Emulating a target trial of surgical removal of uterine fibroids on atherosclerotic cardiovascular disease.

Among 1,076,918 individuals with uterine fibroids, 337,090 individuals were eligible across 589,848 person-months ( Supplemental Fig. 2 ). The study included 11,106 myomectomy, 23,089 hysterectomy, and 27,657 no surgery individuals (5% random sample). Baseline characteristics are shown in Table 1 ; after weighting, all SMD were <0.1 ( Supplemental Fig. 3 ). Over an average follow-up of 2.41 years (myomectomy; SD = 1.73), 2.38 years (hysterectomy; SD1.73), and 2.44 years (no surgery; SD = 1.75), 149 atherosclerotic CVD diagnoses and events following myomectomy, 506 following hysterectomy, and 606 in the no surgery group occurred. Figure 1A shows the cumulative risk curves for risk of atherosclerotic CVD by treatment group. At 5 years, individuals who underwent a myomectomy had a 0.93 percentage-point lower 5-year risk of atherosclerotic CVD (95% CI: −2.04, 1.30) compared with no surgery, although this was imprecise. Hysterectomy was associated with a 1.24 percentage-point increased 5-year risk of atherosclerotic CVD compared with no surgery (95% CI: −0.01, 2.03) ( Table 2 ). After 5-years of follow-up, 2,030 (7.34%) in the no surgery group underwent a myomectomy and 4,152 (15.01%) underwent a hysterectomy, whereas 563 (5.07%) in the myomectomy group underwent a hysterectomy. Results were similar for per-protocol analyses accounting for subsequent surgeries ( Fig. 1B ). Compared with no surgery, undergoing a myomectomy was associated with a 0.98 percentage-point reduction in atherosclerotic CVD (95% −2.10, 1.27) and hysterectomy a 1.39 percentage-point increase in atherosclerotic CVD (95% CI 0.02, 2.16) ( Table 2 ). Estimates for secondary outcomes of individual atherosclerotic CVD diagnoses and events were imprecise because of few events. There were no differences in the 5-year risk of coronary artery disease, myocardial infarction, transient ischemic attack, and peripheral artery disease after myomectomy compared with no surgery. Undergoing a myomectomy was associated with modestly decreased, although imprecise, risk of cerebrovascular disease (adjusted risk difference [aRD] −0.86%, 95% CI: −1.61, 0.66) and ischemic stroke (aRD −0.42%, 95% CI: −1.00, 0.55). Undergoing a hysterectomy was associated with modestly increased risk of coronary artery disease (aRD 0.88%, 95% CI: 0.03, 1.66), myocardial infarction (aRD 0.52%, 95% CI: 0.02, 1.05), and ischemic stroke (aRD 0.62%, 95% CI: −0.06, 1.23) ( Fig. 2 ; Supplemental Table 3 ). There were no differences in the 5-year risk of cardiovascular risk factors among those who underwent myomectomy compared with those who had no surgery. At 5 years, individuals who underwent hysterectomy had a higher risk of obesity (aRD 9.36%, 95% CI: 8.40, 11.73), hypertension (aRD 3.85%, 95% CI: 1.29, 5.62), and hyperlipidemia (aRD 4.45%, 95% CI: 1.49, 5.80) ( Supplemental Table 3 ). Among those with race and ethnicity data, 22,437 (65.0%) were White, 10,131 (29.3%) Black, 1,966 (5.7%) Asian, and 5,163 (14.9%) Hispanic. Estimates were imprecise across groups because of few events ( Supplemental Table 4 ). Among those aged <40, hysterectomy was associated with an increased 5-year atherosclerotic CVD risk (aRD 1.86%, 95% CI: −0.35, 3.08), whereas myomectomy was associated with a decreased risk in the age ≥40 group (aRD −1.50%, 95% CI: −2.94, 0.74) ( Supplemental Table 5 ). Among myomectomies, 34.9% (N = 6,513) were open/abdominal and 55.9% (N = 10,451) were minimally invasive; for hysterectomy, 32.4% were open/abdominal (N = 13,740) and 67.5% (28,583) were minimally invasive. At 5 years, compared with no surgery, minimally invasive myomectomy was associated with a 1.69 percentage-point (95% CI: −2.91, 0.28) lower risk of atherosclerotic CVD. Open hysterectomy was associated with a 2.75 percentage-point higher risk (95% CI: 0.94, 4.06) ( Supplemental Table 6 ). Among 2,441 (9.6%) individuals who underwent hysterectomy with unilateral oophorectomy, 5-year atherosclerotic CVD risk was 2.06 percentage points higher (95% CI: −0.87, 4.60) than no surgery in the ITT analysis ( Supplemental Table 6 ). Results did not change appreciably after excluding 6,915 (11.2%) individuals because of infertility, 1,535 (2.5%) because of <12 months of baseline data, or using IPCW ( Supplemental Table 7 ). When hysteroscopic myomectomy (CPT 58561) was separated from other minimally invasive myomectomy procedures, there was a reduction in atherosclerotic CVD risk compared with no surgery (ITT aRD −1.99%, 95% CI: −3.50, −0.28; Supplemental Table 8 ), whereas estimates for laparoscopic and robotic myomectomy remained directionally similar but less precise.

We used target trial emulation to estimate the effect of surgical removal of fibroids on atherosclerotic CVD risk using observational data ( 36 – 39 ). When few individuals initiate the treatment at any one point in time, this approach can be extended to emulate a sequence of target trials to reduce selection bias and immortal time bias ( 39 ), producing estimates comparable with trials ( 33 , 40 ). The analysis utilized Optum’s deidentified Clinformatics Data Mart Database (Optum CDM) (2000–2024) ( Supplemental Appendix 1 , available online). Variables were identified using International Classification of Diseases (ICD)-9/10, Current Procedural Terminology (CPT), Healthcare Common Procedure Coding System (HCPCS), and National Drug Codes (NDC) ( Supplemental Table 1 , available online). Informed consent was waived, given the use of fully deidentified data and minimal risk of harm. The study was reviewed and approved by the University of Pennsylvania Institutional Review Board. The trial emulation protocol is described in Supplemental Table 2 ( 41 ). Eligible individuals were females in Optum CDM, aged 18–50 years, with a diagnosis of uterine fibroids between December 1, 2000 and October 1, 2019 (allowing at least 5 years of follow-up), defined as ≥1 fibroid claim preceded by a transvaginal ultrasonography or pelvic magnetic resonance imaging claim within 30 days ( Supplemental Table 1 ) ( 42 , 43 ). Individuals with a history of atherosclerotic CVD, hysterectomy, myomectomy, oophorectomy (unilateral or bilateral), menopause, or cancer were excluded. Those with hysterectomy and concurrent oophorectomy were excluded from the primary analysis, as the indications for oophorectomy are typically not related to uterine fibroids ( 44 ), but were considered in exploratory analyses. Pregnant or postpartum individuals were excluded, defined as 10 months before and 3 months after a delivery code. To capture comorbidities before “trial” start, we required continuous health insurance coverage for ≥6 months before (30-day gap allowed) entry into the study cohort. We also required individuals to have a fibroid claim within 1 month of each trial start. To emulate a trial examining the effects of hysterectomy, myomectomy, or no surgery on incident atherosclerotic CVD, eligible individuals were followed until atherosclerotic CVD, disenrollment from Optum CDM, or November 2024. In the trial, death would be a censoring event, yet mortality data were unavailable. Trials were defined on the calendar month (e.g., from December 1, 2000 to December 31, 2000). At baseline of the first trial, defined as December 2000, individuals were classified into the myomectomy arm if they had procedure codes for a myomectomy that month, the hysterectomy arm if they had procedure codes for a hysterectomy that month, and the no surgery arm otherwise. If both procedures were coded in the same month, they were coded in the hysterectomy group (2%) ( 30 ). To reduce computational time, we used a 5% random sample of the no surgery arm, resulting in a sample size comparable with other treatment groups. We implemented a sequential target trial ( 36 , 37 , 39 , 45 – 48 ), emulating 271 trials monthly from December 2000 through October 2019 ( Supplemental Fig. 1 , available online). Eligibility criteria were reassessed at each sequential trial. Individuals who were no longer eligible (i.e., because of age, comorbidity criteria or prior surgical removal) were excluded from that trial; all others were reclassified according to whether they underwent surgery during that month. Individuals could contribute to multiple trials to maximize statistical power ( 36 , 37 , 46 ). Trial emulation can yield observational analogues of the intention-to-treat (ITT) effect and the per-protocol effect. The observational analogue of the ITT effect was defined as the effect of undergoing (rather than being randomized) a treatment strategy at baseline. The observational analogue of the per-protocol effect was defined as the effect of undergoing a given treatment strategy and receiving no further intervention. The per-protocol effect is relevant because surgical reintervention after myomectomy is common ( 19 , 30 , 49 ). The primary outcome was 5-year risk of atherosclerotic CVD diagnosis or event, defined by the American Heart Association ( 3 ), including coronary artery disease (encompassing stable ischemic heart disease and acute myocardial infarction), cerebrovascular disease (including transient ischemic attack and ischemic stroke), and peripheral artery disease. Secondary outcomes examined individual atherosclerotic CVD components and acute events. Outcomes were identified as the first ICD-9/10, CPT, or HCPCS claims. One inpatient or more than one outpatient claim was required for classification as an outcome ( 50 , 51 ). Cardiovascular risk factors, including incident type 2 diabetes, hyperlipidemia, hypertension, and obesity, were also examined, identified as the first ICD-9/10 or NDC code, if applicable. For these analyses, individuals with these conditions at baseline were excluded. Baseline variables were updated at the start of each trial (e.g., baseline variables for the December 2000 trial reflected the most recent measurements of covariates at that time). To emulate randomization at baseline, inverse probability of treatment weights (IPTW) were estimated. Weights were fit across all trials with calendar time and trial number fit using natural splines. Variables used to generate weights included age at fibroid diagnosis and age at trial start (modeled using natural splines), and binary indicators of cardiovascular risk factors (obesity, hyperlipidemia, type 2 diabetes, hypertension, smoking), mental health (anxiety, depression), reproductive health (parity, infertility, endometriosis), fibroid symptoms (pelvic pain or heavy menstrual bleeding), and hormonal contraceptive use (oral and nonoral). Parity (parous vs. nulliparous) was defined as prior delivery in the dataset. Use of tranexamic acid, gonadotropin-releasing hormone (GnRH) agonists, or prior procedural interventions was rare (<0.5%) but summarized descriptively. Balance after IPTW was assessed using standardized mean differences (SMD), with ≤0.1 considered acceptable ( 52 ). For the ITT analysis, we fit pooled logistic regression models for each outcome and used predicted values to estimate 5-year risks, weighted by IPTW. Because the monthly event probability was small, the pooled logistic model parameters closely approximated a Cox proportional hazards model ( 53 ). Data from all trials were pooled into one model, including baseline month, follow-up month, and treatment-time interaction, modeled using natural splines. The per-protocol analysis was similar, except individuals were censored at deviation from the assigned treatment strategy. Specifically, those in the myomectomy group were censored if and when they had a hysterectomy, and the no surgery group was censored if and when they had a hysterectomy or myomectomy. Because a hysterectomy removes the uterus, these individuals were never censored, even if coded with a myomectomy (likely coding error). To adjust for time-varying selection bias induced by artificial censoring, individuals received time-varying weights of the inverse probability of adhering to treatment. At each time point, individuals were weighted by the cumulative product of their IPTW and inverse probability of adhering to treatment. Additional details are provided in the Supplemental Appendix 2 . Weights were truncated at the 99th percentile. We calculated 95% confidence intervals (CIs) using bootstrapping (500 repetitions). Missing data for the exposure, outcome, and confounders were not present, as binary indicators were defined as the presence or absence of a code. Analyses were performed in RStudio (Version 4.0.2, RStudio Team). Race and ethnicity could not be included in the primary adjustment because data were unavailable for the full cohort, although they may be important covariates as indicators of structural factors ( 2 , 54 , 55 ). Among a subcohort (55.8%) with self-reported race and ethnicity data, stratified analyses (White, Black, Asian, Hispanic) were conducted. Given age differences in hormone production ( 56 , 57 ), analyses were stratified by age at trial start (<40 years and ≥40 years). Because of small event counts, only ITT models were fit for stratified analyses. For each of these analyses, IPTW were refit in the stratified cohorts. Two exploratory analyses were conducted. We examined the surgical route, given that minimally invasive approaches may reduce surgical stress and inflammation compared with open procedures ( 58 – 60 ). Route was classified as open/abdominal or minimally invasive (endoscopic, laparoscopic, vaginal) on the basis of procedure codes. Codes that did not specify route were excluded (myomectomy: 9.2%, N = 1,716; hysterectomy: 0.05%, N = 23). Given the known association between oophorectomy and atherosclerotic CVD ( 20 , 61 – 64 ), we included individuals who underwent a hysterectomy with unilateral oophorectomy in exploratory analyses. Although excluded from the primary analysis because clinical indications may differ from the indications for hysterectomy alone, some individuals may elect concurrent unilateral oophorectomy for other indications (e.g., risk of ovarian cancer) ( 44 ). Individuals with bilateral oophorectomy or codes for removing the “remaining” uterus remained excluded. We conducted several sensitivity analyses. First, we excluded individuals who used infertility medications at any time during follow-up as a proxy for baseline pregnancy intentions; such individuals would not enroll in a trial where hysterectomy was an option ( 58 ), although including them in the primary analysis would induce selection bias. Second, we computed inverse probability of censoring weights (IPCW) to adjust for selection bias because of loss to follow-up ( Supplemental Appendix 2 ). For each month, the cumulative product of the IPTW and IPCW was used. Third, we extended the required baseline insurance coverage from ≥6 months to ≥12 months to see if capturing longer baseline data impacted results. Fourth, we note that hysteroscopic myomectomy (CPT 58561) was included within the minimally invasive category in the primary route analysis. Hysteroscopic myomectomy is clinically distinct from laparoscopic or robotic approaches and is typically performed for submucosal fibroids, which differ in number, size, and burden from fibroids treated via other minimally invasive routes. We conducted a sensitivity analysis separating hysteroscopic myomectomy from other minimally invasive approaches to examine whether this grouping influenced results.

Fibroid treatment decisions should be grounded in shared decision-making ( 58 , 68 – 70 ). This study suggests that among individuals with uterine fibroids, myomectomy was not strongly associated with atherosclerotic CVD risk, whereas hysterectomy was associated with modest increases. These associations may inform patient counseling alongside established decision-making factors, yet further research is needed.

Among individuals with uterine fibroids, there were imprecise associations between myomectomy and atherosclerotic CVD risk, while undergoing a hysterectomy (without oophorectomy) was associated with modestly increased atherosclerotic CVD risk. Although confidence intervals around estimates for cardiovascular associations with myomectomy indicate uncertainty, the consistent direction of estimates may suggest potential cardiovascular benefits, highlighting the need for further research. This may be related to preservation of uterine and ovarian functions, with maintained cyclic hormonal patterns and menstrual function after surgery ( 58 ). Additionally, undergoing a myomectomy may provide benefits in preserving reproductive physiology while relieving fibroid-related symptoms, such as heavy bleeding, pelvic pressure, and anemia, that have been associated with increased cardiovascular strain ( 17 ). Interestingly, this does not seem to be driven by reductions in atherosclerotic CVD risk factors, in particular, incident type 2 diabetes, hypertension, hyperlipidemia, or obesity. Future research with more extended follow-up should aim to examine long-term cardiovascular associations after myomectomy. The modestly increased risk of atherosclerotic CVD with hysterectomy is consistent with general reproductive cohorts, potentially explained by reduced ovarian blood flow when uterine arteries are divided ( 65 ). Alterations in cyclic hormonal patterns may have adverse effects on lipid metabolism, endothelial function, and vascular inflammation. ( 66 , 67 ) Consistent with this, undergoing a hysterectomy was also associated with a higher 5-year risk of obesity, hypertension, and hyperlipidemia in this cohort, even after excluding individuals with these conditions at baseline. However, as these outcomes were ascertained via diagnosis codes, it is unclear whether they reflect truly incident conditions or residual differences between groups, including conditions that were present but previously undiagnosed. Our findings extend prior hysterectomy and CVD studies ( 21 , 51 , 62 , 63 ), which included postmenopausal cohorts, did not isolate uterine fibroid cases ( 24 ), and often did not distinguish procedures without oophorectomy, which have a less clear biological rationale for negative cardiovascular impact. A meta-analysis comparing hysterectomy with no surgery reported modest CVD and stroke risk increases (9%–10%) ( 23 ), whereas an analysis of the Nurses’ Health Study reported a 21% increased CVD risk, although in a cohort with an average age of 48 years ( 24 ). A cohort in Australia similarly found an increased risk associated with hysterectomy with ovarian conservation before 35 years ( 22 ). Only one study conducted a subgroup analysis exclusively among individuals with uterine fibroids (N=761), finding no difference in coronary artery disease or stroke risk after hysterectomy with ovarian conservation, despite harmful associations in their general cohort ( 20 ). Our analyses stratified by route of surgery provide additional evidence. Open/abdominal hysterectomy was associated with a substantially higher 5-year atherosclerotic CVD risk compared with no surgery, whereas minimally invasive hysterectomy showed a modest and imprecise increase. This difference may reflect greater surgical trauma and tissue disruption with open procedures, or the selection of patients with more severe disease for open approaches. Hysterectomy with concurrent unilateral oophorectomy was associated with an even greater atherosclerotic CVD risk increase of 2.06%, consistent with the hypothesis that surgical disruption of ovarian function may be associated with cardiovascular risk through premature induction of a hypoestrogenic state. ( 21 – 23 ) Several important limitations must be considered. Confounding by fibroid severity is an important concern and warrants careful consideration for each comparison. For myomectomy compared to no surgery, the theoretical direction of confounding by indication is reassuring, as individuals with more severe fibroids would be expected to have worse cardiovascular risk profiles, which would be expected to bias the myomectomy group toward a higher atherosclerotic CVD risk. Because myomectomy showed a modest protective association, this direction is inconsistent with what confounding by indication alone would produce and might suggest an underestimate. For hysterectomy, this argument does not apply in the same way, and residual confounding by disease severity cannot be excluded. Notably, open hysterectomy was associated with substantially greater atherosclerotic CVD risk than minimally invasive hysterectomy, a pattern that may reflect greater surgical trauma and vascular disruption, although differences in patient selection across surgical routes could also contribute. From a methodological standpoint, we attempted to reduce the potential for confounding by indication using IPTW with balanced observable proxies of disease severity, including fibroid-related symptoms, and we required imaging confirmation of diagnosis and an active fibroid claim within 1 month of each trial start. Unmeasured aspects of fibroid severity, including size, number, and location, remain limitations of claims-based data, and residual confounding cannot be fully excluded. Additional limitations include reliance on billing codes for variable ascertainment and lack of data on surgical indications beyond fibroid diagnosis. Although we required a fibroid claim within 1 month of follow-up start to ensure active disease, hysterectomies may have been performed for cooccurring conditions such as endometriosis or uterine prolapse, which would need an independent causal relationship with atherosclerotic CVD to result in confounding. Additionally, the Optum CDM database population may limit generalizability to broader patient populations with uterine fibroids. Finally, the average length of follow-up in this cohort was shorter than expected, which limited power to detect differences in outcomes and could have led to an underestimation of risk differences, as atherosclerotic CVD events may develop over longer periods. Although our study was designed to emulate a target trial and estimate causal effects, the small number of events and resulting wide confidence intervals mean these findings should be interpreted as exploratory rather than definitive; caution is warranted in drawing strong causal conclusions from these estimates.

Supplemental data for this article can be found online at https://doi.org/10.1016/j.fertnstert.2026.05.151 .