Prevalence, characteristics, and risk factors of occult uterine cancer in presumed benign hysterectomy

A total of 229,536 adult women met sample eligibility criteria. Their median age was 47 years (interquartile range: 42-52) ( Table 1 ). Uterine fibroid (38.9%), menstrual disorders (20.3%), and genital prolapse (13.8%) were the most common indications. Forty-two percent of the women underwent a total abdominal or total laparoscopic hysterectomy, while 28.8% had a vaginal or laparoscopic-assisted vaginal hysterectomy. In this sample, 2,207 patients had occult uterine cancer, resulting in an overall prevalence of 0.96% (95% CI: 0.92-1.00%; Table 2 ). When distinguished by subtype, 1,716 had endometrial carcinoma (0.75%, 95% CI: 0.71-0.78%), and 495 had uterine sarcoma (0.22%, 95% CI: 0.20-0.23%) including 346 who had leiomyosarcoma (0.15%, 95% CI: 0.13-0.17%). (Four patients had both endometrial carcinoma and uterine sarcoma.) Sensitivity analyses excluding radical hysterectomies, excluding patients with any diagnosis of metastatic cancer in the past 9 months, or expanding the time window for measuring occult uterine cancer to 6 months and 24 months after index hysterectomy produced similar estimates ( Appendix D ). The risk of occult uterine cancer varied by surgical indication, age, and surgical route ( Table 2 ). Endometrial carcinoma was found in 0.13% (95% CI: 0.05-0.21%) of women undergoing hysterectomy for endometriosis, 0.33% (95% CI: 0.29-0.37%) for uterine fibroids, and 7.92% (95% CI: 7.08-8.77%) for postmenopausal bleeding. When stratified by age, the risk for occult endometrial carcinoma ranged from 0.10% (95% CI: 0.02-0.29%) in women age 18-29 years to 3.93% (95% CI: 3.47-4.38%) in women age ≥75 years. Likewise, the risk for occult endometrial carcinoma was 0.18% (95% CI: 0.12-0.24%) in women undergoing laparoscopic supracervical hysterectomy versus 1.35% (95% CI: 1.25-1.45%) in women who underwent a total abdominal hysterectomy. Risk for occult uterine sarcoma had a similar pattern. Among women with occult endometrial carcinoma, most had localized disease (71.3%) and grade 1-2 tumor (63.8%) ( Table 3 ). Thirty-five percent had tumors ≤5 centimeters, while 48.2% had unknown tumor size. Adenocarcinoma was the most common subtype (89.2%). In contrast, women with unexpected uterine sarcoma had more severe disease ( Table 3 ). Twenty-one percent had distant stage, 54.1% had grade 3-4 tumor, and 61.8% had tumors larger than 5 centimeters. Leiomyosarcoma accounted for 69.9% of the occult uterine sarcomas. The multivariable prediction models confirmed surgical indication and age as significant risk factors ( Figures 1A and 1B ). For instance, compared to women undergoing hysterectomy for fibroids, those who had hysterectomy for genital prolapse were 0.29 (95% CI: 0.22-0.39) times as likely to have occult endometrial carcinoma and 0.01 (95% CI: 0.001-0.05) times as likely to have occult uterine sarcoma. Moreover, the risk of occult uterine cancer was significantly higher in older women (e.g., adjusted RR for age 60-64 years versus 45-49 years=7.08, 95% CI: 5.34-9.39, for endometrial carcinoma; and 7.06, 95% CI: 4.44-11.21, for uterine sarcoma). Non-Hispanic black women were less likely than non-Hispanic whites to have occult endometrial carcinoma possibly due to more frequent evaluation of uterine fibroids among black women, whereas diabetes, hypertension, obesity, and renal failure were associated with a higher risk for occult endometrial carcinoma ( Figure 1A ). For occult uterine sarcoma, the risk was higher in patients with hypertension, blood loss or deficiency anemia, weight loss, and personal history of malignancy ( Figure 1B ). When evaluated in the validation sample, the model for endometrial carcinoma had a sensitivity of 73.7%, specificity of 86.0%, and area under ROC curve of 0.85; and the model for uterine sarcoma attained a sensitivity of 70.4%, specificity of 75.2%, and area under ROC curve of 0.80 ( Table 4 ). Both models had high NPV (99.8% for endometrial carcinoma and 99.9% for uterine sarcoma) and good clinical utility index score for ruling out the disease, supporting their potential use in identifying women at low risk for occult uterine cancer. However, since occult uterine cancer was rare, these models had low PPV.

In a large population-based sample of women undergoing hysterectomy for presumed benign indications, we found that 0.75% of them had occult endometrial carcinoma and 0.22% had occult uterine sarcoma (including 0.15% with occult leiomyosarcoma). The risk for occult uterine cancer differed by patient age, surgical indication, and surgical route. Several other patient characteristics were also identified that may inform preoperative risk assessment. Estimated risk of occult uterine cancer varied substantially in previous research. For instance, among 23 studies published in April 1, 2014-April 15, 2017, the estimated risk ranged from 0% to 0.51% for unexpected leiomyosarcoma and from 0% to 1.48% for uterine sarcoma. 3 The U.S. Food and Drug Administration (FDA) synthesized evidence from the literature and reported a risk of 1 in 498 women (0.20%) for occult leiomyosarcoma and 1 in 352 (0.28%) for occult uterine sarcoma in its initial statement in 2014, 22 and a risk of 0.09-0.20% for leiomyosarcoma and 0.17-0.45% for uterine sarcoma in an updated assessment in 2017. 3 Meta-analysis performed by the Agency for Healthcare Research and Quality (AHRQ) estimated that 0 021% (95% credible interval: 0-0.094%) of women in prospective studies and 0.085% (95% credible interval: 0.047-0.127%) in retrospective studies had unexpected leiomyosarcoma. 4 These estimates, however, focused on women with fibroids. In our study, the risk of occult leiomyosarcoma and uterine sarcoma was 0.21% and 0.30% respectively in the subset of women undergoing hysterectomy for fibroids, which are on par with the FDA estimates but higher than the AHRQ estimates. One possible explanation is that in addition to hysterectomies, the FDA and AHRQ assessments included myomectomies where the risk for occult uterine cancer is lower because patients tend to be younger and healthier. 23 , 24 In particular, 56.7% of the prospective studies and 31.9% of the retrospective studies included in the AHRQ analysis focused on myomectomies. 4 Our study also extends this literature by examining the risk of occult endometrial carcinoma in hysterectomies. Although most attention has centered on sarcomas due to their poor prognosis, endometrial carcinoma is more common and warrants careful consideration as well. In our study, three-quarters of the occult uterine cancers were endometrial carcinoma. Meanwhile, techniques for identifying endometrial carcinoma (e.g., transvaginal ultrasonography, endometrial biopsy) are more effective than those available for identifying uterine sarcomas. Thus initiatives to reduce unexpected endometrial carcinoma may be one promising area for improvement. These initiatives can benefit from additional research examining reasons for the unexpected malignancies (e.g., due to inadequate preoperative assessment or ineffectiveness of existing techniques) and comparative effectiveness of alternative evaluation strategies (e.g., different endometrium sampling method). 25 , 26 The risk factors and prediction models identified in our study may facilitate patient risk stratification. Consistent with prior research, 3 , 9 , 10 , 24 , 27 we found age and surgical indication as two important risk factors. Race/ethnicity, obesity, comorbidity, and personal history of malignancy were also pertinent factors. Differences in occult cancer risk across surgical routes, however, may be confounded by patient characteristics (e.g., younger women with lower risk for occult cancer were more likely to undergo subtotal hysterectomy, and women with genital prolapse who have lower risk for occult cancer frequently undergo vaginal hysterectomy). Turning knowledge on risk factors into effective decision aids may be one fruitful area of investigation. The prediction models in our study reflect a preliminary step in this direction. For instance, for patients with different age, race/ethnicity, and indication profiles, the models can predict their likelihood of having occult uterine cancer (see examples in Appendix E ). Research to further validate and improve these models using other databases and patient samples will be important for developing rigorous risk assessment tools for use by providers and patients in clinical care. However, given the high NPV and low PPV, these models may be more useful in ruling out occult cancers rather than confirming them. We recognize several limitations of this study. First, our data reflect patient population and clinical practice in one single state. The findings may not generalize to other places in the country. Second, a ten-year study period may mask changes in practice and hence prevalence of occult cancer. However, when stratified by years, we found similar rates of occult endometrial carcinoma (0.71% in 2003-2008 and 0.79% in 2009-2013) and uterine sarcoma (0.20% in 2003-2008 and 0.23% in 2009-2013). Third, we relied on retrospective data to identify hysterectomies performed for benign indications. Since not all preoperative suspicions for malignancy can be coded and identified in claims data, we might overestimate the risk of occult cancer. Indeed, it has been shown that retrospective studies generally report higher prevalence of occult uterine cancer than prospective studies. 4 , 28 Likewise, we measured medical history and comorbidities using hospital discharge records and a 9-month look-back window. As claims data has limited accuracy in capturing certain diseases 29 , 30 and not all patients had hospital encounters over the past 9 months, we might misclassify or underestimate some conditions. In summary, using statewide data from New York, we found that the overall risk of occult uterine cancer was 0.96% in women undergoing hysterectomy for presumed benign indications. Preoperative risk factors such as age and surgical indication may be used to guide risk-stratification and surgical planning. Efforts to enhance preoperative evaluation and develop effective risk assessment tools may facilitate future patient care.

This study used data from the 2003-2013 New York Statewide Planning and Research Cooperative System (SPARCS) – an all-payer data system capturing all inpatient and outpatient encounters at civilian hospitals as well as hospital-based and free-standing ambulatory surgery centers throughout the state. 12 For each encounter, the SPARCS data provided detailed information on patient sociodemographic and clinical characteristics, such as age, gender, race/ethnicity, admitting and discharge diagnosis, and procedure codes and dates. A unique personal identifier and date of birth allowed for longitudinal linkage of data within the same patient. Using the SPARCS data, we identified all hysterectomies performed for adult women in October 1, 2003-December 31, 2013 based on International Classification of Diseases Ninth Revision (ICD-9) procedure codes and current procedural terminology (CPT) codes ( Appendix A ). A 9-month period prior to surgery was used to measure medical histories. To identify occult uterine cancer, we further acquired data from the New York State Cancer Registry on all women diagnosed with corpus uteri cancer in 2003-2015. 13 Additional data were also available on women diagnosed with cancer of the cervix uteri or fallopian tube/uterine ligaments in this time period. For each of these patients, the cancer registry data included her complete diagnosis history (i.e., all cancers ever diagnosed) and diagnosis date and detailed characteristics of each cancer. We linked these cancer registry data to hysterectomy encounters in the SPARCS database using each patient’s unique identifier and date of birth. This study was approved by Yale University Human Investigation Committee. To study hysterectomies performed for presumed benign indications, we excluded patients with: 1) admitting diagnosis of any cancer, 2) discharge diagnosis indicating personal history of gynecologic malignancy, 3) encounters in the previous 9 months indicating a diagnosis of gynecologic cancer or cancer metastasis to female genital organs, or 4) documentation of corpus uteri, cervix uteri, or fallopian tube/uterine ligaments cancer in the cancer registry prior to date of hysterectomy. Patients with an admitting diagnosis of ascites or neoplasm of uncertain behavior/unspecified nature or with a history of endometrial hyperplasia (admitting diagnosis or in previous 9 months) were also excluded as they were often suspected of having cancer. To focus on a gynecologic patient population, we further excluded hysterectomies performed for obstetric conditions or diseases of the digestive system. Sensitivity analyses were performed by additionally excluding patients who underwent a radical hysterectomy or had any diagnosis of metastatic cancer in the 9 months before index hysterectomy. See Appendix B for more details. We defined a patient as having uterine cancer if she had International Classification of Diseases for Oncology 3rd Edition (ICD-O-3) site code 54.x or 55.x (excluding histology code 9050-9055, 9140, or 9590-9992). 14 In addition, we required an ICD-O-3 behavioral code of malignancy (i.e., excluding in situ disease). Uterine cancer newly diagnosed within 28 days after an index hysterectomy was defined as “occult” cancer, with sensitivity analyses conducted using alternative cutoffs at 6 and 24 months, respectively. For each identified occult uterine cancer, we characterized its subtype, stage, grade, and tumor size based on data from the cancer registry. Uterine cancer included two major types: endometrial carcinoma and uterine sarcoma. Within each type, we further distinguished more specific subtypes (e.g., adenocarcinoma, leiomyosarcoma) based on ICD-O-3 histology codes ( Appendix C ). 11 , 15 , 16 To minimize missing data, we categorized cancer stage as localized, regional, distant, or unknown by reconciling information from the American Joint Committee on Cancer stage variable and Surveillance, Epidemiology, and End Results Program stage variable. Grade was categorized as 1-4 or unknown; and tumor size was classified as 5 centimeters, or unknown. For each index hysterectomy, we ascertained patient age, race/ethnicity, surgical indication, and surgical route. We determined surgical indication based on ICD-9 admitting diagnosis code and classified it into categories (e.g., uterine fibroid, menstrual disorders, and genital prolapse) ( Appendix A ). We classified surgical route (e.g., total abdominal hysterectomy, laparoscopic supracervical hysterectomy) based on ICD-9 and CPT procedure codes ( Appendix A ). In addition, to characterize patients’ medical history and comorbidities, we used ICD-9 discharge diagnosis codes from encounters in the 9 months before index hysterectomy, as well as from the index hysterectomy itself (excluding diagnosis codes for gynecologic cancer as they might be newly diagnosed after hysterectomy). We required that the diagnosis code was either from the index admission or on at least one inpatient encounter or two outpatient encounters (>30 days apart) in the past 9 months. 17 This enabled us to measure the following binary indicators: 1) genetic susceptibility to malignant neoplasm of the breast, endometrium, ovary, and other female genital organ, respectively; 2) comorbidities including tobacco use, history of other cancer (breast cancer, colon cancer, melanoma, cancer of urinary organs, other solid tumor, and lymphoma, respectively), and eighteen measures of benign chronic conditions adapted from the Elixhauser index (e.g., hypertension, diabetes, and obesity); 18 and 3) family history of malignant neoplasm of breast, ovary, other female genital organs, and gastrointestinal tract, respectively. We selected these variables based on clinical relevance and availability of data. We summarized patient and tumor characteristics using descriptive statistics, and reported prevalence of occult uterine cancer (overall and by subtype) using point estimates and 95% confidence intervals (CIs). The CIs were calculated using exact binomial method if there were fewer than five cases of occult cancer or based on normal approximation otherwise. We analyzed data in the overall sample, as well as by surgical indication, route, and age category. We further developed a predictive model for patients’ risk of occult endometrial carcinoma and occult uterine sarcoma, respectively, using a random split sample approach (randomly selecting two thirds of the sample for model development while using the remaining sample for validation). In model development, we used a Poisson regression to estimate the risk ratio (RR) of having occult cancer associated with various patient characteristics. 19 Age, race/ethnicity, surgical indication, comorbidities, and genetic susceptibility to and family history of malignant neoplasm were candidate risk factors. Variables that were statistically significant at p<0.10 level in bivariate analysis were included in a backward selection process in multivariable regression to determine the final list of risk factors retained in the model (cutoff p value=0.05). We evaluated the performance of these predictive models in the validation sample. First, the models predicted each patient’s probability of having occult cancer. We then used the Youden index to determine the optimal threshold of the predicted probability for indicating occult cancer, 20 and classified each patient as having or not having occult cancer. By comparing this predicted occult cancer status with the cancer registry data, we calculated the models’ sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), overall accuracy, area under receiver operating characteristic (ROC) curve, and clinical utility index for rule-in and rule-out accuracy. 21 For clinical utility index, a value of 0.64-0.80 was considered good and a value ≥0.81 was considered excellent. 21 All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC).

Over 600,000 women undergo hysterectomy each year, making it one of the most common gynecologic procedures in the U.S. 1 Suspected presence of uterine cancer can have profound impact on surgical planning by influencing choice of surgical approach (e.g., total versus subtotal hysterectomy) and surgical team (e.g., involvement of gynecologic oncologists). 2 Some patients, however, have occult (pre-operatively unrecognized) uterine cancer which can pose unique challenges in care. They may require re-operation for surgical staging or removal of additional anatomy (cervix, adnexal organs, lymph nodes, and omentum) if not performed at initial hysterectomy. Uncontained laparoscopic power morcellation, if used at initial hysterectomy, may also inadvertently disseminate cancerous cells into the abdominopelvic cavity, jeopardizing prognosis. 3 , 4 Unfortunately, current knowledge on the prevalence of and risk factors for occult uterine cancer is still limited. Prior research was mostly based on medical record reviews from selected institutions with small sample sizes and non-representative samples, leading to large variability and uncertainty in estimated prevalence (e.g., a wide range of 0-3.17% for unexpected uterine cancer). 5 – 8 Conversely, studies at larger scales often relied on claims data and diagnosis codes to identify occult cancers, lacking histopathologic detail to confirm or adequately characterize the malignancy (e.g., inability to distinguish cancer subtypes). 9 , 10 Moreover, existing data tend to focus on uterine sarcoma overlooking endometrial carcinoma which is the more prevalent type of uterine cancer; 11 and rigorous assessment of risk factors has been sparse due to the small number of patients with occult uterine cancer in prior studies. These limitations and knowledge gaps, along with limited screening tests available, have hindered our ability in managing occult uterine cancer in hysterectomies. To better inform risk-stratification in patient care, we constructed a large, population-based sample of women undergoing hysterectomy for presumed benign indications in New York State with linkage to cancer registry data, and examined the prevalence of occult uterine cancer (including all major subtypes) and associated tumor characteristics. Moreover, we thoroughly evaluated differences in risk of occult uterine cancer across different patient subgroups (by age, surgical indication, and surgical route) and developed risk prediction models based on patients’ preoperative characteristics.