IN MANY cases of supraglottic cancer, conservative treatment

Transcription

1 Management and Outcome Differences in Supraglottic Cancer Between Ontario, Canada, and the Surveillance, Epidemiology, and End Results Areas of the United States By Patti A. Groome, Brian O Sullivan, Jonathan C. Irish, Deanna M. Rothwell, Karleen Schulze, Padraig R. Warde, Ken M. Schneider, Robert G. Mackenzie, D. Ian Hodson, J. Alex Hammond, Sunil P.P. Gulavita, Libni J. Eapen, Peter F. Dixon, Randy J. Bissett, and William J. Mackillop Purpose: We compared the management and outcome of supraglottic cancer in Ontario, Canada, with that in the Surveillance, Epidemiology, and End Results (SEER) Program areas in the United States. Methods: Electronic, clinical, and hospital data were linked to cancer registry data and supplemented by chart review where necessary. Stage-stratified analyses compared initial treatment and survival in the SEER areas (n 1,643) with a random sample from Ontario (n 265). We also compared laryngectomy rates at 3 years in those patients 65 years and older at diagnosis. Results: Radical surgery was more commonly used in SEER, with absolute differences increasing with increasing stage: I/II, 17%; III, 36%; and IV, 45%. The 5-year survival rates were 74% in Ontario and 56% in SEER for stage I/II disease (P.01), 55.7% in Ontario and 46.8% in SEER for stage III disease (P.40), and 28.5% in Ontario and 29.1% in SEER for stage IV disease (P.28). Cancer-specific survival results mirrored the overall survival results with the exception of stage IV disease, for which 34.6% of Ontario From the Radiation Oncology Research Unit, Departments of Oncology and Community Health and Epidemiology, Queen s University, Kingston; Departments of Radiation Oncology and Surgical Oncology, Princess Margaret Hospital, University of Toronto, Toronto; and Departments of Radiation Oncology at the Regional Cancer Centres of Cancer Care Ontario, Ontario, Canada. Submitted October 22, 2001; accepted October 16, Supported by the National Cancer Institute of Canada with funds from the Canadian Cancer Society. P.A.G. is an Ontario Ministry of Health Career Scientist. Address reprint requests to P.A. Groome, PhD, Radiation Oncology Research Unit, Apps Level 4, Kingston General Hospital, Kingston, Ontario K7L 2V7, Canada; patti.groome@krcc.on.ca by American Society of Clinical Oncology X/03/ /$20.00 patients survived their cancer compared with 38.1% in SEER (P.10). This stage IV difference was more pronounced when we further controlled for possible cause of death errors by restricting the comparison to patients with a single primary cancer (P.01). Three-year actuarial laryngectomy rates differed. In stage I/II, these rates were 3% in Ontario compared with 35% in SEER (P < 10 3 ). In stage III disease, the rates were 30% and 54%, respectively (P.03), and in stage IV disease they were 33% and 64% (P.002). Conclusion: There are large differences in the management of supraglottic cancer between the SEER areas of the United States and Ontario. Long-term larynx retention was higher in Ontario, where radiotherapy is widely regarded as the treatment of choice and surgery is reserved for salvage. In stages I to III, survival was similar in the two regions despite the differences in treatment policy. In stage IV, there may be a small survival advantage in the U.S. SEER areas related to the higher use of primary surgery. J Clin Oncol 21: by American Society of Clinical Oncology. IN MANY cases of supraglottic cancer, conservative treatment can preserve both life and natural speech. In other cases, however, many surgical methods intended to save life require removal of the larynx, with loss of natural voice and cosmetic and functional changes in the neck. This surgery may lead to difficulties in communication, loss of employment, and psychological problems. 1-6 The first decision in any individual curable case is, therefore, whether the larynx can be preserved without compromising survival. When radical radiotherapy is offered for treatment of supraglottic cancer, it is often performed with the understanding that total laryngectomy may be required subsequently for persistent or recurrent disease. Those who advocate this approach do so in the belief that more larynges are being preserved with no compromise to survival. 7,8 Those who advocate the use of initial total laryngectomy do so in the belief that it will increase survival. 9,10 Randomized trials comparing radiotherapy with surgery have not been conducted because of these entrenched beliefs In supraglottic cancer, current practice is based on reports from individual institutions recommending differing treatment policies and on the results of the two trials 14,15 that compared surgical treatment to induction chemotherapy with radiotherapy for advanced disease. With regard to the radiotherapy-surgery debate, these trial results have been difficult to interpret because radiotherapy was only administered if a favorable response to chemotherapy was obtained. 12,14-16 Survival equivalence was demonstrated in the Veterans Affairs study, 14 but in the Groupe d Etude des Tumeurs de la Tête et du Cou trial, survival was worse in the chemotherapy arm, with 69% of patients still alive at 2 years compared with 84% of patients in the surgery arm. 15 With the use of original data, a meta-analysis combined the results of these trials with one other trial in pyriform sinus cancer 17 and showed a trend toward improved survival with surgery that was not statistically significant (relative risk, 1.19; P.10). 13 The role of chemotherapy in advanced laryngeal cancer is being studied in a National Cancer Institute (NCI) high-priority trial (RTOG 91 11) 18 and two European Organization for Research and Treatment of Cancer (EORTC) trials (EORTC and 24954). Recent results from the RTOG 496 Journal of Clinical Oncology, Vol 21, No 3 (February 1), 2003: pp DOI: /JCO

2 SUPRAGLOTTIC CANCER IN CANADA AND THE UNITED STATES 497 Table 1. Data Sources for Variables Used in Study Study Variable Ontario United States Identify cases Ontario Cancer Registry SEER cancer registries Stage Chart review SEER cancer registries Initial treatment Surgery Hospital discharge and chart review data SEER cancer registries Radiotherapy Cancer clinic and chart review data SEER cancer registries Survival Ontario Cancer Registry SEER cancer registries Subsequent surgery (65 years of age and older cohorts) Hospital discharge and chart review data Medicare hospital discharge data trial indicate no evidence of a survival advantage for induction chemotherapy over radiation alone, although local control was improved in the concurrent arm. 18 Also, a randomized trial is being conducted in Singapore that will compare surgery and adjuvant radiotherapy with concurrent chemoradiotherapy in patients with resectable stage III or IV head and neck cancers (NMRC-SHN01, EV-97015). 19 We previously addressed the similar controversy in glottic cancer by doing a population-based comparison of the management and outcome of this disease in Ontario, Canada, and the United States Surveillance Epidemiology and End Results (SEER) registry areas. 20,21 In that study, we found that the policy of primary radiotherapy, reserving surgery for salvage, was associated with a higher rate of retention of the larynx, and it was not associated with any detriment in survival. In the current study, we repeated the same design in supraglottic cancer. When complete registration of incident cases is available, and case mix can be assessed, it is possible to take advantage of the natural experiments that occur when health care policies vary. By grouping patients with differing patterns of care by some variable not associated directly with treatment selection or outcome (in this case, geographic region), we can study the effect of that difference in care on outcome. The study of such natural experiments is a well-known epidemiologic approach to the assessment of community-level interventions, and similar thinking underlies the instrumental variable approach to outcomes research, a term that has its origins in economics. 22,23 Also, because the approach is population based, this study design is free of the referral biases that plague reports of institutional experience. Study Design METHODS This is a retrospective, population-based cohort study comparing the management and outcome of supraglottic cancer in Ontario and the SEER registry areas in the United States. We compared initial treatment and overall and cancer-specific survival between Ontario and SEER areas stratified by disease stage. We also compared the ultimate laryngectomy rates in those patients 65 years and older at diagnosis (the age restriction was because of data availability, as described below). To address the study objectives, we needed data about the diagnosis, stage of disease, initial treatment with radiotherapy and surgery, and follow-up information with respect to survival and the occurrence of subsequent laryngectomies. The details of the study design were dictated by the differences in the availability of these data in Ontario and the United States. The sources for the study variables are summarized in Table 1. In both regions, information about the diagnosis and vital status was available in the cancer registries. Stage of the cancer was available in the SEER registries but was not available in the Ontario Cancer Registry. We, therefore, conducted a chart review on a sample of cases in Ontario to match the staging information that was available in SEER. Information about initial treatment was already available in the SEER registries, and we were able to similarly classify Ontario cases using data collected from the chart review enhanced by electronic records. In Ontario, information on subsequent laryngectomies also was available to us from the chart review and electronic records. For the SEER patients, we obtained information on subsequent laryngectomies from the Linked Medicare-Tumor Registry Database at the U.S. National Cancer Institute (NCI) for patients 65 years of age and older at diagnosis. These data enabled us to make a comparison of ultimate laryngectomy rates between Ontario and the United States for patients 65 years of age and older. Study Population Our target population consisted of all case patients with the primary cancer diagnosis of International Classification of Diseases (ICD9) code (supraglottic carcinoma) identified in the registries. Those patients with non-squamous cell histologies were excluded. Those patients with a clinical diagnosis (no histologic information) or nonspecific histologies were retained on the assumption that most of these histologies would have proved to be squamous cell histologies. There were only a few such patients. The patients from Ontario were diagnosed from 1988 through The patients from the SEER registries were diagnosed from 1988 through Patients were followed to the end of 1996 in Ontario and to the end of 1995 in SEER. The Ontario study population is a subset from a larger Ontario Patterns of Practice study that collected data on both glottic and supraglottic patients diagnosed in Ontario from 1982 through The population for this report constitutes a stage-stratified, random sample of 265 supraglottic cancer patients diagnosed from 1988 through These years were chosen to more closely match the years for which SEER data were available. The study sampling for the larger study was done in two phases. First, we ran a simple random sample of 900 glottic and supraglottic cases picked from all such cases diagnosed in the registry between 1982 and There were 158 patients from this larger sample who had supraglottic cancer and who were diagnosed in the appropriate years. Data on these 158 subjects were used in the current report to describe the case mix characteristics of the Ontario supraglottic cancer population. In a second phase of the larger study, we collected data on more patients in the stage groups that were sparsely populated in the random sample to improve study power. We targeted 115 patients in each of the eight groups defined by subsite (glottic or supraglottic) and T category (T1 to T4). This was the maximal number that we deemed feasible given our study resources and the size of the patient pool for some of the subgroups. With the exception of T4 glottic, we met this target in all of the subsite T-category subgroups. As stage was not available in the Ontario registry, we obtained the stage-stratified sample by using the information from the simple random sample to set the proportion of patients from each subsite-stage group seen at each cancer center. We then set a quota for the number of patients in each subsite-stage group to be obtained at each center. We put the list of glottic and supraglottic cancer patient charts in random order at each center, and the data abstractor continued to abstract until the subsite-stage quotas were reached. After restricting for subsite and diagnosis year, this second phase added 107 subjects to the study population used in this report. In SEER, we kept all supraglottic cases who had sufficient data from the target population. Those SEER cases with unknown T or N stage (when applicable for stage grouping assignment) were excluded. This meant that 31.6% (n 759) of SEER patients are not included in the stage-stratified results. Only 0.8% (n 2) are missing from the Ontario sample for this

3 498 GROOME ET AL reason. The potential effect of the missing stage data in SEER was mitigated by our decision to run stage-stratified analyses. We also assessed the possible effect of missing stage in SEER by comparing the characteristics and outcome of staged and unstaged cases. To compare the ultimate laryngectomy rates, we were restricted to studying patients who were 65 years of age and older at diagnosis, because the electronic follow-up data used to identify subsequent laryngectomies in the SEER population were only available from Medicare. We included only those SEER patients who had continuous Medicare Part A coverage (inpatient hospitalization and skilled nursing facility coverage) for at least 2 years after diagnosis and who were not enrolled in a health maintenance organization (HMO). We made the HMO exclusion because Medicare data from HMOs are incomplete. 24 In the end, our study group included 82.6% of those supraglottic cancer patients in the SEER study population from that age category. Data Sources The U.S. SEER program registries are active registries, employing data abstractors to collect information about all cancer cases diagnosed in nine areas of the United States, comprising approximately 9.5% of the total U.S. population 25. The Ontario registry is a passive one, collecting and linking existing sources of data to form the case record. 26 Ontario is the largest province in Canada, comprising 11.4 million people or 38% of the Canadian population. 27 The quality of the Ontario registry is comparable to SEER with respect to case ascertainment and vital status In addition to the data contained within the SEER public use database, 31 we used U.S. Health Care Financing Administration (HCFA) Medicare hospital discharge data for the SEER laryngeal cancer patients aged 65 and older who were diagnosed from 1988 through This Linked Medicare- Tumor Registry linked Medicare data to 94% of SEER eligible cases, and unique identifiers were assigned at the U.S. NCI. 24 These hospitalization records provided information about subsequent laryngectomies. We obtained Ontario hospital separation records for all records in which cancer was mentioned from the date of the cancer diagnosis to the end of These data are sent to the Ontario registry by the Canadian Institute for Health Information and are linked by the registry to their case records using the Generalized Iterative Record Linkage System. 26 These hospitalization records provided information about initial and subsequent surgeries. We conducted a retrospective chart review in Ontario that provided information about disease stage and initial treatment. Here, we report on the 265 supraglottic cancer cases from that sample. The Ontario Cancer Registry provided all patients cancer clinic and hospital chart numbers associated with the registry case identification number. The nine cancer centers, which provide radiotherapy exclusively for the province, together saw 95.4% of the supraglottic cancer cases, and these charts served as our main source of data. Hospital charts were accessed for those patients never seen in a cancer center or where necessary, to obtain up to 5 years of follow-up information on each case. Two data abstractors with previous chart abstraction experience and medical training (a registered nurse and a health records administrator) abstracted the data and worked with a study collaborator at each cancer center to assign stage and verify critical information. Bias Assessments Our results may have been biased by differences in case mix within the stage groups. We compared the study populations by age and sex. We addressed the role of possible differences in comorbidity on our survival comparisons by considering cancer-specific survival separately from comorbidity-specific survival. 32 We also calculated relative survival, which accounts for differences in general population life expectancy. 33 To assess possible biases as a result of the data restrictions, patient and treatment characteristics were compared. We compared the SEER patients with stage information with those without stage information. We compared the SEER patients 65 years of age and older with complete Medicare data with those without complete Medicare data. We also compared the actuarial laryngectomy rates between the Ontario patients aged younger than 65 years with those aged older than 65 years to assess whether the laryngectomy findings could also be applied to younger patients. Assignment of Stage Following the SEER protocol for assigning extent of disease, 34 stage assignment in Ontario was based on pathologic stage when available and clinical stage otherwise. SEER extent of disease codes were translated to T categories; this process was straightforward for all but the SEER extent of disease code 60 because it includes aspects of both T3 and T4 disease: extension to pre-epiglottic tissues, postcricoid area, pyriform sinus, hypopharynx not otherwise specified, vallecula, or base of tongue. In this instance, we used the American Joint Committee on Cancer/International Union Against Cancer (AJCC/UICC) rule to downstage when stage is uncertain. 35,36 This group represents 20.6% of all of the staged cases in SEER and 84% of the SEER stage III cases. In Ontario, stage was assigned by study collaborators who are oncologists treating this disease. They used the AJCC fourth edition summary tumor-node-metastasis form 35 and consulted the chart and verbatim transcriptions, describing the extent of disease, recorded by the data abstractors. We were concerned that the SEER assignment of extent of disease might differ from the data we were able to abstract in Ontario because SEER reports pathologic stage when available, 25,34 and pathologic stage in the U.S. data would be more common because of the expected higher use of surgery. In the Discussion, we explore the possible role that differential stage assignment may have played in our results. Assignment of Initial Treatment Initial treatment was available directly from the SEER database. The SEER registry defines initial treatment as all cancer-directed treatment documented as the planned first course of treatment. 34 In Ontario, we collected initial treatment information in our chart review, using the same definition as SEER. Treatment categories include radiotherapy alone, local excisions and partial laryngectomies with and without radiotherapy, neck dissection with radiotherapy, total and radical laryngectomies with and without radiotherapy, and no initial treatment identified. We could not address the role of chemotherapy in this study because it was rarely used as part of initial treatment during this era. Assignment of Subsequent Laryngectomy For SEER, we identified total laryngectomies in the linked Medicare hospital discharge data that occurred more than 4 months after diagnosis in those patients who did not have a laryngectomy as initial treatment. For the analogous 65 years and older group in Ontario, we used abstracted data on subsequent laryngectomies from the charts that were enhanced by the hospitalization data to ensure complete ascertainment. Because the Medicare data source (electronic hospital discharge summaries) was not as comprehensive as the Ontario source (chart review plus hospital discharge data), we calculated error rates for the detection of subsequent laryngectomies in the Medicare data. To do this, we used the initial treatment information in SEER as a standard because these data were collected similarly to our Ontario data: directly from charts by cancer registrars. We compared the number of initial laryngectomies found in the SEER registry data with those found in the Medicare data. The resulting Medicare initial laryngectomy assignment error rate is a surrogate for the subsequent laryngectomy assignment error rate. Conversely, we also took the opportunity to check the SEER initial treatment assignment error rate by comparing it with the Medicare data. There were 779 supraglottic cancer patients from SEER who were 65 years or older at diagnosis and who had Medicare coverage in the period of interest after diagnosis. Of these, 255 patients had a total or radical laryngectomy in the initial treatment period according to one or both of the data sources. The SEER registry data identified 236, for an error rate of 8%. The Medicare hospital discharge data identified 179, for an error rate of 43%. We applied these error rates to the initial and subsequent laryngectomy rates by multiplying the number of surgeries by one plus the error rate. Because the initial laryngectomy information came from the SEER registry data, we corrected this rate using the SEER error rate of 8%. Because the information about subsequent laryngectomies came from the Medicare data, we corrected this rate using the Medicare error rate of 43%. We also report the uncorrected results.

4 SUPRAGLOTTIC CANCER IN CANADA AND THE UNITED STATES 499 Table 2. Patient Characteristics by Region Table 3. Stage Distributions by Region, % Ontario (n 265) SEER (n 1643) P Ontario (n 158)* SEER (n 1643) p Stage I Mean age, years (SD) 62.6 (8.6) 63.4 (10.0) 0.64 % male Stage II Mean age, years (SD) 60.5 (9.5) 63.0 (10.3) 0.55 % male Stage III Mean age, years (SD) 64.2 (10.4) 62.4 (10.6) 0.23 % male Stage IV Mean age, years (SD) 63.3 (10.4) 62.4 (10.6) 0.23 % male Follow-Up and Vital Status Assignment We determined survival and cause of death directly from the registry data in both regions. Vital status obtained during chart review was compared with that in the Ontario Cancer Registry, and six additional deaths were added to the data set. To diminish differences in cause of death assignment, we grouped all cancer deaths together to report cancer-specific survival rather than cause-specific survival. We tested the stability of our findings with this strategy by also comparing cancer-specific survival among those who had a single primary cancer only. Statistical Analysis The age, sex, and stage distributions of the Ontario and SEER study populations were compared, as were variations in the initial treatment distributions. For categorical comparisons, 2 tests were computed, and t tests compared mean age. All of our main analyses were stratified by the UICC/AJCC stage groups: I/II, III, and IV. We grouped stages I and II because of small numbers in the Ontario sample. We measured the effect of differences in initial treatment practice by comparing the outcomes experienced by the patients in Ontario with those experienced by the patients in SEER. Actuarial survival curves were statistically compared using the Wilcoxon statistic. We chose the actuarial method rather than the Kaplan- Meier method because SEER only provides follow-up in months. We also ran Cox proportional hazards models to control for small differences in the age and sex distributions between the regions. The actuarial laryngectomy rates (censoring for death) at 3 years were compared by calculating a z statistic on the difference. Relative survival is the ratio of the observed survival to the expected. To calculate expected survival, we used Ontario and U.S. general population life expectancy tables and assigned a life expectancy for each patient on the basis of their age and sex in Ontario and on their age, sex, and race in SEER. The relative survival rates at 5 years were compared by calculating a z statistic on the difference in rates. Corrections for missing laryngectomy data were made using a life-table that was created in a spreadsheet from the output of the actuarial analysis. This spreadsheet included columns for the number of events in each month, the number censored, and the effective sample size. The number of events was increased using the computed correction factor (one plus the error rate). The effective sample size was then reduced by the number of extra events, and the standard error for each month was calculated from the adjusted numbers. RESULTS The median follow-up was 31.5 months in Ontario and 27 months in SEER. Referring to Table 2, the mean age was not statistically significantly different between the two regions within any of the stage groups. The percentage of males was 16% and 14% higher in Ontario for stage II and III cases (P.04), respectively. Table 3 presents the disease stage distributions in the two regions. The Ontario results are for the random subset Stage group I II III IV T stage TX T T T T N stage N0/X N N N N M stage M0/X M *This reduced number constitutes the simple random sample subset of the study population. of the study population. In SEER, more patients had stage I disease and fewer had stage IV disease than in Ontario (overall P.007). The differences were largely the result of a higher proportion of T1 cases in SEER, more T4 cases in Ontario, and more N-positive cases in Ontario (44.6% of cases in Ontario compared with 35.6% of cases in SEER). The proportion of study subjects diagnosed in each year was similar between the two regions, ranging from 10% to 15% in Ontario and from 13% to 15.5% in SEER (P.71). Initial Treatment Table 4 presents the comparison of the initial treatment distributions. Stage I and stage II results are combined because of small numbers in the Ontario sample. In stage I/II disease, the use of radical surgery was much higher in SEER, with 20.7% having total laryngectomy compared with 3.6% in Ontario. This pattern persisted across all three stage groups, with the gap in treatment getting wider with advancing stage so that 56.0% of SEER stage IV patients had a laryngectomy as initial treatment compared with 10.7% in Ontario. Also, initial combined radiation with total or radical laryngectomy was used more often in SEER, especially for advanced disease. For those treated conservatively, radiotherapy was the most common treatment used, but there was also more use of partial laryngectomy in SEER, with absolute differences of 13% in stage I/II, 11% in stage III, and 7% in stage IV. Outcome Comparisons Figure 1 reports actuarial overall survival curves by stage group. Ontario patients with stage I/II disease experienced better survival that was statistically significant, with 74.3% surviving to 5 years, compared with 55.8% in SEER (Wilcoxon P.01). The survival rates for those with stage III and stage IV disease were not statistically different, with 55.7% of stage III Ontario

5 500 GROOME ET AL Table 4. Initial Treatment by Stage Grouping and Region, % I/II III IV Initial Treatment Ontario (n 83) SEER (n 623) Ontario (n 60) SEER (n 323) Ontario (n 122) SEER (n 697) Radical surgery XRT Total laryngectomy Total laryngectomy XRT Radical laryngectomy Radical laryngectomy XRT Conservative treatment Radiotherapy Local excision Local excision then XRT Partial laryngectomy Partial laryngectomy XRT Neck dissection XRT* Other surgery NOS XRT No treatment identified P Abbreviations: SEER, Surveillance, Epeidemiology, and End Results; XRT, radiotherapy; NOS, not otherwise specified. *All had XRT except one stage IV Ontario patient and four stage IV SEER patients. P values reported are for comparisons of major treatment categories (radical surgery, conservative treatment, and no treatment identified), with the other surgery NOS excluded from the analysis because of an absence of subjects in this category in Ontario, Canada. patients surviving to 5 years compared with 46.8% of SEER patients (P.40), and 28.5% stage IV Ontario patients compared with 29.1% in SEER (P.28). The Cox proportional hazards models, run for each stage group separately and controlling for age and sex, yielded the same conclusions. The adjusted relative risk of death in the SEER group was 1.86 (95% Fig 1. Actuarial survival by stage groupings. Ontario, solid line; SEER, dashed line. confidence interval [CI], 1.14 to 3.00) for stage I/II, 1.26 (95% CI, 0.82 to 1.95) for stage III, and 0.96 (95% CI, 0.75 to 1.21) for stage IV. When we accounted for general population life expectancy by calculating relative survival, the direction and magnitude of the differences between the two regions remained the same (results not shown). Figure 2 presents the cancer-specific survival. The same pattern is seen in stage I/II and stage III as was seen in the overall survival results. In stage I/II, Ontario patients experienced fewer cancer deaths, with a 5-year rate of 84.7% compared with 71.9% in SEER (P.05). The difference in the rates for stage III were not statistically significant (P.86). In stage IV, the SEER patients did slightly better, with a rate of 38.1% compared with 34.6% in Ontario, but the difference was only marginally statistically significant (P.10). The Cox proportional hazards models, run for each stage group separately and controlling for age and sex, yielded the same conclusions. The adjusted relative risk of death in the SEER group was 1.89 (95% CI, 0.99 to 3.62) for stage I/II, 1.11 (95% CI, 0.67 to 1.85) for stage III, and 0.89 (95% CI, 0.69 to 1.16) for stage IV. To minimize differences in death registration, we used any cancer death as an event rather than larynx cancer death. We tested the stability of our findings by looking at the results in those patients who only had one primary cancer registered (results not shown). The results follow the same pattern, and the stage IV, 5-year, cancer-specific survival difference reached statistical significance with a rate of 33.0% in Ontario and 41.8% in SEER (P.01). Figure 3 presents survival from noncancer causes of death. Across all three stage groups, the patients in the two regions did not experience statistically significant different rates of death from other causes, although the SEER patients had more observed events in all stage groups. Controlling for small difference in age and sex in Cox proportional hazards models did not alter these findings. Referring to Fig 4, for stage I/II disease, 3% of Ontario patients had their larynges removed by 3 years postdiagnosis

6 SUPRAGLOTTIC CANCER IN CANADA AND THE UNITED STATES 501 Fig 2. Actuarial survival from cancer causes of death by stage groupings. Ontario, solid line; SEER, dashed line. Fig 3. Actuarial survival from noncancer causes of death by stage groupings. Ontario, solid line; SEER, dashed line. compared with 35% (corrected rate) in SEER-Medicare (P.0001). In stage III disease, 30% of Ontario patients had a laryngectomy by 3 years postdiagnosis compared with 54% (corrected rate) of SEER-Medicare patients (P.03). For stage IV disease, 33% of Ontario patients had a laryngectomy by 3 years postdiagnosis compared with 64% (corrected rate) of SEER-Medicare patients (P.002). In all but stage I/II disease, the initial difference in the use of laryngectomy is reduced because of a higher rate of subsequent laryngectomy in Ontario, but the remaining differences are still clinically important. The observed, uncorrected laryngectomy rate curves also are presented, and the SEER curves remain higher than the Ontario curves over all subgroups. Bias Assessments To assess the representativeness of our study sample, we compared the characteristics of the random subset of the Ontario study population with the Ontario target population of all patients in the registry for the years 1988 through We compared them by age, sex, treatment, survival, and laryngectomy rates. To ensure comparable data, we used the electronic information only. The mean age was 63 years in both groups; the sex distributions differed by 4%, but this difference was not statistically significant (P.30). The initial treatment distributions were marginally significantly different (P.15), with a 4.5% lower rate of no treatment identified and a 3.9% lower rate of laryngectomy in the study group. The survival rate at 5 years was higher in the study population: 49.1% compared with 37.3% in the rest, but the difference did not approach statistical significance (P.29). At 3 years postdiagnosis, the probability of laryngectomy was 20.3% in the study group and 31.0% in the rest (P.008), but these rates began to converge so that by 5 years postdiagnosis, the probability of laryngectomy was 29.6% in the study group and 33.3% in the rest (P.48). We had to exclude 32% of the SEER study population because of missing stage. We compared those with stage in SEER with those without by age, sex, treatment, and survival. The age and sex distributions in the disease stage absent group were not significantly or clinically different from the disease stage present group, the absent group being 0.7 years younger on average and containing 1% fewer males. There were differences in the treatment distributions (P.001). Those with staging data were more likely to have had treatment, with 6.6% having no treatment recorded versus 12.5% in those with missing stage. Also, more initial laryngectomies occurred in the staged group: 40.5% compared with 21.9% in those without stage. Survival was higher for those with stage information: 42.9% compared with 38.8% (P.009). The study subcohort aged 65 years and older from SEER did not include 17% of the patients in the relevant diagnosis year/age group because of incomplete Medicare coverage. We compared the 102 case patients aged 65 years and older with staging

7 502 GROOME ET AL Fig 4. Actuarial laryngectomy rate by stage groupings for > 65 study cohorts. Ontario, solid line; SEER-Medicare, dashed line. Thin lines are observed rates; thick lines are main results that were adjusted for missing data when necessary. Lines are truncated when the sample size drops below five. information but incomplete Medicare data with the 65 years of age and older staged study subcohort by age, sex, stage, treatment, and survival. Sex, age, and treatment were not statistically or clinically different. In the subgroup of patients 65 years or older, there were 4% fewer males and 4.6% more initial laryngectomies, and the mean age was 71 and 72 years. The excluded group had 6.4% lower survival at 5 years, which was not statistically significant (P.58). The stage distributions were marginally significantly different (P.09), with more cases in stage III in the excluded group (26.5% compared with 17.2%). Last, because our actuarial laryngectomy comparisons were restricted to the subcohorts of patients 65 years of age and older, we compared the probability of laryngectomy over time in Ontario (where the data were available for all cases) in those younger than 65 years of age at diagnosis with those older than 65 years of age at diagnosis. The laryngectomy rates were slightly higher (2.8% at 3 years) in those younger than 65 years of age, but they were not statistically significantly different (P.52). DISCUSSION We observed strikingly different practice patterns between the U.S. SEER areas and Ontario, a finding that repeats the selfreported practice that was documented in our international survey, 37 and in our previous observational work in glottic cancer. 20,21 The difference in the rates of use of laryngectomy as initial treatment was statistically significant and clinically important for each stage group. These large practice variations repeat a pattern seen in other clinical settings when the level of evidence is weak What was particularly useful in our context is the difference of opinion represented in the U.S. and Canadian medical cultures regarding organ preservation in this disease, because it allowed us to take advantage of the resulting natural experiment to learn about treatment effectiveness. That is, we looked at the effect of the U.S. and Canadian policies by comparing the outcomes experienced in the whole population of patients with the disease in the two regions. This design is in contrast to that of the randomized clinical trial, in which outcomes would be compared between those who were treated with radiotherapy and those who were treated with surgery. The approach has been recently used in another cancer study to test the generalizability of randomized control trial results in an elderly lung cancer population. 44 Our results indicate that stage I/II and stage III Ontario patients are not experiencing a survival disadvantage from the predominant use of radiotherapy, with surgery reserved for salvage. Also, the ultimate rate of laryngectomy was lower in Ontario, supporting the contention that these patients are more likely to retain their natural voice. In stage IV supraglottic cancers, our data indicate that there may be a small cancerspecific survival advantage when total laryngectomy is the dominant treatment choice. Our results in stages I to III agree with the findings from the induction chemotherapy trials. The trial results have led some to the conclusion that initial laryngectomy is no longer the standard of care, 17 and subsequent trials, such as the recent report on RTOG 91 11, 18 are refining the role of chemotherapy without the use of a surgery-only arm. Even though our results are population based, our study populations may not be equally representative of the wider supraglottic cancer target population for two reasons: first, the quality of case ascertainment in the registries may have varied, and second, SEER does not represent a random sample of U.S. cases. We are reasonably certain that the case ascertainment is similar in the SEER registry and Ontario Cancer Registry; the degree of certainty about the diagnosis is similar, in that 96.2% of Ontario supraglottic cancer cases and 96.3% of SEER cases were histologically confirmed, 31,45 and no cases were ascertained by death certificate only. Degree of capture is similar in that two studies of Ontario cancer registration estimated that the proportion of cancer cases captured was 95% for all cancer sites, 28,29 whereas the SEER registries achieved 97.5% completeness in case ascertainment in a study conducted in 1989 across six of the nine SEER areas. 30 The characteristics of the SEER population of supraglottic patients are probably similar to U.S. supraglottic patients as a whole. The age and sex distributions that we observed in the SEER regions are strikingly similar to those reported for the National Cancer Data Base (NCDB), which represents 65% of laryngeal cancer patients in the United States diagnosed in The grouped stage distribution in the NCDB study was somewhat similar to SEER, with 36.5% and 37.9% having stage I/II disease, respectively, but fewer had stage IV disease: 37.6% in NCDB compared with 42.4% in SEER. NCDB treatment

8 SUPRAGLOTTIC CANCER IN CANADA AND THE UNITED STATES 503 practice was similar to SEER, with 32.4% of supraglottic cancer patients getting total or radical laryngectomy compared with 34.6% in SEER. We noted some case mix differences between SEER and Ontario that we controlled for in our analyses. We controlled for stage distribution differences by stratifying the analyses. Cox proportional hazards models controlled for age and sex. We were unable to address possible comorbidity differences, but when we ran a competing cause analysis by looking at deaths from cancer separate from deaths from other causes, we found no convincing evidence of large differences in comorbidity between the two regions. When we controlled for population-level differences in life expectancy by calculating relative survival, 33 we also found no differences from the crude survival comparisons either in the direction or magnitude of the results. We addressed a number of possible biases. We examined the representativeness of our findings from the Ontario chart-reviewed sample. We explored the implications of missing data. We addressed possible effects from differential data acquisition between the two regions. Differences between the random subset of our Ontario study population and the wider population of patients with supraglottic cancer indicate that our results may underestimate the initial Ontario laryngectomy rate by as much as 3.9%. Because of the exclusions from the SEER study population resulting from the stage being missing, our SEER results may overestimate the initial laryngectomy rate by as much as 5.9% (calculated as the difference in rates multiplied by the proportion with missing stage). If we apply these bias estimates to the observed initial laryngectomy rates, they explain only a portion of the differences observed between the regions. Our survival results may have been compromised by the higher survival rate in our Ontario study population compared with that of the Ontario target population. Although this finding did not reach statistical significance, the observed difference was large. Mitigating this possible bias is the same finding in SEER: those patients with missing stage who were, therefore, not included in our analyses, had worse survival. Availability of information in the SEER-Medicare data restricted the subcohort of patients 65 years of age and older to those who were most likely to have complete Medicare data. When we compared the 83% of patients with Medicare data with those without, we found that the latter group had slightly worse stage of disease, with a corresponding survival disadvantage that did not reach statistical significance. Given their more severe disease, inclusion of this group may have slightly increased the difference in laryngectomy rates observed between the areas. We adjusted the rate of subsequent surgeries by a correction factor based on comparisons of data from the different sources. If we had restricted our conclusions to the observed rates only (shown in Fig 4), our conclusions would not change. We assessed whether our laryngectomy rate comparisons from those 65 years and older could be applied to younger patients by comparing the rates in Ontario, where the information was available for all cases. We found that Ontario laryngectomy rates in the older and younger patients were similar, providing indirect support for a generalization of our laryngectomy rates for patients 65 years of age and older to the entire study population in both regions. We observed differences in the stage distributions, with more stage I disease in SEER and more stage IV disease in Ontario. These differences could be the result of true differences in stage distribution or of differences in the attribution of cases. We stratified our results by stage group to control for true distributional differences in stage, but stage differences within these groups may exist because of differences in how the staging was conducted. In the SEER registries, pathologic stage is used when it is available. Although we mimicked this approach in our Ontario stage assignment, the use of pathologic information for staging would have been more common in the SEER study population because more surgeries were conducted. We reviewed articles reporting staging accuracy in laryngeal cancer and found that, when combined with computed tomography (CT), clinical T-stage assignment is accurate about 82% of the time (this is an average of the reports reviewed) Our Ontario sample contained clinical with CT stage assignment for 52% of cases. When clinical information alone is used to stage, the accuracy is about 61% compared with pathologic staging Clinical (with and without CT) staging errors are more likely to understage rather than overstage, with the most common errors being assignment of T2 rather than T3, and T3 rather than T4. A higher rate of such errors in the Ontario data probably biased our results toward worse survival and, therefore, would not have reversed our findings. Last, staging accuracy may have differed because clinicians assigned stage in Ontario and cancer registrars assign stage in SEER. We were unable to find any reports in the literature on staging error rates in the SEER laryngeal carcinoma data. SEER does report extent of disease accuracy rates from reabstraction studies conducted in breast and lung cancer of 89% and 95%, respectively. 53 A similar approach to the current work was undertaken by Vermund et al 54 in Norway. These investigators compared treatment and outcome between two centers for patients with supraglottic carcinomas. The center in Norway had a treatment policy of primary radiotherapy, with surgery for failures, whereas the center in Wisconsin had a policy that placed more emphasis on primary surgery. The patients reviewed in the two centers were treated in different time periods: 1958 through 1978 in Wisconsin and 1978 through 1983 in Norway. Overall, 39% of the Wisconsin patients were cancer free with an intact larynx during 5 years of follow-up versus 56% of patients in Norway, suggesting superior results with the Norwegian policy. Among the possible reasons for the superior outcomes in Norway, the authors list improvements in radiotherapy delivery over time and better radiotherapeutic control of locoregional disease with attendant reduction in distant metastases. In a classic paper demonstrating the need to consider quality of life in treatment decisions, McNeil et al 55 observed that, if diagnosed with laryngeal cancer, about 20% of their study group of healthy volunteers would prefer radiotherapy to retain their natural voice. This finding occurred despite the 20% survival disadvantage that the investigators attributed to the radiotherapy treatment option. Recent work by Sharp et al 56 has reinforced the presence of important treatment issues other than survival that patients want to consider in the treatment decision in laryngeal

9 504 GROOME ET AL cancer. Given our demonstration of survival equivalence, the current work supports radiotherapy, with surgery reserved for salvage, as a reasonable treatment option for stages I to III supraglottic cancer. Oncologists disagree about best treatment in supraglottic cancer at all stages, and it is only since the publication of the Veterans Affairs trial, 14 in which induction chemotherapy followed by radiation therapy demonstrated survival equivalence to surgery, that advocates of radical surgery have considered larynx-preservation approaches for advanced disease Refinement of the chemoradiation approach is currently being studied, with the recent RTOG results indicating that induction chemotherapy does not confer a survival advantage over radiation alone and that concurrent chemoradiation may provide such an advantage. The Singapore trial 19 may provide evidence regarding the radiotherapy versus surgery question, although the answer will not be specific to laryngeal cancer. We have demonstrated that, for stages I to III supraglottic cancer, survival is not compromised in Ontario, Canada, where radiotherapy, reserving surgery for salvage, is widely regarded as the treatment of choice. We have also demonstrated that, with this conservative approach, the initial higher rate of larynx retention is maintained over the follow-up period. The larynx retention advantage resulting from the more common use of radiotherapy in stage IV Ontario patients may be undermined by the modest cancer survival advantage afforded the stage IV patients in SEER. Perhaps the decision to use conservative treatment is dependent on the trade-off between patient values for natural voice retention and his/her values for an increased chance of initial cure. 1. Harwood AR, Rawlinson E: The quality of life of patients following treatment for laryngeal cancer. Int J Radiat Oncol Biol Phys 9: , Natvig K: Study No. 1: Social, personal, and behavioural factors related to present mastery of the laryngectomy event. J Otolaryngol 12: , Mathieson CM, Henderikus JS, Scott JP: Psychosocial adjustment after laryngectomy: A review of the literature. J Otolaryngol 19: , DeSanto LW, Olsen KD, Perry WC, et al: Quality of life after surgical treatment of cancer of the larynx. Ann Otol Rhinol Laryngol 104: , Hillman RE, Walsh MJ, Wolf GT, et al: Functional outcomes following treatment for advanced laryngeal cancer. Part I Voice preservation in advanced laryngeal cancer. Part II Laryngectomy rehabilitation: The state of the art in the VA System Research Speech-Language Pathologists Department of Veterans Affairs Laryngeal Cancer Study Group. Ann Otol Rhinol Laryngol 172:1-27, McQuellon RP, Hurt GJ: The psychosocial impact of the diagnosis and treatment of laryngeal cancer. Otolaryngol Clin North Am 30: , Davis LW, Fazekas JT: Controversy in the management of laryngeal tumors: Radiation therapy perspective, in Thawley SE, Panje WR (eds): Comprehensive Management of Head and Neck Tumors. Philadelphia, PA, W.B. Saunders, 1987, pp Harwood AR: Cancer of the larynx the Toronto experience. J Otolaryngol 11:1-21, 1982 (suppl) 9. DeSanto LW: Controversy in the management of laryngeal tumors: Surgical perspective, in Thawley SE, Panje WR (eds): Comprehensive Management of Head and Neck Tumors. Philadelphia, PA, W.B. Saunders, 1987, pp Myers EN, Alvi A: Management of carcinoma of the supraglottic larynx: Evolution, current concepts, and future needs. Laryngoscope 106: , O Sullivan B, Mackillop WJ: An approach to the interpretation of the literature of head and neck cancer. Clinics Oncol 5: , Lefebvre J, Bonneterre J: Current status of larynx preservation trials. Curr Opin Oncol 8: , Pignon JP, Bourhis J, Domenge C, et al: Chemotherapy added to locoregional treatment for head and neck squamous-cell carcinoma: Three meta-analyses of updated individual data. Lancet 355: , Department of Veterans Affairs Larynx Cancer Study Group: Induction chemotherapy plus radiation compared with surgery plus radiation in patients with advanced laryngeal cancer. N Engl J Med 324: , Richard JM, Sancho-Garnier H, Pessey JJ, et al: Randomized trial of induction chemotherapy in larynx carcinoma. Oral Oncol 34: , 1998 ACKNOWLEDGMENT The acknowledgment is available online at REFERENCES 16. Tannock IF, Cummings BJ: Neoadjuvant chemotherapy in head and neck cancer: No way to preserve a larynx. J Clin Oncol 10: , Lefebvre JL, Chevalier D, Luboinski B, et al: Larynx preservation in pyriform sinus cancer: Preliminary results of a European Organization for Research and Treatment of Cancer phase III trial. J Natl Cancer Inst 88: , Forastiere AA, Berkey B, Maor M, et al: Phase III trial to preserve the larynx: Induction chemotherapy and radiotherapy versus concomitant chemoradiotherapy versus radiotherapy alone, Intergroup Trial R Proc Am Soc Clin Oncol 20:4, 2001 (abstr) 19. National Cancer Institute US: PDQ clinical trial abstracts. Bethesda, MD. nci. nih. gov/ 20. Groome PA, Mackillop WJ, Rothwell D, et al: The management and outcome of glottic cancer: A population-based comparison between Ontario, Canada and the SEER areas of the United States J Otolaryngol 29:67-77, Groome PA, O Sullivan B, Irish JC, et al: Glottic cancer in Ontario, Canada and the SEER areas of the United States: Do different management philosophies produce different outcome profiles? J Clin Epidemiol 54: , Imbens A, Angrist J: Identification and estimation of local average treatment effects. Econometrica 62: , Newhouse JP, McClellan M: Econometrics in outcomes research: The use of instrumental variables. Annu Rev Public Health 19:17-34, Potosky AL, Riley GF, Lubitz JD, et al: Potential for cancer related health services research using a linked medicare-tumor registry database. Med Care 31: , Kosary CL, Ries LAG, Miller BA, et al: SEER cancer statistics review, : Tables and graphs. Bethesda, MD, National Cancer Institute, NIH publication , Clarke EA, Marrett LD, Kreiger N: Cancer registration in Ontario: A computer approach, in Jensen OM, Parkin DM, MacLennan R, et al (eds): Cancer Registration Principles and Methods. Lyon, France, IARC publication No. 95, 1991, pp Statistics Canada. CANSIM, matrices : Population statistics. (January 1999) Robles SC, Marrett LD, Clarke EA, et al: An application of capturerecapture methods to the estimation of completeness of cancer registration. J Clin Epidemiol 41: , Holowaty EJ, Moravan V, Lee G, et al: Accuracy of data elements in the Ontario Cancer Registry: A reabstraction study. Report to Health Canada. Toronto, Ontario Cancer Registry, 1995