in the United States, as a study that used computational phenotypes to identify transmasculine patients found 88% of transmasculine patients were under the age of 45. 21 The study was approved by the Institutional Review Board in North Carolina. The study utilized data from EHR for patients who had undergone a hysterectomy in one of ten hospitals in a not-for-profit health system in North Carolina. Hospital data was integrated into the health system’s data warehouse and used by data abstractors to search for procedure and diagnostic codes, as well as patient demographics to create the study cohort. Abstractors gathered data recorded in the electronic medical record for up to 1 year before the procedure when the records were available. Data were abstracted in years 2018 and 2019 by four abstractors. The team combed through clinical data including provider pre-operative and operative notes, provider text-string notes, imaging reports, lab values, medication records, previous treatment history, and fertility information. While this health care system’s EHR had a field for sex since the EHR’s inception, self-reported gender identity was not a field in this EHR system until October 2017. Therefore, we identified transmasculine patients through a CP which used diagnostic codes, hormone prescriptions, and an inductive open-ended process with keyword text-strings. Transmasculine patients were first identified by presence of a gender-affirming care (referred to as gender-affirming care as the aforementioned terminology used for the ICD codes during our study time period are now considered outdated) diagnosis in the ICD-9 or ICD-10 diagnostic coding system (see Supplementary Table S1 ). Then, through the abstractors’ manual chart review of medical notes associated with the hysterectomy, we flagged the presence of seven text-strings from past literature: “transgender,” “trans,” “gender identity,” “gender dysphoria,” “FTM,” “female to male,” and “testosterone” (also shown in Figure 1 ). 18 , 21 Notably, the term “FTM” is an outdated term that was used in our search for historical relevance. Testosterone prescriptions were also included in the manual chart review. If any of these indicators were present, the patient was categorized as transmasculine. For the purpose of this analysis, all other patients were coded as cisgender (gender identity congruent with their sex assigned at birth). Gender was operationalized as a dichotomized variable (cisgender = 0; transmasculine = 1). The main reasons for hysterectomy were documented by the provider team in EHR free text as noted above, then captured by manual abstraction. Gender dysphoria, if listed as one of the main reasons for hysterectomy, is defined as distress involving difference between one’s gender identity and one’s sex assigned at birth. Gender dysphoria was separated from other main reasons for surgery to create a mutually exclusive category for surgery. Testosterone use (yes=1, no=0), an important contributing factor to the health of many transmasculine people, was also abstracted from the EHR free text and manual chart review. We used a three-level race/ethnicity variable: Non-Hispanic Black; Non-Hispanic White, and Other. This collapsed race/ethnicity variable was derived from two variables in the EHR structured data: 6-level race (White, Black, Asian, Native, Other, Refused/Unknown) and dichotomous Hispanic ethnicity (yes=1, no=0). We collapsed our race/ethnicity variable into the non-Hispanic White versus non-Hispanic Black and Other for analyses because of previous literature suggesting non-Hispanic Whites may have patterns of care that are different from other racial groups. 22 Patients who had multiple values for the EHR’s race variable (which in our study was consistently White and a non-White value) were classified as the non-White value in our collapsed race/ethnicity variable to center patients of color who are often underrepresented in gynecologic research. Due to low frequency and data user agreement restrictions on reporting results with cell sizes <5, Hispanic, Asian, Multi-racial, and other race/ethnicities were combined as an Other category. Median household income of patients’ residential census tract at time of surgery, insurance type, and hospital type (community, academic, or rural) were collected. Insurance type was documented as private, Medicaid, and other insurance types, which were combined due to low frequency (Medicare, agency/health coverage for people who are incarcerated, tri-care, multi-payor, and self-pay). Reports of past live births were dichotomized into (0=no live births, 1= at least one live birth). Bivariate distributions by gender category were calculated to assess differences in age at surgery, insurance type, hospital type, census tract median household income, gender dysphoria, past live births, and main reason for surgery. Chi-square tests for categorical variables and student’s t-tests for continuous variables were utilized to assess the relationship between gender identities and racial groups among the transmasculine subgroup with p-value <.05 as statistically significant. We assessed differences by race within the transmasculine group as racial disparities have been reported for cisgender women who have had a hysterectomy. 22 We include testosterone use among transmasculine patients. The data analysis for this paper was generated using SAS software (Copyright, SAS Institute Inc., Cary, NC, USA). Beginning on October 18, 2017, the hospital health system began collecting information on gender identity. The self-reported gender identity question included classification fields: Male, Female, Transgender Female/Male-to-Female, Transgender Male/Female-to-Male, Other, and Choose not to disclose. We used this variable to compare our efforts to identify transmasculine individuals at the end of our study period. We assessed concordance and discordance of gender category for patients who continued care in the same health system.

Using key text-strings, diagnosis codes, and prescription information, the CP method identified 35 patients (1.9%) as transmasculine among our cohort of 1,857 patients treated with hysterectomy. Therefore, 1822 patients were assumed to be cisgender women. Key-text strings alone were able to identify all 35 transmasculine patients (100%), and diagnosis codes and prescription information found 34 (97%) of transmasculine patients. Demographics by gender category are in Table 1 . Across both cisgender and transmasculine groups, there is a similar distribution of three-level race/ethnicities (p-value =.20), however, when the racial categories were not collapsed (six-level), the difference between cisgender and transmasculine groups was significant (p-value = <.01). The average age at surgery for hysterectomy for transmasculine patients was 29, 10 years younger than the cisgender average age at surgery of 39 years, and statistically significant (p-value<.001). All transmasculine patients received their hysterectomy at academic medical centers (100%), while cisgender patients were seen at various hospital types, such as community (56.6%), rural (0.3%) and academic medical center (43.1%) hospitals (p-value<.001). The median household income of patients’ residential census tracts was significantly lower for transmasculine patients than for cisgender patients (p-value=0.03). Less than five transmasculine patients reported having a live birth prior to hysterectomy, while 60% of cisgender women reported having at least one live birth. Transmasculine patients had greater missingness (85.0%) in reported live births than cisgender patients (35.7%). Both groups had similar distributions for insurance type, with 72.2% of all patients having private insurance. Of the transmasculine patients, 12 (34.3%) had medical notes documenting gender dysphoria. Transmasculine patients had a significantly higher percentage of chronic pelvic pain (48.6%, p-value<.001), and PCOS (p-value<.001) in comparison with the cisgender group (22.1%). Cisgender women had significantly higher percentages of fibroids (37.9%, p-value<.001) and menorrhagia (35.0%, p-value<.001). Less than half (48.6%) of the transmasculine population had a non-cancerous gynecologic reason for surgery (i.e., fibroids, endometriosis, and pelvic pain). However, almost all (95.4%) cisgender women had at least one of these main reasons for surgery. Six pre-operative main indications for hysterectomy for the transmasculine group contained a sample size too small to report (<5). To further our analyses, we assessed racial/ethnic differences among the transmasculine sub-group in Supplementary Table S2 , which stratified transmasculine patients by race and ethnicity. Among the 35 transmasculine patients, 20 were reported as non-Hispanic white and 11 were patients of color. Four transmasculine patients refused to answer or their race was unknown and were excluded from analyses. There were no statistically significant differences between the two racial groups (not reported due to low sample size). Descriptively, the median census tract household income was slightly lower for patients of color than for non-Hispanic white transmasculine patients. 81.8% of transgender patients of color were prescribed testosterone compared to 80.0% of non-Hispanic white patients (p = 1.00). Less than 45.5% of transgender patients of color were diagnosed with gender dysphoria compared to 35.0% of non-Hispanic white patients (p = 0.566). Non-Hispanic white transmasculine patients (61.1%) had lower reports of a main reason for surgery categorized as “other”, PCOS, or “failed” than transmasculine patients of color (77.8%), though not statistically significant. Age at hysterectomy, proportion of patients with private insurance, hospital type, and most of the main reasons for hysterectomy (e.g., chronic pelvic pain) were similar between both racial/ethnic groups. When we compared our CP method with a gender identity variable that was added to the EHR in 2017, the new EHR gender identity variable only classified 16 (45.7%) of transmasculine patients (out of the 35 patients from our CP method) as transmasculine. Due to missingness in the gender identity variable, the other 19 people had missing values, so the CP could not confirm these patients are transmasculine. It did not identify any patients as transmasculine who were not already categorized as transmasculine by the CP. Therefore, the CP retained 100% specificity compared to the gender identity variable, as it did not find any false negatives (none of the patients identified as transmasculine in CP indicated they were cisgender in the gender identity variable).

Transmasculine patients (n=35) were younger and had higher reports of chronic pelvic pain than cisgender patients (n=1,822). All transmasculine patients were treated in academic medical centers while most cisgender patients were treated at community hospitals. An algorithm of diagnosis codes, keyword text-strings and testosterone prescriptions identified more transmasculine patients treated with hysterectomy than the self-reported gender identity variable. As self-reporting gender identity information becomes more common place, computational phenotypes may still be needed because some gender variables may have relatively low sensitivity for identifying transgender patients.

Using our CP methods, we identified 35 (1.9%) transmasculine individuals from the parent study population. Our findings are consistent with literature that have found significant differences in characteristics and main reasons for hysterectomy between transmasculine and cisgender patients. 3 We found significant differences in characteristics such as age at surgery, hospital type, and median census tract household income. Transmasculine patients were much younger than their cisgender counterparts, aligning with past literature. 3 , 23 This may be in part due to transmasculine patients seeking a hysterectomy for gender-affirming care versus for gynecologic conditions whose severity increases with age. Higher pain and PCOS among transmasculine patients compared to cisgender patients have been reported. 3 , 4 Pain may be reported at a higher level than cisgender patients due to pelvic floor musculoskeletal changes or inflammation potentially due to the onset of menstrual symptoms with taking testosterone and testosterone-related cramping. One cross-sectional study found 20% of transgender men desired a hysterectomy to reduce testosterone-related cramping. 24 These findings have significant implications for clinical care and equity, as there is strong evidence of unequal health burdens of pain for marginalized populations and health disparities in pain care. 25 The majority of transmasculine patients had an indication for gender dysphoria or an indication not specified as a main reason for hysterectomy in our study (e.g. labeled in our study as “other”), which is inconsistent with research in California from 2000–2012 that found pain to be the highest reported main reason for transmasculine patients. 3 This may be due to lack of coverage for gender-affirming surgery before 2014. In 2014, the Affordable Care Act’s main provisions took into effect, and the Medicaid removed its exclusion of transition-related surgery. 26 , 27 Providers may have noted pain as a surgical indication when gender dysphoria was not an option. In our study population from 2014 to 2017, we were able to differentiate main indications for pain and gender dysphoria. 3 Our sample has higher proportions of non-Hispanic Black transmasculine patients (30.3%) than previous literature, which may be because our study population is from the US South, where there are larger numbers of Black transgender people than in other regions. 10 Testosterone use was more common among non-Hispanic white transmasculine patients than transmasculine patients of color, which may represent differences in access to medication and insurance coverage. Gender dysphoria was also more common among non-Hispanic white transmasculine patients. This may represent more access to insurance coverage. The compounding effects of racism and anti-transgender stigma, as well as higher reports of negative health care experiences for transmasculine patients of color could impact trust in providers and lead to less disclosure of transmasculine status. 9 , 11 While we did not have a gold standard, we were able to compare our CP method against a gender identity variable added to the EHR after our study period (starting October 2017). All 16 transmasculine hysterectomy patients who self-reported their transgender gender identity were identified through the CP as transgender, and those who self-reported their cisgender gender identity in the EHR variable aligned with the cisgender hysterectomy patients identified in the CP. Our algorithm identified 54.3% more transmasculine patients than the gender identity variable while also retaining 100% specificity (compared to the gender variable). This aligns with previous research that found their CPs had a specificity of 99% or higher when compared to self-reported gender identity. 28 , 29 All transmasculine patients in our study were confirmed through key text-strings, which is similar to a CP that used key-text string data to validate the use of transgender-related diagnostic codes using Veteran’s Affairs data, which had a false negative rate of 0.05%. 29 Our use of EHR data instead of claims allowed us to look at the distribution of insurance used for this form of gender-affirming care. Insurance access issues are a major issue in gender-affirming care, especially in non-Medicaid expansion states. While examination of privately insured and Medicare-insured populations captures a great deal of the population, it is possible that diagnosis coding and even prescription utilization differs for those on Medicaid or who are uninsured. Therefore, work that includes these populations is important for identifying deviations from coding practices that might be more common for commercially insured patients. While our findings show that diagnosis codes and prescription information alone or key text-strings alone identify nearly all of the transmasculine patients in our study sample who were being treated with hysterectomy, previous literature shows that different aspects of CPs vary in performance by data source. For example, Roblin and colleagues used Kaiser EHR data and found that key text-strings alone, diagnosis codes alone, and both key text-strings and diagnosis codes together led to PPVs of 45%, 56% and 100%, respectively. 16 In Veterans Health Administration data, Wolfe and colleagues found that gender identity related diagnosis codes had the strongest PPV (83%), while Chyten-Brennan and colleagues found only 13.5% of transgender patients were identified with diagnosis codes in their sample of transgender people living with HIV. 20 , 30 Therefore, we believe it is still important to assess utilize different ways of measuring transgender gender identity in EHR data. We were able to obtain a highly specific CP as our parent study was able to collect information through manual chart review abstraction, and we selected participants based on a procedure that can function as a form of gender-affirming care. In addition, all transmasculine patients were seen in academic medical centers, which may provide more gender-affirming services for transmasculine patients than other hospital settings. Our small sample size of transmasculine patients limits our ability to generalize results from our study to the general transgender population. Additionally, differences between gender and racial groups in this sample may be due to confounding by age at surgery, socioeconomic status, or unmeasured variables, as our small sample size does not permit more rigorous statistical analysis. Particularly for our results by racial group, we are likely underpowered due to our low sample size. We hope our results are helpful for future studies that use larger sample sizes, as reporting differences by race among transmasculine patients is an important gap in the literature. We assumed all other patients as cisgender patients, which could impact the sensitivity of our analyses as some transmasculine patients may have been misclassified as cisgender due to lack of information in the EHR. There may have also been patients who were not “out” as transmasculine to their provider, so we likely have obtained an undercount. The lack of documentation of transmasculine status within provider notes may be due to stigma, as some transmasculine people might fear outing themselves to their provider. Lack of insurance coverage for gender-affirming surgeries may also influence documentation of transmasculine status. Therefore, there could potentially be more transmasculine patients within the dataset than we could identify. As the self-reported gender identity variable in the EHR was recently implemented, missingness from this variable may be due to its recent introduction and might not reflect the performance of a gender identify variable established for longer in health system.

Transmasculine people may seek treatment of hysterectomy as part of their gender-affirming care plan. 1 , 2 As hysterectomies are often not considered beyond their role in cisgender women’s health, it is important to identify patients who are transmasculine rather than lumping their health outcomes with cisgender women. Transmasculine people have been reported to have higher levels of pain and polycystic ovary syndrome (PCOS) as main reasons for hysterectomy compared to cisgender women. 3 – 6 With most studies conducted in the Northern United States (US), transgender people in the South are underrepresented. 3 , 7 , 8 According to the 2015 US Transgender Survey, the US South contains the second largest proportion of transgender people in the US after the West. 9 The South contains the highest reports of uninsured transgender people, highest amount of reported job loss due to being transgender, as well as the largest number of Black and Latino transgender populations. 9 – 11 The high reports of structural and interpersonal discrimination could translate to less disclosure of transgender status to providers in the South, making data on transgender populations harder to find. Transmasculine people may also receive hysterectomies for common gynecologic conditions, such as abnormal uterine bleeding, pelvic pain, or cervical pathology. Although these conditions are identifiable by International Classification of Disease (ICD) code, such coding does not provide information on transmasculine status. 3 , 12 Prior efforts to identify transmasculine people who have had a hysterectomy have included ICD-9 and ICD-10 codes, male pronouns, and some form of documentation of male status. 3 A major methodological challenge that remains within transgender health research is identifying transgender individuals within larger electronic health record (EHR) data. EHR data often do not use the recommended two-step method of asking for assigned sex at birth and current gender identity. 13 – 15 There may also be underreporting based on the social stigma associated with identifying as openly transgender, as one survey found 71% of respondents hid their transgender identity to avoid discrimination. 11 Without accurate ascertainment of gender identity, researchers are unable to properly address potential health disparities in comparison to cisgender populations. An important advancement in transgender health is the use of computational phenotypes to identify transgender people within EHR. Computational phenotypes (CPs) are algorithms comprised of diagnostic and procedure codes, medications, and keyword text-strings to locate a sub-population within health care utilization data. 16 – 20 In this study, we focus on identifying transgender people assigned female at birth (referred to as transmasculine hereafter) who are 19–44 years at surgery who have had a hysterectomy. 18 Cohorts of people undergoing hysterectomy may be a valuable source of transmasculine identification and CP research. This study aimed to describe the clinical, reproductive, sociodemographic, and health care system characteristics of a diverse transmasculine patient population compared to cisgender patients among a cohort treated with hysterectomy in a large health care system in the US South from 2014 to 2017. We then compared the CP used to identify transmasculine patients with a self-reported EHR gender identity variable.