at SciVerse ScienceDirect

Clinical Oncology 24 (2012) e113ee124

Contents lists available

Clinical Oncology

journal homepage: www.cl in icaloncologyonl ine.net

Original Article

A Population-Based Study of Factors Affecting the Use of Radiotherapy forEndometrial Cancer

T.P. Hanna *y, H. Richardson *z, Y. Peng *x, W. Kong *, J. Zhang-Salomons *, W.J. Mackillop *y*Division of Cancer Care and Epidemiology, Queen’s University Cancer Research Institute, Kingston, Ontario, CanadayCancer Centre of Southeastern Ontario, Kingston, Ontario, CanadazNational Cancer Institute of Canada, Clinical Trials Group, Kingston, Ontario, CanadaxDepartment of Mathematics and Statistics, Queen’s University, Kingston, Ontario, Canada

Received 8 December 2010; received in revised form 14 December 2011; accepted 27 January 2012

Abstract

Aims: To describe the use of adjuvant radiotherapy for endometrial cancer in Ontario, and identify factors associated with its use, and to determine whethervariation in the use of radiation is associated with differences in survival.Materials and methods: This was a retrospective, population-based, cohort study of all patients who had a hysterectomy for endometrial cancer in Ontariobetween 1992 and 2003. We used multiple logistic regression to identify health system-related factors associated with the use of radiotherapy, while controllingfor disease- and patient-related factors. Survival and cancer cause-specific survival were compared among regions of the province with higher and lower rates ofuse of radiotherapy.Results: The study population included a total of 9411 women with a median age of 63 years. Overall, 26.2% received adjuvant radiotherapy. Patients livingfurther from regional cancer centres were slightly less likely to receive radiation (P ¼ 0.02). Patients who had their surgery during longer prevailing waitingtimes for radiotherapy were less likely to receive radiation (P ¼ 0.04). The use of radiotherapy varied widely from 18.0 to 34.3% among the catchment areas ofprovincial radiotherapy centres (P < 0.0001). In the overall population, there was no difference in survival among regions with higher and lower rates of use ofradiotherapy. However, in the subgroup of cases with clear cell and serous carcinomas, both overall survival and cancer cause-specific survival were significantlylower in regions with lower rates of use of radiotherapy (P < 0.05). This difference remained significant after controlling for other factors (P < 0.05; hazard ratio1.43; 95% confidence limits 1.06e1.93).Conclusions: Health system-related factors unrelated to patients’ needs affect the use of adjuvant radiotherapy in Ontario. Lower rates of use of adjuvantradiotherapy are associated with lower rates of survival in patients with serous and clear cell carcinomas.� 2012 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

Key words: Endometrial neoplasms; health services accessibility; health services research; outcome assessment (healthcare); radiotherapy; survival rate

Introduction

It is estimated that between 34 to 58% of all patients withendometrial cancer will have an indication for externalbeam radiation at some point in the course of their disease[1]. There is no doubt that adjuvant radiotherapy reducesthe risk of pelvic failure after surgery for early endometrialcancer and there is some evidence that it improves survivalin high-risk stage I cases [2,3]. Randomised trials have,however, failed to show a survival advantage from

Author for correspondence: W.J. Mackillop, Division of Cancer Care andEpidemiology, Queen’s University Cancer Research Institute,10 Stuart St. Level2, Kingston, ON, Canada K7L 3N6. Tel: þ613-533-6895; Fax: þ613-533-6794.

E-mail address: William.Mackillop@krcc.on.ca (W.J. Mackillop).

0936-6555/$36.00 � 2012 The Royal College of Radiologists. Published by Elsevdoi:10.1016/j.clon.2012.01.007

radiotherapy in intermediate-risk stage I patients and therole of postoperative radiotherapy in this large groupremains controversial [4e7]. A guideline published byOntario’s Program in Evidence-Based Care recommendsadjuvant radiotherapy for stage I patients with a high risk ofrecurrence (stage IC, grade 3), and recommends againstradiotherapy in those at low risk of recurrence (stage IA, IB,grades 1 and 2) [8]. This guideline states that radiotherapy’is a reasonable treatment option’ for patients at interme-diate risk of recurrence (stage IA, IB, grade 3 or stage IC,grades 1 and 2), and recommends that the decisions aboutradiotherapy in this situation should be based on consid-eration of the risks and benefits in the individual patient [8].Given the continuing scientific controversies about theappropriate role of radiotherapy in early endometrial

ier Ltd. All rights reserved.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124e114

cancer, it is not surprising that variations in the beliefs ofindividual doctors and variations in access to radiotherapyhave been shown to influence the use of adjuvant radio-therapy in this situation [9].

Previous studies have shown variations in the use ofradiotherapy in a number of clinical situations, and haveidentified potential barriers to access to radiotherapy inOntario [10e12]. Older patients, residents of poorercommunities, and those who reside further from a radio-therapy centre, have consistently been shown to be lesslikely than others to receive radiotherapy. Moreover, insome situations, there are large inter-regional variations inthe use of radiotherapy, which are unexplained by varia-tions in case mix [10,11]. It is not known whether suchfactors affect the use of postoperative radiotherapy forendometrial cancer, or whether variations in the use ofradiation in this context affect outcomes at the populationlevel. These questions can only be answered by population-based outcome studies [13].

The first objective of this study was to describe the use ofpostoperative radiation for endometrial cancer in Ontariobetween 1992 and 2003, and to identify potential barriersto the use of radiotherapy in this context. The generallyaccepted indications for the use of radiotherapy at that timewere as described in Ontario’s current practice guideline[8], but we hypothesised that the use of postoperativeradiotherapymight also vary with factors affecting access toradiotherapy, such as distance of patient’s residence to thenearest radiotherapy centre, or the prevailing waiting listfor radiotherapy. We also hypothesised that there might bedifferences in the use of postoperative radiotherapybetween patients who have their surgery in hospitals withspecialist expertise in gynaecological oncology and thosemanaged in less specialised general hospitals. The secondobjective of this study was to explore the relationshipbetween rates of use of radiotherapy and survival in thegeneral population, particularly in the context of high-riskhistologies. Unfortunately, we were unable to measurerates of pelvic recurrence in this study because the relevantinformation was not available at the population level.

Materials and Methods

Ontario’s Health System for the Management ofGynaecological Malignancies

Ontario’s one-tier, publically funded health care systemprovides cancer surgery, radiotherapy and hospital care,without direct charge, for all Ontario residents. There is noparallel private sector that offers these services. Bothgynaecological oncologists and general gynaecologists areinvolved in the care of patients with endometrial cancer inOntario. Gynaecological oncologists are usually confined tolarge hospitals in urban areas. Radiotherapy is availableonly at provincial cancer centres located in Ontario’s largercities. At the time of this study, there were nine radio-therapy centres in Ontario. There were no formal guidelinesfor the use of radiotherapy in endometrial cancer in Ontario

during the study period. We believe, however, that thegeneral approach to the use of radiotherapy in stage Iendometrial cancer in that era is accurately reflected in theguideline published not long afterwards by Ontario’sProgram in Evidence-Based Care. This was summarisedabove and is readily available online [8]. There are still noformal provincial guidelines for stage II and III cases, or forthose with clear cell or serous histologies, but in most casesthese characteristics would probably have been consideredas indications for radiotherapy in that era. Some radio-therapy centres also have their own in-house treatmentguidelines, but ultimately all decisions about the use ofradiotherapy were, and still are, made by individual radia-tion oncologists and patients.

Study Population

This study was a retrospective, population-based studyof all patients diagnosed with endometrial cancer inOntario between 1992 and 2003, who underwent a simpleor radical hysterectomy. Only patients with adenocarci-noma, clear cell carcinoma or serous carcinoma by ICD-0 criteria were included (adenocarcinoma 8140/3, 8210/3,8380/3, 8570/3, clear cell carcinoma 8005/3, 8310/3, andserous carcinoma 8460/3, 8461/3). Patients with anyprevious malignancy were excluded. Patients known tohave distant metastases at diagnosis and those whose firstradiation after their primary surgery was described aspalliative were excluded. Cases of endometrial cancerdiagnosed by autopsy or death certificate were excluded.Patients dying within 90 days of surgery were excludedbecause they had not lived long enough to ascertainwhether they would have received postoperative radiation,had they lived longer.

Sources of Data

Ontario Cancer RegistryThe Ontario Cancer Registry (OCR) is a passive,

population-based registry that collects administrative dataabout all cancer diagnosed in the province and identifiesindividual cases through a process of probabilistic linkage.The OCR is known to be at least 95% complete for all sitescombined [14,15]. The OCR record includes: date of birth;postal code; Ministry of Health (MOH) residence code; dateof cancer diagnosis; histological diagnosis in ICD-0 morphology code; and primary site in ICD-9, ICD-9-CMor ICD-10 topography code. The OCR database is regularlylinked to the provincial and national death registries toprovide information about date and cause of death. Figure 1describes the sources of additional administrative data thatwere linked to the OCR to describe the management ofendometrial carcinoma, as described previously [10,11].

Canadian Institute of Health InformationThe Canadian Institute of Health Information (CIHI)

Discharge Abstract database includes information about allhospital admissions in Ontario, including: admission anddischarge dates; diagnosis most responsible for admission

Fig. 1. The Endometrial Cancer Database.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124 e115

and other diagnoses, in ICD-9 or ICD-10 code; and surgicalor other procedures coded using the Canadian Classificationof Diagnostic, Therapeutic and Surgical Procedures (CCP)code.

Radiotherapy databaseThis database was assembled at the Queen’s University

Cancer Research Institute Division of Cancer Care andEpidemiology with permission from each of the nineprovincial cancer centres operating during the period understudy. These centres have kept electronic radiation therapyrecords in uniform format since 1982. The databaseincludes records of: anatomic site irradiated, type of radi-ation; treatment; dose; number of fractions; and dates of alltreatments. Over the period of this study, radiotherapy foralmost all residents of Ontario was provided exclusively bynine provincial cancer centres. However, residents of themost westerly county of Kenora were usually referred forradiotherapy to Winnipeg in the neighbouring province ofManitoba from which radiotherapy records were notavailable. The rate of radiotherapy in Kenora was thereforenot reported (see below).

Definition of Variables Used in the Analysis

Disease characteristicsNeither tumour stage nor grade was available for most

patients in this cohort. Clinical and pathologic TNM stageare now routinely collected and compiled at all provincialcancer centres, but this practice was only uniformly adop-ted after 2002. There is still no information about stageavailable for patients who are not seen at a cancer centre. Inthis study, information about stage was, therefore, onlyavailable for patients seen at a cancer centre who were

diagnosed in 2003. For that period, the fifth edition of UICCTNM was used, which corresponds precisely to the 1988FIGO Staging System. In this study, pathological TNM wasused to assign stage group in most cases, and clinical TNMwas used in the remainder.

Neighbourhood income quintileThe neighbourhood income quintile was used as an

ecological measure of each patient’s socioeconomic status.Information about the median household income fromStatistics Canada was linked to each case in the OCR. Thelinkage was from census enumeration area to the patient’spostal code, or from census subdivision to the patient’sMOH residence code, as described previously [16].

ComorbidityComorbidity was scored using the Elixhauser comor-

bidity score, based on all diagnoses in the CIHI record of thepatient’s admission for hysterectomy. The Elixhauser scoreis a validated scoring system for comorbidity data extractedfrom electronic records [17e19]. Its performance charac-teristics have been compared with the Deyo adaption of theCharlson score and found to be superior in predicting in-hospital mortality for cancer patients and other patientpopulations [17e20].

Classification of surgical proceduresPatients were subdivided using CIHI procedure data into

those receiving a simple hysterectomy (CCP codes 80.3,80.4) or radical hysterectomy (CCP codes 80.5, 80.6). CCPcodes for surgical staging of lymph nodes were 52.19, 52.2,52.4, 52.41, 52.44, 52.49, 52.81 and 52.89. These had tooccur during the patient’s admission for hysterectomy forendometrial cancer to be accepted as a surgical staging oflymph nodes. CCP procedure code 66.82 was used to iden-tify cases of peritoneal biopsy, the definition of whichincludes biopsy of mesentery, omentum or peritonealimplant and does not include peritoneal washings.

Classification of hospitalsHospitals that had both expertise in gynaecological

oncology and a radiotherapy centre on site were classifiedas comprehensive gynaecological oncology centres(CGOCs). We used the 1 April 2006 Ontario Ministry ofHealth and Long Term Care Master Numbering System toidentify hospital amalgamations and name changes [21].

The most responsible radiotherapy centre was deter-mined for each hospital in the province. This was defined asthe radiotherapy centre at which most endometrial cancerpatients diagnosed at that hospital received adjuvantradiotherapy. If hospitals had less than 20 patients whoreceived radiation and less than 70% of these receivedradiation at the same cancer centre, the most responsibleradiotherapy centre was defined using information onradiation use for other common cancer surgeries using CCPsurgical codes specified by Bardell et al. [22]. If no patientsfrom a given hospital had ever received adjuvant radiationfor any indication, the closest cancer centre was assigned asthe most responsible cancer centre.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124e116

Prevailing waiting time for radiotherapyThe concept of prevailing waiting time for radiotherapy

was first described by Huang [23]. For the purposes of thisstudy, it was defined as themedian time from surgery to thestart of adjuvant radiation for cancer at the most respon-sible radiotherapy centre, at the time of the index patient’ssurgery. Specifically, the prevailing waiting time attributedto each case in this study was the median waiting time foradjuvant radiotherapy observed at the most responsibleradiotherapy centre over a period starting 4 months beforeher surgery and ending 2 months after it.

Distance to most responsible radiotherapy centreThis was defined as the linear distance between the

geometric centre of the postal code, or MOH residence code,within which the patient resided, and the most responsibleradiotherapy centre.

Adjuvant radiotherapy was defined here as radiotherapyto the pelvis givenwith curative or adjuvant intent within 6months of a hysterectomy for endometrial cancer. The 6month cut-off was chosen using available information ontiming of adjuvant radiation and treatment of recurrenceavailable for a representative subset of the data. Externalbeam radiation and brachytherapy were both included asforms of adjuvant radiation for the purpose of this study.

Overall survival and cancer-specific survivalOverall survival and cancer-specific survivalweredescribed

by the KaplaneMeier method. Equality of strata was testedusing theWilcoxon test. Factors associatedwith survival wereevaluated in a Cox model. Local control could not be assessedin this study because information about local recurrence is notroutinely collected at the population level in Ontario.

Statistical Analysis

We used SAS version 9.1 for statistical analysis. Relation-shipsbetween independentvariables and theuse of radiationwere assessed using chi-squared statistics. Exploration ofconfounding and effect modification was carried out usingstratified contingency tables and logistic regression. Thestrength of association between independent variables andthe use of radiation was estimated using logistic regression.Both crude and adjusted odds ratios were computed toexamine the probability of radiation use for each factor ofinterest. The Hosmer and Lemeshow Goodness-of-Fit Testwas used to assess the degree of fit of the fixed effects modeland the Likelihood Ratio Test was used to evaluate thepredictive value of the model. Multi-level multiple logisticregression was used to assess and control for the potentialeffects of data clustering by operating hospital.

Results

Patient Characteristics

In total, 15 291 patients with endometrial cancer werediagnosed in Ontario between 1 January 1992 and 31

December 2003. Of these, 2314 were excluded because theydid not have a hysterectomy (some of the excluded casesalso lacked a microscopic diagnosis and a few were knownto the registry from death certificate only). Also excludedwere: 936 patients who had a previous malignancy; 447who underwent initial palliative radiotherapy, 410 patientswho died within 90 days of surgery; and 1773with a uterinesarcoma or other histology. The remaining 9411 patientscomprised the study population; all these cases had beenconfirmed microscopically and were known to the registrywhile they were alive. About 28% of the patients had theirinitial surgery at CGOCs and 72% had their surgery at othergeneral hospitals. Table 1 shows that the age distribution ofpatients differed very little between CGOCs and otherhospitals, but there was a slightly higher proportion ofpatients with high-risk histologies at the CGOCs. Table 1also shows that surgical staging of lymph nodes and peri-toneal biopsies were more frequently done at the CGOCsthan in other general hospitals. Table 1 also shows the stagedistribution of the subset of 505 cases seen at a cancercentre in 2003. Note that the stage distribution differs littlebetween patients who had their surgery at a CGOC andthose who had their surgery at other hospitals.

Factors Associated with the Use of Radiotherapy

The results of our analysis of factors associated with theuse of postoperative radiotherapy are described in Tables 2and 3. Table 2 provides the rates of use of radiotherapy forsubgroups defined by patient-related, disease-related andtreatment-related variables. Table 3 provides rates forgroups defined by health system-related variables. Thesetables also include the results of a fixed effect, multiplelogistic regression. The data were a good fit with the fixedeffects model according to the Hosmer and LemeshowGoodness-of-Fit Test (P ¼ 0.7932), and the Likelihood RatioTest (P < 0.0001) confirmed the predictive value of themodel. Multilevel modelling was also done to assess andaccount for the potential effects of data clustering byoperating hospital, but this approach did not alter our mainfindings or conclusions and, therefore, these data are notpresented.

Patient-related Factors

Just over one-quarter of all patients in the study pop-ulation received adjuvant radiotherapy (26.2%). Table 2shows that the use of radiation increased with increasingage up to the age of 80 years. The radiotherapy rateincreased significantly from 16.7% in patients under 40years to 30.4% in patients aged 71e80 years (P < 0.0001),and this trend remained significant in the multivariateanalysis. Table 2 shows that there was no discernible asso-ciation between the use of radiation and the medianhousehold income in the neighbourhood where the patientlived. About two-thirds of patients were classified as havingno comorbidity using the Elixhauser score. Table 2 showsthat therewas no significant decrease in the use of radiation

Table 1Characteristics of patients managed at comprehensive gynaecological oncology centres (CGOCs) and other hospitals

Characteristics Surgery at CGOC(n ¼ 2605)

Surgery at otherhospitals (n ¼ 6806)

All cases combined(n ¼ 9411)

Age (years)�60 42.7% 41.0% 41.5%

61e70 29.6% 30.7% 30.4%>70 27.7% 28.2% 28.1%

High-risk Histologyy (%)* 10.2% 4.0% 5.7%Surgical lymph node staging (%)* 30.1% 4.8% 11.8%Peritoneal biopsy (%)* 9.8% 6.5% 7.4%Radiation* 29.3% 24.9% 26.2%Stage (%) for subset of 505 patients withcomplete stage informationz

Surgery at CGOC(n ¼ 172)

Surgery at otherhospitals (n ¼ 333)

All cases combined(n ¼ 505)

IA 18.0% 12.3% 14.3%IB 34.9% 47.4% 43.2%IC 20.3% 22.2% 21.6%IIA 4.7% 5.1% 5.0%IIB 8.7% 6.6% 7.3%IIIA 5.8% 3.0% 4.0%IIIB 0.6% 0.9% 0.8%IIIC 5.2% 1.5% 2.8%IVA 1.2% 0.3% 0.6%IVB 0.6% 0.6% 0.6%

* P < 0.0001.y Serous or clear cell carcinoma.z FIGO 1988 stage distribution for subset of patients with complete stage information.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124 e117

with increasing levels of comorbidity in either the univar-iate or the multivariate analysis.

Disease-related Factors and Use of Radiation

Most patients in the study population (94.3%) wereclassified as having an adenocarcinoma, 4.1% had a serouscarcinoma and 1.5% had a clear cell carcinoma. Table 2shows that rates of use of radiotherapy in the clear cell(50.3%) and serous carcinomas (46.4%) were significantlyhigher than the rate of 24.9% observed in the adenocarci-nomas (P < 0.0001), and these differences remainedsignificant in the multivariate analysis. We were unable toreport on the use of radiotherapy by stage in the overallpopulation because this information was only available ona minority of patients (see above). However, in the stagedsubgroup of 505 cases, adjuvant radiotherapy rates were asfollows: stage IA, 1.4%; stage IB, 11.9%; stage IC, 68.2%; stageIIA, 84.0%; stage IIB, 83.8%; stage III, 91.9%.

Treatment-related Factors

Surgical staging of lymph nodes, as defined above, wascarried out in 11.8% of patients, but Figure 2 shows that ratesof surgical staging varied widely across the province. Theproportion of patients who were surgically staged in eachcounty ranged from 0 to 51.9%, with a median 7.9% and aninterquartile range of 4.7e16.6%. The rate of surgical stagingat CGOC hospitals (30.1%) was significantly higher than therate observed at other hospitals (4.8%). Table 2 shows thatpatients who had surgical staging of lymph nodes, those

who had a peritoneal biopsy, and those who had a radicalhysterectomy, were all significantly more likely than othersto have postoperative radiotherapy.

Health System-related Factors

Figure 3 illustrates that the proportion of patients whoreceived adjuvant radiation varied widely among hospitalsin Ontariowith amedian of 24.3% and an interquartile rangeof 17.2e30.7%. Rates of radiotherapy varied widely amongCGOCs, and also among non-CGOC hospitals.

Table 3 shows that the rate of use of radiotherapy amongpatients who had their surgery at a CGOC (29.3%) wassignificantly higher than among patients who had surgeryat other hospitals (25.0%). Table 3 shows that, when patient-related, disease-related and surgery-related factors wereincluded in a multivariate analysis together with otherhealth system-related factors, no association was foundbetween having surgery at a CGOC and the use of radio-therapy (odds ratio 1.00, 95% confidence limits 0.89e1.13).To explorewhy the apparent association between surgery ata CGOC and the use of radiotherapy disappeared in themultivariate analysis, we assessed the effect of removingother variables from the model. When the surgery-relatedvariables (type of hysterectomy, lymph node staging andperitoneal biopsy) were removed from the model, surgeryat a CGOC became borderline significant (odds ratio 1.12,95% confidence limits 1.00e1.26), and when histology wasalso removed, the association became stronger (odds ratio1.22, 95% confidence limits 1.09e1.36). We also carried outstratified analyses on cases that had surgical lymph node

Table 2Patient-related, disease-related and treatment-related factors affecting use of adjuvant radiation in Ontario

Patient factors Number Use of radiation (%) Use of radiation multivariatelogistic regression estimateodds ratio (95% confidence limit)*

Patient age (years)y�40 221 16.74 Reference41e50 893 19.60 1.23 (0.83, 1.83)51e60 2791 24.58 1.66 (1.15, 2.41)61e70 2863 27.84 1.92 (1.32, 2.77)71e80 2096 30.44 2.16 (1.49, 3.13)>80 547 23.40 1.46 (0.97, 2.21)Neighbourhood income quintile1 (Poorest) 1950 26.36 Reference2 2102 25.69 0.98 (0.85,1.13)3 1904 27.84 1.09 (0.94,1.26)4 1778 24.75 0.93 (0.80,1.09)5 (Richest) 1668 26.02 0.98 (0.83,1.15)Elixhauser comorbidity0 6587 25.84 0.96 (0.90,1.02)z1 1771 28.182 730 25.343 234 25.21�4 89 17.98

Disease- and treatment-related factors

HistologyyAdenocarcinoma 8876 24.86 ReferenceClear cell 145 50.34 2.71 (1.93, 3.80)Serous 390 46.41 2.19 (1.77, 2.71)Surgical stagingyNone 8303 25.04 ReferenceStaged 1108 34.48 1.45 (1.23, 1.70)Hysterectomy typeySimple hysterectomy 9249 25.89 ReferenceRadical hysterectomy 162 40.74 1.66 (1.18, 2.33)Peritoneal biopsyyNo 8715 25.44 ReferenceYes 696 35.06 1.48 (1.24, 1.76)

* Fixed effects multivariate logistic regression controlling for patient age, neighbourhood income quintile, Elixhauser comorbidity score,histology, surgical staging, hysterectomy type, peritoneal biopsy, hospital type, prevailing waiting time at the most responsible cancercentre, distance from residence to most responsible cancer centre, year of diagnosis and most responsible cancer centre.y P < 0.0001 in univariate analysis.z Elixhauser comorbidity was modelled as a continuous variable. The odds ratio reported corresponds to one unit increase in comorbidity.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124e118

staging and those that did not. Among the subgroup thathad lymph node staging, patients who had their surgery ata CGOC were more likely to receive radiation than thosewho had their surgery in other hospitals (37.0% versus28.3%, P < 0.01), and this association was significant in themultivariate logistic regression (odds ratio 1.53, 95% confi-dence limits 1.12e2.09). Among those who did not havesurgical lymph node staging there was no significantdifference in the use of radiation between patients who hadtheir surgery at a CGOC and those who had surgery in otherhospitals, in either the univariate or the multivariate anal-ysis (26.0% versus 24.8%, P ¼ 0.30, odds ratio 0.94, 95%confidence limits 0.82e1.07).

There was a small but significant inverse relationshipbetween the prevailing waiting time for radiotherapyand the chance of receiving postoperative radiotherapy

(Table 3). When surgery was carried out during prevailingwaiting times of �78 days, patients were less likely toreceive radiation after controlling for other factors inmultivariate logistic regression (odds ratio 0.853, 95%confidence limits 0.730e0.996).

The median distance from the patient’s home tothe most responsible radiotherapy centre was 26.9 km(interquartile range, 8.5e68.8 km). Table 3 shows thatlower rates of use of radiotherapy were observed amongpatients who lived further from a radiotherapy centre. Themagnitude of this association was minimal except in thesmall group who lived more than 180 km from a radio-therapy centre, of whom 18.8% received radiotherapy,compared with 27.2% of those who lived within 15 km oftheir radiotherapy centre. The association was barelystatistically significant in the multivariate analysis, but the

Table 3System-related factors affecting use of adjuvant radiation in Ontario

System factors Number Use of radiation (%) Use of radiation multivariatelogistic regression estimateodds ratio (95% confidence limit)*

Hospital typeyOther hospital 6806 24.95 ReferenceComprehensive gynaecological oncology centre 2605 29.29 1.00 (0.89, 1.13)Prevailing waiting time at the most responsible cancer centre (days)<78 2326 26.74 Reference�78 7085 25.96 0.853 (0.730, 0.996)Distance to most responsible cancer centre (km)0.0e15.0 3629 27.20 Reference16e60 3027 25.27 0.99 (0.88, 1.11)61e120 1723 26.70 0.85 (0.74, 0.99)121e180 599 25.21 0.76 (0.61, 0.94)>180 337 23.74 0.76 (0.56, 1.02)NA 96 18.75 0.61 (0.36, 1.06)Year of diagnosis1992 650 28.62 Reference1993 715 29.65 1.05 (0.83, 1.34)1994 709 26.80 0.91 (0.71, 1.16)1995 670 28.36 0.94 (0.73, 1.19)1996 766 24.67 0.76 (0.60, 0.97)1997 809 24.47 0.79 (0.62, 1.01)1998 778 25.84 0.89 (0.70, 1.14)1999 809 26.70 0.90 (0.71, 1.15)2000 884 23.30 0.78 (0.61, 1.00)2001 850 26.82 0.99 (0.78, 1.26)2002 880 24.32 0.79 (0.62, 1.01)2003 891 25.93 0.88 (0.69, 1.12)Most responsible cancer centreyA 2494 26.06 ReferenceB 845 27.57 0.89 (0.73, 1.09)C 1494 18.01 0.56 (0.47, 0.66)D 715 26.43 1.04 (0.83, 1.30)E 1416 34.25 1.51 (1.29, 1.76)F 1610 25.71 0.93 (0.79, 1.08)G 144 27.08 1.00 (0.67, 1.51)H 243 26.34 0.94 (0.69, 1.29)I 450 26.22 1.12 (0.85, 1.46)

* Fixed effects multivariate logistic regression controlling for patient age, neighbourhood income quintile, Elixhauser comorbidity score,histology, surgical staging, hysterectomy type, peritoneal biopsy, hospital type, prevailing waiting time at the most responsible cancercentre, distance from residence to most responsible cancer centre, year of diagnosis and most responsible cancer centre.y P < 0.0001 in univariate analysis.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124 e119

relationship was consistent over the whole range ofdistances studied, suggesting that this association is prob-ably real.

The proportion of patients who received radiation variedamong the catchment areas of the provincial cancer centres(median 26.3%). The rates observed in the catchment areasof most centres were close to 26%, but one centre hada significantly lower rate of 18.0%, and another centre hada significantly higher rate of 34.3%, and these differencesremained significant in the multivariate analysis (Table 3).Figure 4 illustrates geographical variations in the use ofradiotherapy at county level. The proportion of patientswho received radiotherapy varied widely from 13 to 40.3%with a median rate of 25.6% and an interquartile range of22.0e30.4%. Table 3 also shows that therewas no significant

change in the rate of use of radiotherapy over the 12 yearperiod of this study.

Survival

Figure 5 compares survival among residents of countieswith high, medium and low rates of use of radiotherapy.Figure 5A illustrates overall survival for the entire cohortand shows that there was no discernible difference insurvival among residents of counties with high, mediumand low rates of use of radiotherapy. Figure 5B shows that,in the overall cohort, there was no difference in cancercause-specific survival among residents of counties withhigh, medium and low rates of use of radiotherapy. Thesefindings are summarised in the top section of Table 4, which

Fig. 2. Geographical variations in the proportion of patients with endometrial cancer who had surgical staging of lymph nodes. The mapillustrates variations in the proportion of endometrial cancer patients who underwent surgical lymph node staging among Ontario counties.Counties were grouped into quintiles based on their rate of surgical staging. Circles indicate locations of radiotherapy centres.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124e120

also shows that there was no significant difference inoverall survival among these groups when they werecompared in a multivariate Cox analysis that controlled forother prognostic factors. Figure 5C and D provide similardata for the subgroup of patients with the higher risk, clearcell and serous histologies. For this subgroup of cases, both

Fig. 3. Variations in the use of adjuvant radiotherapy for endometrialcancer among Ontario hospitals. The dot plot shows the proportion ofpatients who received adjuvant radiotherapy at every Ontariohospital that operated on more than 60 cases of endometrial cancerover the study period. The bars show the 95% confidence limits.Comprehensive gynaecological oncology centres are shown as blackdots, and other hospitals are shown as grey dots.

overall survival and cancer cause-specific survival wereworse among residents of counties where the lowest ratesof use of radiotherapy were observed. Table 4 summarisesthese findings and also shows that overall survivalremained significantly worse in counties with a low rate ofuse of radiotherapy, when other factors associated withsurvival had been controlled for in a Cox analysis.A secondary analysis of the subgroup of 505 staged casesfrom the 2002e2003 period showed that there was nosignificant difference in stage distribution between theregions with low rates of radiotherapy and other regions ofthe province (P ¼ 0.43).

Discussion

In this study, we identified a number of health system-related factors associated with the use of adjuvant radia-tion for endometrial cancer in Ontario. For example,patients who had their surgery where longer radiotherapywaiting times prevailed, and those who lived farther froma radiotherapy centre, were less likely to receive radiation.Although our study was confined to Ontario, we suspectthat similar factors may affect the use of radiotherapy in thisdisease, particularly in other publicly funded health caresystems. The observed differences in the use of radio-therapymay be important becausewe also found that lowerrates of radiotherapy were significantly associated withlower survival in a subgroup of high-risk patients in whom

Fig. 4. Geographical variations in the proportion of patients with endometrial cancer who received adjuvant radiation. The map illustratesvariations in the proportion of patients with endometrial cancer who received adjuvant radiotherapy among Ontario counties. Counties weregrouped into quintiles based on their rate of use of adjuvant radiotherapy. Circles indicate locations of radiotherapy centres.

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124 e121

postoperative radiation might be expected to improvesurvival (serous and clear cell carcinoma).

This study shares the strengths and weaknesses of manyother registry-based, outcome studies. The OCR is an

Fig. 5. Comparison of survival among regions with higher and lower ratepatients with endometrial cancer who reside in regions of Ontario withtherapy. For the purpose of this analysis, counties were assigned to tertilescombined. (B) Cancer cause-specific survival for all cases combined. (C) Ovspecific survival for patients with high-risk histologies.

accurate source of basic diagnostic information aboutalmost every incident case of cancer in Ontario, and it alsoprovides reliable information about vital status and date ofdeath. The administrative databases of CIHI and CCO are

s of use of adjuvant radiotherapy. The graphs illustrate the survival ofhigh (:), medium (C), and low rates (-) of use of adjuvant radio-based on rates of use of radiotherapy. (A) Overall survival for all caseserall survival for patients with high-risk histologies. (D) Cancer cause-

Table 4Variations in survival by rates of radiation use in Ontario

Tertile (% radiation use) n 5-year overall survival(95% confidence limits)

Hazard ratio foroverall survival(95% confidencelimits)

5-year cancer-specificsurvival (95% confidencelimits)

Hazard ratio forcancer-specificsurvival (95%confidence limits)

Total cohort 9369 *

Low-use counties (<23.1%) 2399 86.7% (85.3%, 88.0%) 0.92 (0.82, 1.03) 90.4% (89.1%, 91.5%) 0.92 (0.78, 1.09)Medium-use counties(23.1e29.2%)

4522 87.4% (86.4%, 88.3%) 0.93 (0.84, 1.03) 90.9% (90.0%, 91.7%) 0.90 (0.78. 1.05)

High-use counties (>29.2%) 2448 85.8% (84.3%, 87.1%) Reference 89.4% (88.1%, 90.6%) Reference

High-risk histology 531 * *

Low-use counties (<40.0%) 121 44.6% (35.6%, 53.2%) 1.43 (1.06, 1.93) 46.6% (37.0%, 55.6%) 1.52 (1.09, 2.12)Medium-use counties(40.0e52.9%)

178 58.4% (50.8%, 65.3%) 1.10 (0.83, 1.44) 61.1% (53.4%, 67.9%) 1.12 (0.82, 1.53)

High-use counties (>52.9%) 232 59.5% (52.9%, 65.5%) Reference 63.5% (56.7%, 69.5%) Reference

For the total cohort, stepwise selection identified a Cox model controlling for age, neighbourhood income quintile, Elixhauser comorbidity,histology, hysterectomy type, peritoneal biopsy, prevailing waiting time and distance to the most responsible cancer centre. For high-riskhistologies, the Cox model identified by stepwise selection controlled for age, surgical staging and hysterectomy type.* Differences between groups were statistically significant (Wilcoxon test for equality of strata P < 0.05).

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124e122

a fairly accurate source of basic information about the use ofsurgery and radiotherapy in almost every patient. There-fore, the linkage of these province-wide databases allowedus to provide a simplified description of the treatment andoutcome of endometrial cancer in a large population. Ourstudy was therefore free of the problem of referral biasinherent to hospital-based case series, and had greatstatistical power to detect inter-regional differences intreatment and outcome. However, the absence of anyinformation about tumour grade and the absence ofcomplete information about stage made it impossible tocontrol for these important factors in our analyses oftreatment and outcome. Nonetheless, there is no a priorireason to suspect that there might be large inter-regionalvariations in the distributions of tumour grade or stage,and the limited evidence available suggests that thesecharacteristics vary little across the province [24]. It istherefore improbable that the variations in the use ofradiotherapy reported here are explained by variations incase mix. Lack of the details of the care path in the indi-vidual case also made it impossible to determine if a patientwas not treated with radiotherapy because she was notreferred to a radiation oncologist or because radiotherapywas not recommended by her radiation oncologist. Thus,we were unable to determine the extent to which gynae-cologists and radiation oncologists were responsible for theobserved variations in the use of radiotherapy across theprovince. Administrative databases provide no informationabout the patients’ preferences or their role in treatmentdecisions and we cannot comment directly on the extent towhich rates of use of radiotherapy are influenced bypatients’ wishes. Our reliance on administrative data alsoprecluded assessment of rates of pelvic failure, becausenone of the available databases provided any direct infor-mation about the date or site of recurrence. In reality, thisoutcome is not consistently evaluated or reliably reported inroutine practice, and is difficult even to measure in clinicaltrials. Whether and when a patient is found to have a pelvic

recurrence depends on the type and frequency of follow-upinvestigations. We doubt if it will ever be possible to ensuresufficiently consistent follow-up in the community at largeto permit pelvic failure rates to be reliably reported inpopulation-based studies.

Overall, we found that 26% of women who had a hyster-ectomy for endometrial cancer received adjuvant radiationtherapy. Delaney et al. [1] estimated, in a population witha surgical staging rate similar to Ontario’s, that 58% ofwomen with endometrial cancer will develop an indicationfor radiotherapy at some point in the course of the illness. Inestimating the appropriate rate of use of radiotherapy, theseinvestigators included indications for subsequent treatmentfor recurrence and metastasis, and excluded indications forbrachytherapy [1]. Thus, we cannot directly compare ourobserved rate with their estimate of the appropriate rate.However, given that relatively few patients with endome-trial cancer go on to develop recurrence or metastases, therate we have observed seems to fall far short of the Delaneyet al. [1] estimate of the appropriate rate. Moreover,although the rate of use of adjuvant radiotherapy variedwidely across Ontario, the observed rate did not approachthe estimated appropriate rate in any part of the province.

Nonetheless, we hesitate to conclude that radiotherapy isunder-utilised in Ontario, because the indications forradiotherapy in this disease remain controversial and theevidentiary basis of the evidence-based rate is not verystrong. An alternative approach to establishing the ‘right’rate of radiotherapy is to set a ‘benchmark’ based on therates of use of radiotherapy observed in circumstanceswhere access to radiotherapy is not constrained, and deci-sions about its use are made in a multidisciplinary setting[25]. However, we found that the rate of use of radiotherapyin this disease varies widely even among the CGOCs, wherepatients have ready access to radiotherapy and gynaeco-logical and radiation oncologists work closely with oneanother. In fact, the variation in rates observed in thesemore sophisticated hospitals was just as great as it was

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124 e123

among other hospitals. This makes it impossible to identifya robust benchmark rate for the use of radiotherapy inendometrial cancer, as we have done successfully in severalother diseases [25e27].

The results of our study suggest that longer prevailingwaiting times for adjuvant radiotherapy may be associatedwith lower rates of use of adjuvant radiation for endome-trial cancer. This may be due to lower rates of referral, or toless frequent prescription of radiation after referral whenprevailing waiting times are longer. Prevailing waitingtimes have been associated with variations in the use ofradiation in some, but not all studies. Zhang-Salomons et al.[28] found that elderly patients were found less likely toreceive post-lumpectomy radiation for breast cancer duringlonger prevailing waiting times for radiotherapy in Ontario.However, Huang [23] found no association between pre-vailing waiting times for radiotherapy and the choice ofradiation versus surgery for prostate cancer. Prevailingwaiting times may also affect the way radiotherapy isdelivered; Kong et al. [12] found that a longer prevailingwaiting time was associated with the use of shorter frac-tionation schemes for bone metastases.

Inkeepingwith the results of the relevant clinical trials,wefoundno association between the rate of use of postoperativeradiotherapy and survival in the overall study population.Regrettably, for reasons discussed above, we were unable todetermine whether lower rates of radiotherapy were asso-ciated with higher rates of pelvic recurrence, as predicted bythe results of clinical trials. However, in a subpopulation ofhigher risk patients with clear cell or serous histologies, weobserved worse survival among residents of regions withlower rates of use of radiotherapy. The association betweenlow radiotherapy use and poorer survival in this subgroupwas clinically and statistically significant in both the univar-iate and multivariate analyses. Although the stage distribu-tion observed in regionswith low rates of use of radiotherapywas similar to that observed in other regions of the province,it remains possible that the observed differences in survivalreflect differences in case mix rather than differences intreatment. The association between radiation use andsurvival might have been confounded by the use of adjuvantchemotherapy, butwe think this is unlikely. The periodunderstudy (1992e2003) was before the publication of thecommonly cited 2007 study by Hogberg et al. [29] that sug-gested a small benefit in disease-free survival with chemo-therapy inpatientswithhigh-risk endometrial cancer. Even ifchemotherapy had been widely used between 1992 and2003, it would have to be used much more in regions withhigh radiation use for any confounding to occur, and there isno reason to suspect this. Finally, if the evidence for an overallsurvival benefit of adjuvant chemotherapy for high-riskendometrial cancer is limited, it is even more so specificallyfor clear cell and serous carcinoma [30,31].

Conclusions

This study provides evidence that factors unrelated topatient’s needs influenced the use of adjuvant radiation for

endometrial cancer in Ontario between 1992 and 2003, andthat lower rates of use of radiation were associated withpoorer survival among a subset of the patients who wouldbe expected to benefit most from radiation.

Acknowledgements

The authors wish to acknowledge the support of theCanadian Institute of Health Research (CIHR) Fellowship inPolicy and Health Services Research, Queen’s University andCIHR Transdisciplinary Training Program in CancerResearch Fellowship, CIHR New Emerging Team (NET)Grant, Cancer Care Ontario (CCO) and the Ontario CancerRegistry (OCR).

References

[1] Delaney G, Jacob S, Barton M. Estimation of an optimal radio-therapy utilization rate for gynecologic carcinoma: part II ecarcinoma of the endometrium. Cancer 2004;101:682e692.

[2] Johnson N, Cornes P. Survival and recurrent disease after post-operative radiotherapy for early endometrial cancer: system-atic review and meta-analysis. BJOG 2007;114:1313e1320.

[3] Lee CM, Szabo A, Shrieve DC, et al. Frequency and effect ofadjuvant radiation therapy among women with stage Iendometrial adenocarcinoma. JAMA 2006;295:389e397.

[4] Blake P, Swart AM, Orton J, et al. Adjuvant external beamradiotherapy in the treatment of endometrial cancer (MRCASTEC and NCIC CTG EN.5 randomised trials): pooled trialresults, systematic review, and meta-analysis. Lancet 2009;373:137e146.

[5] Creutzberg CL, van Putten WL, Koper PC, et al. Surgery andpostoperative radiotherapy versus surgery alone for patientswith stage-1 endometrial carcinoma: multicentre randomisedtrial. PORTEC Study Group. Post Operative Radiation Therapyin Endometrial Carcinoma. Lancet 2000;355:1404e1411.

[6] Keys HM, Roberts JA, Brunetto VL, et al. A phase III trial ofsurgery with or without adjunctive external pelvic radiationtherapy in intermediate risk endometrial adenocarcinoma:a Gynecologic Oncology Group study. Gynecol Oncol2004;92:744e751.

[7] Nout RA, Smit VT, Putter H, et al. Vaginal brachytherapy versuspelvic external beam radiotherapy for patients with endo-metrial cancer of high-intermediate risk (PORTEC-2): anopen-label, non-inferiority, randomised trial. Lancet 2010;375:816e823.

[8] Lukka H, Chambers A, Fyles A, et al. Adjuvant radiotherapy inwomen with stage I endometrial cancer: a clinical practiceguideline 2006. Available at: https://www.cancercare.on.ca/common/pages/UserFile.aspx?serverId¼6&path¼/File%20Database/CCO%20Files/PEBC/pebc4-10f.pdf.

[9] Kwon JS, Carey MS, Cook EF, et al. Patterns of practice andoutcomes in intermediate- and high-risk stage I and IIendometrial cancer: a population-based study. Int J GynecolCancer 2007;17:433e440.

[10] Paszat LF, Mackillop WJ, Groome PA, et al. Radiotherapy forbreast cancer in Ontario: rate variation associated withregion, age and income. Clin Invest Med 1998;21:125e134.

[11] Huang J, Zhou S, Groome P, et al. Factors affecting the use ofpalliative radiotherapy in Ontario. J Clin Oncol 2001;19:137e144.

[12] Kong W, Zhang-Salomons J, Hanna TP, et al. A population-based study of the fractionation of palliative radiotherapy

T.P. Hanna et al. / Clinical Oncology 24 (2012) e113ee124e124

for bone metastasis in Ontario. Int J Radiat Oncol Biol Phys2007;69:1209e1217.

[13] Booth CM, Mackillop WJ. Translating new medical therapiesinto societal benefit: the role of population-based outcomestudies. JAMA 2008;300:2177e2179.

[14] Robles SC, Marrett LD, Clarke EA, et al. An application ofcaptureerecapture methods to the estimation of complete-ness of cancer registration. J Clin Epidemiol 1988;41:495e501.

[15] Clarke EA, Marrett LD, Kreiger N. Cancer registration inOntario: a computer approach. IARC Scientific Publications1991:246e257.

[16] Mackillop WJ, Zhang-Salomons J, Groome PA, et al. Socio-economic status and cancer survival in Ontario. J Clin Oncol1997;15:1680.

[17] Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measuresfor use with administrative data. Med Care 1998;36:8e27.

[18] Southern DA, Quan H, Ghali WA. Comparison of the Elix-hauser and Charlson/Deyo methods of comorbiditymeasurement in administrative data. Med Care 2004;42:355e360.

[19] Stukenborg GJ, Wagner DP, Connors Jr AF. Comparison of theperformance of two comorbidity measures, with and withoutinformation from prior hospitalizations. Med Care 2001;39:727e739.

[20] de Groot V, Beckerman H, Lankhorst GJ, et al. How to measurecomorbidity. A critical review of available methods. J ClinEpidemiol 2003;56:221e229.

[21] Ministry of Health and Long Term Care Finance & InformationManagement Branch Information Management Unit (editors).Ministry of Health and Long Term Care Master NumberingSystem (CIHI Institution Numbers). Ontario, 2006.

[22] Bardell T, Belliveau P, Kong W, et al. Waiting times for cancersurgery in Ontario: 1984e2000. Clin Oncol (R Coll Radiol)2006;18:401e409.

[23] Huang M. The influence of waiting time on treatment choicesamong prostate cancer patients. Queen’s University, Depart-ment of Community Health and Epidemiology, 2006: 120.

[24] Kwon JS, Carey MS, Cook EF, et al. Are there regional differ-ences in gynecologic cancer outcomes in the context ofa single-payer, publicly-funded health care system? Apopulation-based study. Can J Public Health 2008;99:221e226.

[25] Barbera L, Zhang-Salomons J, Huang J, Tyldesley S,Mackillop WJ. Defining the need for radiotherapy for lungcancer in the general population: a criterion-based, bench-marking approach. Medical Care 2003;41:1074e1085.

[26] Kerba M, Miao Q, Zhang-Salomons J, Mackillop WJ. Definingthe need for breast cancer radiotherapy in the general pop-ulation: a criterion-based benchmarking approach. Clin Oncol2007;19:481e489.

[27] Kerba M, Miao Q, Zhang-Salomons J, Mackillop W. Definingthe need for prostate cancer radiotherapy in the generalpopulation: a criterion-based benchmarking approach. ClinOncol 2010;22:801e809.

[28] Zhang-Salomons J, Fourodi F, Huang J, et al. Health systemeffects on the use of radiotherapy following surgery for breastcancer.Abstr AcadHealth ServResHealthPolicyMeet2002;19:45.

[29] Hogberg T, Rosenberg P, Kristensen G, et al. A randomizedphase-III study on adjuvant treatment with radiation (RT) �chemotherapy (CT) in early-stage high-risk endometrialcancer (NSGO-EC-9501/EORTC 55991). J Clin Oncol 2007;25:5503.

[30] Hogberg T, Rosenberg P, Kristensen G, et al. A randomizedphase-III study on adjuvant treatment with radiation (RT) �chemotherapy (CT) in early-stage high-risk endometrialcancer (NSGO-EC-9501/EORTC 55991). Online oral presenta-tion. Available at: http://www.asco.org/ASCOv2/MultiMedia/VirtualþMeeting?&vmview¼vm_session_presentations_view&confID¼47&sessionID¼2213.

[31] Randall ME, Filiaci VL, Muss H, et al. Randomized phase IIItrial of whole-abdominal irradiation versus doxorubicin andcisplatin chemotherapy in advanced endometrial carcinoma:a Gynecologic Oncology Group Study. J Clin Oncol 2006;24:36e44.