Background
High-volume center surgery and gynecologic oncology care are associated with improved outcomes for women with uterine cancer. Referral patterns, from biopsy through to chemotherapy, may have patients interacting with high-volume centers for all, a portion, or none of their care. The relative frequency, the underlying factors that contribute to referral, and the potential impact of these referral patterns on treatment outcomes are unknown.
Objective
We sought to analyze the referral patterns and subsequent impact of care sites on treatment for women with high- and low-risk uterine cancer.
Study Design
This is a population-based retrospective cohort study of uterine cancer cases from 2004 through 2009 in North Carolina. Using state cancer registry files linked to Medicare, Medicaid, and private payer insurance claims, we analyzed referral and treatment patterns by annual surgical volume (high ≥12/y). We examined clinical and demographic factors associated with referral and used modified Poisson regression to evaluate risk of referral, lymphadenectomy, and chemotherapy. Stratified Kaplan-Meier plots and Cox proportional hazard models were used to examine survival.
Results
A total of 2053 women were analyzed, including 34% (n = 677) with grade 3 histology. Of 1630 (80%) women with preoperative biopsies, referral patterns (biopsy to surgery) were: low volume to high volume (n = 652, 40%), followed by high volume to high volume (n = 605, 37%), then low volume to low volume (n = 318, 20%), and the rare high volume to low volume (n = 50, 3%). Women retained in low-volume centers after biopsy were older, were less likely to have private insurance, and had more comorbidities. High-risk histology (aRR, 1.14; 95% confidence interval, 1.04–1.25) was positively associated with referral, while Medicaid insurance was negatively associated with referral (aRR, 0.64; 95% confidence interval, 0.42–0.96). Most women (74%, n = 1557) had surgery at high-volume centers. Lymphadenectomy was less likely at low-volume centers (aRR, 0.71; 95% confidence interval, 0.64–0.77). Similarly, for high-risk patients, the relationship between low-volume center surgery and subsequent chemotherapy was aRR, 0.71 (95% confidence interval, 0.48–1.02). Of 290 women who received chemotherapy, the referral patterns (surgery to chemotherapy) were: high volume–all (high volume to high volume), high volume–hybrid (high volume to low volume, or low volume to high volume), and high volume–none (low volume to low volume). In all, 36% (n = 104/290) received chemotherapy at a low-volume center, the majority (68%, n = 71/104) of whom were referred from high-volume centers after surgery. Crude, unadjusted mortality risk of chemotherapy recipients differed by referral pattern (surgery to chemotherapy): high volume–all patients (hazard ratio, 1.0; referent), followed by high volume–hybrid (hazard ratio, 1.33; 95% confidence interval, 0.93–1.91) then high volume–none patients (RR, 1.95; 95% confidence interval, 1.24–3.08).
Conclusion
Most women with uterine cancer treated at high-volume centers arrive through referral, which is affected by age and type of insurance, in addition to histology. For high-risk women who require chemotherapy, survival may be related to the extent of treatment received at high-volume centers.
Introduction
The majority of uterine cancer patients undergo surgery by general gynecologists, despite documented benefits of gynecologic oncology treatment. Women most likely to benefit from a gynecologic oncologist are older women (age >70 years) and women with high-risk histology (grade 3 and/or nonendometrioid). In addition, gynecologic oncology training is highly correlated with high annual surgical volume, which in turn has been associated with improved perioperative outcomes in uterine cancer patients.
Despite these data, the frequency with which women with uterine cancer are referred to high-volume (HV) centers is unclear. To our knowledge, all studies assessing surgical volume in uterine cancer have focused on the single surgical episode. This excludes information on the referral patterns both before and after surgery. One critical transition point in uterine cancer care is the step from diagnostic biopsy to receipt of surgery. Women may be initially diagnosed at HV centers and remain there for surgery. On the other hand, perhaps more often, they may be initially diagnosed in a primary care, low-volume (LV) setting, and then subsequently referred to a HV center. Another critical transition point, for high-risk patients, is from surgery to chemotherapy. For patients with high-risk histology, it is especially important to understand these referral patterns. They experience higher mortality and often require adjuvant chemotherapy for the best chance of cure.
There is no evidence describing the proportion of uterine cancer patients who follow the different referral patterns; the influence of histology, demographics, or other factors on these patterns; or their impact on clinical outcomes.
The goal of this study was to examine these questions at a population level, including women of all ages, enrolled in Medicare, Medicaid, and privately insured health plans. Specifically, we sought to: (1) describe the relative frequency and demographic features of women in each referral pattern, and (2) analyze the association of demographic and clinical factors with likelihood of referral and subsequent treatment and outcomes.
Materials and Methods
Data source and study population
This study was approved by the North Carolina Institutional Review Board (no. 13-2863). The North Carolina Central Cancer Registry (NCCCR), a state-level mandatory reporting registry, was used to identify all women in North Carolina diagnosed with a primary uterine cancer, from Jan. 1, 2004, through June 30, 2009. Women who were diagnosed at death or autopsy, had a prior cancer diagnosis, or had a major primary disability were excluded using NCCCR flags and International Classification of Diseases for Oncology, Third Edition site and morphology codes. The North Carolina Integrated Cancer Information and Surveillance System links identified cancer cases from the NCCCR with administrative data from Medicare, Medicaid, and beneficiaries in privately insured health plans across the state. Over 80% of the unique cancer registry patients link to at least 1 of the administrative databases captured in the Integrated Cancer Information and Surveillance System data. The NCCCR is a gold-certified registry based on the North American Association of Central Cancer Registries standards and includes 99% valid, complete social security numbers, as well as other key identifiers such as first, maiden, and last names; date of birth; and address. The Medicare and Medicaid beneficiaries are linked to the registry via the Centers for Medicare and Medicaid Services contractor General Dynamics using a deterministic, exact match on social security numbers. The privately insured beneficiaries are linked through both deterministic (Surveillance, Epidemiology, and End Results–Medicare algorithm) and probabilistic algorithm incorporating Bayes formula, which results in a linkage of 100% sensitivity, 98% specificity, and 95% positive predictive value. Due to administrative lag in data availability, the Medicaid enrollment file extends to 2008, while all other enrollment data are through 2009. We restricted the sample to women with linked, continuous enrollment, in any payer, from 6 months prior to diagnosis date to 6 months after diagnosis date. This allowed for accurate capture of comorbidity, diagnosis, and treatment data. Due to small numbers, women with missing or unknown race/ethnicity information were excluded.
Tumor histology and morphology codes were grouped into the following categories: endometrioid adenocarcinoma, serous carcinoma, carcinosarcoma, sarcoma, and other ( Supplemental Table A.1 ). High-risk histology was defined as grade 3–4 and/or any nonendometrioid histology. Grade 3 (poorly differentiated) and 4 (undifferentiated) were combined into 1 category of grade 3 as this reflects how they are defined and treated clinically. For the purposes of assessing frequency of chemotherapy delivery, the population denominator was restricted to nonendometrioid histology, as this group is uniformly recommended to have chemotherapy by national guidelines.
Exposure and outcome variables
Histology type and treatment sites were the primary exposure variables. Treatment sites were identified using ZIP codes on provider billing claims for each episode of care (biopsy, surgery, chemotherapy). The ZIP codes were then categorized into HV centers based on uterine cancer hysterectomy volume (≥60 cases) during the 5-year study period. In addition, the presence of a gynecologic oncology specialist practicing at that location as determined from the Society of Gynecologic Oncology membership database was determined. There was high concordance (89%) between these 2 criteria, and case volume was used first to define the HV and LV groups. The numerical cut-off for surgical volume was consistent with prior studies, using cut-offs ranging from 10-14.5 mean cases/y. ZIP codes that identified outreach practice locations of known gynecologic oncologists were classified with the HV center of the gynecologic oncologist’s primary practice location (n = 6). The 4 potential referral patterns included HV biopsy to HV surgery (HV-HV), HV biopsy to LV surgery (HV-LV), LV biopsy to HV surgery (LV-HV), and LV biopsy to LV surgery (LV-LV). Volume status, at each point of care (biopsy, surgery, chemotherapy) was therefore based on uterine cancer hysterectomy volume of that ZIP code. For example, a biopsy done by a general gynecologist at a tertiary center that had high uterine cancer hysterectomy volume would be classified as HV biopsy.
Primary study outcomes were referral pattern, performance of lymphadenectomy, and chemotherapy administration for high-risk histology. Exploratory analysis of all-cause mortality was also performed. Hysterectomy was defined using Current Procedural Terminology codes for hysterectomy, occurring at the time of or after cancer diagnosis ( Supplemental Table A.1 ). Lymphadenectomy was defined using the nodal staging information from the cancer registry, which specifies whether or not lymph nodes were removed. Chemotherapy administration was defined by the presence of International Classification of Diseases, Ninth Revision administration or Healthcare Common Procedure Coding System medication codes in the claims, using an algorithm validated in many cancer types, including ovarian. Mortality information came from the state cancer registry, updated through 2014.
Covariates
Covariates included age, race/ethnicity, population density of residence county, stage at diagnosis, comorbidity, and insurance payer. Age at diagnosis and race/ethnicity were reported from the NCCCR. Due to small sample size with granular race and ethnicity categories, all non-white, non-Hispanic categories were grouped into “racial and ethnic minorities.” Stage at diagnosis was reported with the summary staging variable (local, regional, distant, unknown) common to state and national cancer registries, and broadly corresponding to International Federation of Gynecology and Obstetrics stage 1 (local), stage 2-3 (regional), and stage 4 (distant) categorization. There were no changes in staging classification during the study period. Rural/urban classification from the US Department of Agriculture was dichotomized at the county level into metro vs nonmetro based on rural/urban continuum codes from 2013. Comorbidity was assessed using methods reported by Gagne et al, which incorporate both the Charlson comorbidity index and the Elixhauser comorbidity score, to provide the most comprehensive assessment of health status. Data from outpatient and inpatient clinical settings are incorporated into the scoring system. Those with scores ≥1 represent patients with a comorbidity profile associated with greater hospitalization and health care utilization and/or greater mortality risk. In this scale comorbidity values can be <0 for conditions actually associated with decreased health care use. This group was nearly identical to those with scores of 0, and therefore comorbidity was dichotomized into ≤0 vs ≥1. For multivariate analysis, insurance payers were grouped into 3 mutually exclusive categories of any private payer, Medicare only, or any Medicaid.
Statistical analysis
We performed univariate and bivariate analysis of histology type, the covariates and the primary outcomes of referral pattern, lymphadenectomy, and chemotherapy receipt. The χ 2 statistic, Student t test, and analysis of variance tests were used to assess the relationships between independent variables and outcome variables. We constructed multivariable models using modified Poisson regression to generate risk ratios of referral pattern, lymphadenectomy, and chemotherapy receipt. Survival was explored by generating Kaplan-Meier curves stratified by care model type and Cox proportional hazard models to generate crude hazard ratios (HR) for time to death (mortality). Statistical significance was set at P < .05. Analysis was performed using software (SAS v9.3; SAS Institute Inc, Cary, NC).
Materials and Methods
Data source and study population
This study was approved by the North Carolina Institutional Review Board (no. 13-2863). The North Carolina Central Cancer Registry (NCCCR), a state-level mandatory reporting registry, was used to identify all women in North Carolina diagnosed with a primary uterine cancer, from Jan. 1, 2004, through June 30, 2009. Women who were diagnosed at death or autopsy, had a prior cancer diagnosis, or had a major primary disability were excluded using NCCCR flags and International Classification of Diseases for Oncology, Third Edition site and morphology codes. The North Carolina Integrated Cancer Information and Surveillance System links identified cancer cases from the NCCCR with administrative data from Medicare, Medicaid, and beneficiaries in privately insured health plans across the state. Over 80% of the unique cancer registry patients link to at least 1 of the administrative databases captured in the Integrated Cancer Information and Surveillance System data. The NCCCR is a gold-certified registry based on the North American Association of Central Cancer Registries standards and includes 99% valid, complete social security numbers, as well as other key identifiers such as first, maiden, and last names; date of birth; and address. The Medicare and Medicaid beneficiaries are linked to the registry via the Centers for Medicare and Medicaid Services contractor General Dynamics using a deterministic, exact match on social security numbers. The privately insured beneficiaries are linked through both deterministic (Surveillance, Epidemiology, and End Results–Medicare algorithm) and probabilistic algorithm incorporating Bayes formula, which results in a linkage of 100% sensitivity, 98% specificity, and 95% positive predictive value. Due to administrative lag in data availability, the Medicaid enrollment file extends to 2008, while all other enrollment data are through 2009. We restricted the sample to women with linked, continuous enrollment, in any payer, from 6 months prior to diagnosis date to 6 months after diagnosis date. This allowed for accurate capture of comorbidity, diagnosis, and treatment data. Due to small numbers, women with missing or unknown race/ethnicity information were excluded.
Tumor histology and morphology codes were grouped into the following categories: endometrioid adenocarcinoma, serous carcinoma, carcinosarcoma, sarcoma, and other ( Supplemental Table A.1 ). High-risk histology was defined as grade 3–4 and/or any nonendometrioid histology. Grade 3 (poorly differentiated) and 4 (undifferentiated) were combined into 1 category of grade 3 as this reflects how they are defined and treated clinically. For the purposes of assessing frequency of chemotherapy delivery, the population denominator was restricted to nonendometrioid histology, as this group is uniformly recommended to have chemotherapy by national guidelines.
Exposure and outcome variables
Histology type and treatment sites were the primary exposure variables. Treatment sites were identified using ZIP codes on provider billing claims for each episode of care (biopsy, surgery, chemotherapy). The ZIP codes were then categorized into HV centers based on uterine cancer hysterectomy volume (≥60 cases) during the 5-year study period. In addition, the presence of a gynecologic oncology specialist practicing at that location as determined from the Society of Gynecologic Oncology membership database was determined. There was high concordance (89%) between these 2 criteria, and case volume was used first to define the HV and LV groups. The numerical cut-off for surgical volume was consistent with prior studies, using cut-offs ranging from 10-14.5 mean cases/y. ZIP codes that identified outreach practice locations of known gynecologic oncologists were classified with the HV center of the gynecologic oncologist’s primary practice location (n = 6). The 4 potential referral patterns included HV biopsy to HV surgery (HV-HV), HV biopsy to LV surgery (HV-LV), LV biopsy to HV surgery (LV-HV), and LV biopsy to LV surgery (LV-LV). Volume status, at each point of care (biopsy, surgery, chemotherapy) was therefore based on uterine cancer hysterectomy volume of that ZIP code. For example, a biopsy done by a general gynecologist at a tertiary center that had high uterine cancer hysterectomy volume would be classified as HV biopsy.
Primary study outcomes were referral pattern, performance of lymphadenectomy, and chemotherapy administration for high-risk histology. Exploratory analysis of all-cause mortality was also performed. Hysterectomy was defined using Current Procedural Terminology codes for hysterectomy, occurring at the time of or after cancer diagnosis ( Supplemental Table A.1 ). Lymphadenectomy was defined using the nodal staging information from the cancer registry, which specifies whether or not lymph nodes were removed. Chemotherapy administration was defined by the presence of International Classification of Diseases, Ninth Revision administration or Healthcare Common Procedure Coding System medication codes in the claims, using an algorithm validated in many cancer types, including ovarian. Mortality information came from the state cancer registry, updated through 2014.
Covariates
Covariates included age, race/ethnicity, population density of residence county, stage at diagnosis, comorbidity, and insurance payer. Age at diagnosis and race/ethnicity were reported from the NCCCR. Due to small sample size with granular race and ethnicity categories, all non-white, non-Hispanic categories were grouped into “racial and ethnic minorities.” Stage at diagnosis was reported with the summary staging variable (local, regional, distant, unknown) common to state and national cancer registries, and broadly corresponding to International Federation of Gynecology and Obstetrics stage 1 (local), stage 2-3 (regional), and stage 4 (distant) categorization. There were no changes in staging classification during the study period. Rural/urban classification from the US Department of Agriculture was dichotomized at the county level into metro vs nonmetro based on rural/urban continuum codes from 2013. Comorbidity was assessed using methods reported by Gagne et al, which incorporate both the Charlson comorbidity index and the Elixhauser comorbidity score, to provide the most comprehensive assessment of health status. Data from outpatient and inpatient clinical settings are incorporated into the scoring system. Those with scores ≥1 represent patients with a comorbidity profile associated with greater hospitalization and health care utilization and/or greater mortality risk. In this scale comorbidity values can be <0 for conditions actually associated with decreased health care use. This group was nearly identical to those with scores of 0, and therefore comorbidity was dichotomized into ≤0 vs ≥1. For multivariate analysis, insurance payers were grouped into 3 mutually exclusive categories of any private payer, Medicare only, or any Medicaid.
Statistical analysis
We performed univariate and bivariate analysis of histology type, the covariates and the primary outcomes of referral pattern, lymphadenectomy, and chemotherapy receipt. The χ 2 statistic, Student t test, and analysis of variance tests were used to assess the relationships between independent variables and outcome variables. We constructed multivariable models using modified Poisson regression to generate risk ratios of referral pattern, lymphadenectomy, and chemotherapy receipt. Survival was explored by generating Kaplan-Meier curves stratified by care model type and Cox proportional hazard models to generate crude hazard ratios (HR) for time to death (mortality). Statistical significance was set at P < .05. Analysis was performed using software (SAS v9.3; SAS Institute Inc, Cary, NC).
Results
Study population and sites of care
A total of 6180 women were diagnosed with uterine cancer in North Carolina from Jan. 1, 2004, through June 30, 2009. Of these, 1396 (23%) were excluded for: diagnosis at autopsy (n = 37), prior cancer diagnosis (n = 771), and primary disability (n = 588). Of the remaining 4784 women, 2461 had linked, continuous enrollment in 1 of the 3 payers for at least 6 months prior to diagnosis to 6 months after diagnosis. After further exclusions at the claims level, 2053 women comprised the cohort ( Figure 1 ).
In all, 24 sites of care were identified and grouped into 18 distinct practices. Half (9/18) were identified as HV centers and half as LV centers. Most patients had surgery at HV centers (n = 1557 [74%]). LV centers had more patients who were in the tails of the age distribution (20-49 and ≥75 years), who resided in nonmetro counties, who had public insurance, and who had higher comorbidity scores ( Table 1 ).
| Characteristic | All N = 2053 | High volume n = 1557 | Low volume n = 496 | P |
|---|---|---|---|---|
| Age, y b | 66.5 ± 12 | 66.2 ± 12 | 67.2 ± 13 | .106 |
| 20–49 | 172 (8) | 123 (8) | 49 (10) | |
| 50–64 | 609 (30) | 492 (32) | 117 (24) | |
| 65–75 | 732 (36) | 546 (35) | 186 (38) | |
| >75 | 540 (26) | 396 (25) | 144 (29) | .006 |
| Race and ethnicity | ||||
| White, non-Hispanic | 1714 (83) | 1308 (84) | 412 (83) | |
| Racial/ethnic minority | 339 (17) | 255 (16) | 84 (17) | .771 |
| Population density of residence county | ||||
| Metro c | >1368 (67) | >1082 (69) | >286 (58) | |
| Nonmetro | 664 (32) | 464 (30) | 200 (40) | |
| Missing c | ≤ 20 | ≤10 | ≤10 | <.001 |
| Insurance payer | ||||
| Medicare only | 1096 (53) | 798 (52) | 298 (60) | |
| Private | 896 (43) | 718 (46) | 174 (35) | |
| Medicaid | 65 (3) | 41 (3) | 24 (5) | <.001 |
| Year of diagnosis | ||||
| 2004 | 337 (16) | 234 (15) | 103 (21) | |
| 2005 | 371 (18) | 247 (16) | 124 (25) | |
| 2006 | 339 (17) | 255 (16) | 84 (17) | |
| 2007 | 363 (18) | 289 (19) | 74 (15) | |
| 2008 | 417 (20) | 339 (22) | 78 (16) | |
| 2009 | 226 (11) | 193 (12) | 33 (7) | <.001 |
| Summary stage | ||||
| Local | 1513 (74) | 1143 (73) | 370 (75) | |
| Regional | 399 (19) | 306 (20) | 93 (19) | |
| Distant | 111 (5) | 92 (6) | 19 (4) | |
| Missing/unknown c | ≤20 | ≤10 | ≥10 | .009 |
| Grade | ||||
| 1 | 793 (39) | 596 (38) | 197 (40) | |
| 2 | 551 (27) | 421 (27) | 130 (26) | |
| 3 | 592 (29) | 478 (31) | 114 (23) | |
| Unknown | 117 (6) | 62 (4) | 55 (11) | <.001 |
| Histology | ||||
| Low risk | 1342 (65) | 999 (64) | 343 (69) | |
| High risk d | 711 (35) | 558 (36) | 153 (31) | .042 |
| Comorbidity | ||||
| 0 | 1585 (77) | 1219 (78) | 366 (74) | |
| ≥1 | 468 (23) | 338 (22) | 130 (26) | .037 |
| Biopsy before surgery | ||||
| No | 418 (20) | 295 (19) | 123 (25) | |
| Yes | 1635 (80) | 1262 (81) | 373 (75) | .005 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
