{"paper_id":"e33660ae-06ed-4966-ac3f-ab2d9343cc6e","body_text":"Levonorgestrel-releasing intrauterine devices (LNG-IUD) are among recommended\ninitial treatments for abnormal uterine bleeding (AUB; i.e. uterine bleeding with\nirregular frequency, duration or volume) and similar benign conditions such as\nendometriosis, adenomyosis and uterine fibroids [ 1 – 4 ]. These conditions are\ncommon, especially among Black and Hispanic populations [ 5 – 8 ]. For\nexample, prevalence estimates for AUB for reproductive-aged women range between\n10–30% [ 9 – 11 ]. AUB is associated with considerable quality of life\nimpacts and economic costs [ 9 ]. LNG-IUDs can\nbe effective at controlling symptoms [ 1 , 12 – 14 ], but may be under-utilized relative to more invasive treatments such\nas hysterectomy [ 3 , 15 , 16 ], especially\nfor Black and Hispanic patients [ 17 ].\nAfter LNG-IUD placement, re-intervention with hysterectomy is common. For\ninstance, one study found that 46% of patients had undergone a hysterectomy within\n10 years of insertion [ 18 ]. Furthermore,\nLNG-IUDs are more likely to fail and be expelled when fibroids are present [ 1 , 14 ].\nEvidence on risk of hysterectomy following LNG-IUD placement primarily consists of\nsmall studies with non-generalizable patient samples, and few studies follow\npatients for more than 1–2 years [ 12 , 19 , 20 ]. Information for patients or clinicians on the\nclinical and sociodemographic patient characteristics that are associated with\nresponse to LNG-IUD treatment or need for surgical re-intervention, and whether\nthese differ by race and ethnicity, is scarce.\nDifferential progression to more invasive treatment may be expected among\nminoritized patients. For example, social and economic disadvantages more common\namong minoritized groups may incentivize permanent, curative resolution of symptoms\n[ 21 ]. This could motivate patient\npreference for hysterectomy as an initial treatment, but also for more rapid\nprogression to hysterectomy if an LNG-IUD is inserted but not immediately effective\nin managing symptoms. A higher prevalence of fibroids among minoritized patients\n[ 8 ] might also mean LNG-IUDs are less\neffective [ 14 ] and increase risk of\nprogression to hysterectomy. Despite these known patterns, there are sparse data on\nwhich LNG-IUD patients eventually undergo hysterectomy, and whether hysterectomy\nrisk factors differ by race or ethnicity. Therefore, our primary aim is to assess\nwhether risk of hysterectomy following LNG-IUD placement differs by race and\nethnicity. Our secondary aim is assessing whether associations between the clinical\nand sociodemographic characteristics of the patient (age, symptom severity, prior\nIUD use, parity and insurance status) and hysterectomy risk following LNG-IUD\nplacement differ by race and ethnicity.\n\nWe use a retrospective cohort design among patients treated for benign\ngynecological conditions with an LNG-IUD. We identified eligible patients using\nthe North Carolina Translational and Clinical Sciences Institute (NC TraCS)\n“data warehouse” that leverages structured data from electronic\nhealth records (EHR). Data pertained to a large and expanding healthcare system\nin the US South, with an extensive financial assistance program for underinsured\nand uninsured patients. Available data primarily came from hospitals and\noffice-based gynecology providers who were employees of the system’s\nacademic medical center. We abstracted additional unstructured clinical data\nfrom surgical notes and other medical records over four months preceding LNG-IUD\ninsertion [ 22 – 24 ] for matching with structured EHR data. Eligible\npeople were aged 18–44 years, and had an LNG-IUD insertion associated\nwith a diagnostic code for a benign gynecological condition between April\n4 th  2014 and September 18 th  2019 (specifically:\nabnormal vaginal bleeding, fibroids, endometriosis, and gynecological pain; see\n Appendix Table A.1 \nfor medical codes). Patients with LNG-IUDs inserted for other reasons who later\ndeveloped benign gynecological conditions were not considered within scope.\nThe data warehouse extraction identified 1,081 patients across the\nhospital system with an LNG-IUD insertion and an associated diagnostic code for\na benign gynecological condition within the period studied. This was reduced to\nan analytical sample of 783 LNG-IUD patients following exclusions, which were\nlargely related to inaccurate or incomplete data, or LNG-IUD placement for\nprimary reasons other than benign gynecological conditions, such as cancer or\npost-partum contraception. Patients missing data on race and ethnicity were\nexcluded as were patients identifying as Non-Hispanic and either American\nIndian, Alaska Native, Asian, Native Hawaiian, or Pacific Islander, or\nidentifying with multiple non-White races. No single one of these groups\ncontained enough patients to produce stable estimates and avoid risk of\nidentifiability.  Figure 1  details numbers\nof patients excluded for specific reasons. The remaining cohort includes all\npatients who met inclusion criteria.\nStructured sociodemographic data including age, race, Hispanic\nethnicity, marital status and insurance status were extracted from EHR. Age in\nyears was centered on its mean value (33.2 years) for regression models and\ncategorized into ages 35+ and <35 years for estimating censoring-adjusted\nhysterectomy rates. Race and ethnicity, reported by patients at registration,\nwere categorized as: Non-Hispanic Black, Hispanic, Non-Hispanic White. Insurance\nstatus was coded as: Private (including Tricare for members of the US military),\nMedicaid, Medicare (including Medicare Advantage), or uninsured (recorded as\n“self-pay”). Further variables indicated presence of diagnostic\ncodes for fibroids or endometriosis.\nManual abstraction of unstructured EHR data (with a pre-specified\nprotocol) [ 22 – 24 ] provided information on prior IUD treatment,\nparity and gynecological symptom severity. Prior IUD treatment was coded as: no\nprior use, prior use of an IUD (type unknown), and prior use of a hormonal IUD.\nThis represents past (i.e., non-current) IUD use, as current users, who were\nhaving an IUD removed as the new one was being inserted, had already been\nexcluded (see  Figure 1 ). Parity was coded\nas: not recorded, nulliparous, primiparous or multiparous, based on the number\nof prior deliveries. No data were available on LNG-IUD dosage.\nWe coded gynecological symptom severity using adapted versions of\nsymptom severity indices previously validated among hysterectomy patients [ 25 ]. Symptom indices for hysterectomy\npatients drew on unstructured physician notes, diagnostic and procedure billing\ncodes, pre-operative physician notes, laboratory and pathology test results, and\npharmacy prescriptions.  Appendix Table A.2  details how these indices were adapted to LNG-IUD\npatients for whom we had no pre-operative notes, laboratory or pathology test\nresults or pharmacy prescription data. Three distinct scales represented\nseverity of uterine bleeding symptoms (range 0–21; median: 4), pelvic\npain symptoms (range: 0–16; median: 3), and symptoms related to uterine\nbulk (range: 0–7; median: 1). Higher scores indicate more severe\nsymptoms. Binary classifications of high vs low-moderate severity were used in\nestimation of hysterectomy rates (high severity categorized based on\n75 th  percentile as scores of 7+ for bleeding, 5+ for pelvic pain,\nand 4+ for bulk).\nFurther treatments identified (in administrative or abstracted data) as\noccurring between LNG-IUD insertion and study end included further\nLNG-IUD-related encounters and related surgical procedures: myomectomy, uterine\nartery embolization (UAE), endometrial ablation (EA), and hysterectomy (see\n Figure 2  and  Appendix Table A.1  for medical\ncodes).\nWe focus on hysterectomy as very few other surgical procedures were\nobserved during follow up (<10 EAs, myomectomies, or UAEs) and we were\nnot able to capture IUD expulsions and removals comprehensively. Patient follow\nup was until hysterectomy or study end (September 18, 2019), whichever came\nfirst, irrespective of LNG-IUD removal, or other procedures being performed.\nAnalyses therefore assess risk of re-intervention with hysterectomy\nspecifically, and no other potential measures of treatment success or failure.\nWe used a Kaplan-Meier estimator to calculate cumulative incidence of\nhysterectomy at one, two, three and four years following LNG-IUD insertion (with\ndays as the unit of time). We did similar calculations of hysterectomy rates\nstratified by race and ethnicity, symptom severity, having diagnostic codes for\nfibroids or endometriosis, prior IUD use and insurance status, and used a\nlog-rank test to compare rates between groups.\nWe used Cox regression models to further investigate associations\nbetween patient characteristics and hysterectomy rates. We estimated unadjusted\nassociations for race and ethnicity and each covariate, before testing for\ninteractions between race and ethnicity and each other covariate. We considered\nmain effects significant at the p<0.05 level but used a more liberal\np<0.10 threshold for interactions as statistical power may have been low.\nA final model included all covariate main effects, regardless of significance\n(i.e. race and ethnicity, age at treatment, bleeding, pain and bulk severity\nscores, diagnostic codes fibroids and endometriosis, prior IUD use, parity and\ninsurance status) and any significant interactions.\n\nOur cohort included patients who were White (58.1%), Black (26.6%) and\nHispanic (15.3%) and approximately half (48.2%) were aged 35 years or over ( Table 1 ; See  Appendix Table A.3  for patient\ncharacteristics by race and ethnicity). There were 76 hysterectomies observed during\nfollow-up (9.7% of patients). Median follow-up time (to either hysterectomy or\ncensoring) was 1.69 years (interquartile range 0.8–2.9 years, 90th\npercentile: 4.0 years). Minimum follow up time was 0 days (<10 patients had\nan insertion on the last day of the study period). In no instances did follow-up\ntime differ significantly where hysterectomy rates did not, suggesting that\nfollow-up rates were similar regardless of patient characteristics. The cumulative\nrisk of hysterectomy within one year following LNG-IUD insertion was 7.0% and this\nincreased to 14.9% by four years post-insertion.\nThere was little evidence of rates differing by race and ethnicity ( Table 1 ), but several other factors were\nassociated with significantly higher hysterectomy rates, including older age, higher\nseverity uterine bleeding and bulk symptoms, being multiparous (compared to\nnulliparous), having fibroids, and being on Medicaid or being uninsured (compared to\nbeing privately insured). Hysterectomy rates were also significantly higher among\npatients who had used hormonal IUDs before than among patients with no recorded\nprior use and the largest magnitude differences were at three (16.0% vs 12.9%) and\nfour (21.6% vs 15.0%) years post-insertion.\nTable 2  presents the unadjusted and\nmultivariable-adjusted hazard ratio estimates from Cox regression models. In\nunadjusted models, higher age at treatment, more severe uterine bleeding and bulk\nsymptoms, diagnostic codes for fibroids, and being multiparous or on Medicaid were\nassociated with increased hysterectomy risk, similar to findings from  Table 1 . In adjusted models, only bleeding severity\nremained significantly associated with hysterectomy rates (HR:\n1.05[1.00–1.10]), though age at treatment and bulk severity were borderline\nsignificant with estimates of meaningful magnitude (age HR: 1.04[1.00–1.09];\nbulk HR: 1.15[1.00–1.32]). Notably also, while confidence intervals for\nassociations with pelvic pain symptoms over-lapped the null, the associations were\nof similar magnitude and direction as those for symptoms of uterine bleeding. We\ntested for interactions between race and ethnicity and each other covariate but none\nwere significant at the p<0.10 level and in our final model with all\ncovariates, including interactions with race and ethnicity for all variables did not\nimprove model fit (p=0.287).\n\nAmong patients aged 18–44 treated with LNG-IUDs for non-cancerous\nconditions we did not find statistically significant differences in hysterectomy\nrisk by race and ethnicity over a median follow-up time of 1.7 years. Cumulative\nincidence rates for hysterectomy were 7.0% within the first year following LNG-IUD\nplacement, and 14.9% at four years post-insertion. Hysterectomy rates were higher\namong patients who were older, had more severe pre-insertion bleeding and bulk\nsymptoms, had diagnostic codes for fibroids, were multiparous, and were Medicaid\nbeneficiaries. Associations with these patient characteristics did not appear to be\nmoderated by race and ethnicity. Bleeding severity remained significantly associated\nwith hysterectomy after adjustment for other patient characteristics in Cox\nregression models, and there were also borderline associations with hysterectomy\nrisk for age and bulk severity.\nComparisons of LNG-IUDs and hysterectomy as treatments for benign\ngynecological conditions, found LNG-IUDs to be relatively cost effective, even\nthough patients randomized to LNG-IUD treatment often eventually undergo\nhysterectomy (e.g. in one large trial starting in 1994 in Finland, 20% underwent\nhysterectomy after one year, and 42% after five years) [ 12 , 19 , 26 ]. We observed lower rates which could be due\nto a number of factors including our observational design, larger sample size,\ndiffering medical system and differing geographic, social and historical context.\nEvidence is sparse on rates and patient characteristics associated with hysterectomy\namong large observational samples of patients receiving non-randomized LNG-IUD\ntreatment. While observational data may lack measurement of factors that are\nimportant for understanding causal pathways of non-randomized exposures, detailed\ndescription of naturalistic, “real-world” treatment may still help\nidentify potential structural or systemic barriers contributing to healthcare\ndisparities. Our study contributes to such detailed description, with rich clinical\nmeasures of bleeding, pain and bulk symptoms, combined with sociodemographic\ninformation such as age, parity and insurance status.\nAlthough we did not find statistically significant evidence of disparities\nby race and ethnicity in risk for hysterectomy following LNG-IUD insertion, we\ncannot rule out an association. Following adjustment for symptom severity and other\ncovariates, point estimates indicated considerably lower risk of hysterectomy for\npatients who were Black or Hispanic than for those who were White (HRs were\n0.58–0.71), but confidence intervals were wide and we may have been\nunder-powered to detect this association. A study of re-intervention rates following\nuterine-sparing surgeries (rather than LNG-IUD placement) found little evidence for\nrates differing by race and ethnicity [ 27 ].\nTheir study focused on fibroid patients in an integrated healthcare delivery system\nbut findings were consistent in not showing increased risk of hysterectomy for\npatients who were Black or Hispanic. This contrasts with overall hysterectomy rates\nwhich are higher among Black and Hispanic than White populations in the United\nStates [ 5 – 7 ]. Disparities may arise earlier in the care pathway,\nbut there is little evidence they arise via less favorable responses to treatment\nfor patients who are Black or Hispanic compared to those who are White.\nTreatment recommendations emphasize consideration of the relative cost and\nburden of treatment options, symptom severity and response to treatments, patient\npreferences, e.g. with regards to preservation of fertility, and patient age in\nrelation to the expected onset of menopause [ 21 , 28 , 29 ]. Patients from different racial and ethnic groups may\nvalue and prioritize these factors differently, for example, medical costs and\ninterference with work capacity may be more salient concerns among minoritized\nracial and ethnic groups, who also experience socioeconomic disadvantages [ 30 , 31 ].\nWe did not find evidence that age, symptom severity, diagnoses of fibroids or\nendometriosis, prior IUD use, parity or insurance status were differentially\nassociated with hysterectomy following LNG-IUD placement for patients who were\nWhite, Black and Hispanic. However, our study may have been under-powered to detect\nsuch interactions and they should not be ruled out for future investigations in\nlarger samples.\nOur study has some limitations: it is observational and potential\nrelationships between patient characteristics and LNG-IUD treatment are unknown (we\nonly observed the treated). The lack of clear evidence for racial or ethnic\ndisparities in hysterectomy risk following LNG-IUD placement does not preclude\npotential disparities occurring earlier in the care pathway. For example, there is\nevidence that Black and Hispanic patients have more severe symptoms and are more\nlikely to be treated directly with hysterectomy, without first attempting\nalternative treatments such as the LNG-IUD [ 17 ]. With no reliable measure for all patients of how effectively\nLNG-IUDs controlled symptoms, we could not determine whether Black and Hispanic\npatients simultaneously experience worse symptom control with LNG-IUDs but greater\nbarriers to accessing hysterectomy treatment (balancing out to no clear evidence of\ndifferences in hysterectomy risk). Moreover, we primarily captured data from\nhospitals rather than primary care, so observed LNG-IUD patients may have been those\nreceiving advanced or specialty care, potentially using the LNG-IUD as a stop-gap\nwhile awaiting surgery. Analyses assumed patients were only treated within the same\nhospital system. Observation of hysterectomy could have been less likely for\npatients receiving care from multiple healthcare systems, which potentially\nunderestimates hysterectomy rates. We may have differentially underestimated rates\nif patients with particular characteristics were more likely to receive additional\ncare outside the study healthcare system.\nHysterectomy was approximately twice as likely for patients aged 35 or more,\nwith fibroids, or more severe bleeding and bulk symptoms. Clinicians may want to\nconsider that LNG-IUDs may be less successful at delaying hysterectomy for older\npatients with severe symptoms. Lesser effectiveness with bulk symptoms is consistent\nwith previous findings that LNG-IUDs are more likely expelled for patients with\nlarge fibroids [ 1 , 14 ]. Multiparous patients also had higher hysterectomy\nrates, but this was less evident when controlling for covariates, potentially\nsuggesting that parity relates to hysterectomy via associations with age and symptom\nseverity. Indeed, age and parity might both influence patient desires for future\nfertility.\n\nThe data analyzed in this study are owned and administered by the North\nCarolina Translational and Clinical Sciences Institute (NC TRACS). Access to the\ndata for research purposes can be requested at:  https://tracs.unc.edu/index.php . Questions about data access may be\ndirected to  nctracs@unc.edu .","source_license":"public-domain-us","license_restricted":false}