Analytical Studies: Methods and References
Canadian Cancer Treatment Linkage Project

Skip to text

Text begins

Acknowledgements

Acknowledgements to Working Group members: Mary Jane King, Cancer Care Ontario; Heather Stuart-Panko, Saskatchewan Cancer Agency; Sheila Fukumura, Cancer Care Manitoba; Kim Vriends, Prince Edward Island Cancer Registry; Maureen MacIntyre, Cancer Care Nova Scotia; Ryan Woods, BC Cancer Agency; Gordon Walsh, Cancer Care Nova Scotia; Greg Webster, Director, Acute and Ambulatory Care Information Services, Canadian Institute for Health Information (CIHI); Janet Manuel CHIM, Classification Specialist, Classifications and Terminologies, CIHI; and Alana Lane CHIM, Classification Specialist, Classifications and Terminologies, CIHI.

Acknowledgements to consulted medical clinicians for providing expert guidance on selection of surgical treatment of the six cancers investigated. Dr. Ralph Gilbert, MD, FRCSC, Head & Neck Surgeon at the University Health Network, provided recommendations for listed surgical interventions to treat thyroid cancer. Dr. Geoffrey Gotto, MD, MPH, FRCSC, Clinical Associate Professor, Department of Surgery, The University of Calgary, provided recommendations for listed surgical interventions to treat urinary bladder cancer. Dr. Christian Finley, MD, FRCSC, Associate Professor in the Department of Surgery Division of Thoracic Surgery, McMaster University, provided recommendations for listed surgical interventions to treat cancer of the lung and bronchus.

Executive summary

Record linkage has been identified as a potential mechanism to add treatment information to the Canadian Cancer Registry (CCR). The purpose of the Canadian Cancer Treatment Linkage Project (CCTLP) pilot is to add surgical treatment data to the CCR. The Discharge Abstract Database (DAD) and the National Ambulatory Care Reporting System (NACRS) were linked to the CCR, and surgical treatment data were extracted. The project was funded through the Cancer Data Development Initiative (CDDI) of the Canadian Partnership Against Cancer (CPAC).

The CCTLP was developed as a feasibility study in which patient records from the CCR would be linked to surgical treatment records in the DAD and NACRS databases, maintained by the Canadian Institute for Health Information. The target cohort to whom surgical treatment data would be linked was patients aged 19 or older registered on the CCR (2010 through 2012) with a primary diagnosis of the following: female breast, colorectal, prostate, thyroid, urinary bladder, or lung cancer. To identify primary surgical treatments for these cancers, code sets were developed for each site using current standards (for example, Canadian Classification of Health Interventions [CCI]). With this linkage, two new core data elements were developed and added to the linked CCR-DAD-NACRS analytical file: Procedure date and Primary procedure.

The linkage was completed in Statistics Canada’s Social Data Linkage Environment (SDLE). Within the SDLE, each file (CCR, DAD, NACRS) was linked to the Derived Record Depository (DRD), a regularly updated repository of personal identifiers for all Canadians. Linkage keys extracted through this process were used to create the linked CCR–DAD–NACRS file, from which the cohort was extracted (records with only one tumour from among the six target cancers coupled with surgical interventions identified in the treatment code set). Linkage rates for the three files were robust, with each having a rate greater than 90% in the years covered.

The CCTLP demonstrated the feasibility of using record linkage to add surgical treatment data to patient records for six cancers. Opportunities for further development were identified, including the need to improve linkage rates to minimize the number of surgical treatments that are lost. In addition, a protocol for assigning one or more surgical treatments to patient records where multiple tumours are present in the same organ, within the follow-up period, will be required. Nevertheless, record linkage has been shown to be an effective means of increasing the analytical value of Canadian cancer data holdings.

1 Introduction

The Canadian Cancer Registry (CCR), established in 1992, is a collaborative undertaking between Statistics Canada and the 13 provincial and territorial cancer registries to create a single database to report annually on cancer incidence and survival at the national and jurisdictional levels (Statistics Canada n.d.b, 2011). The Registry produces high-quality information on cancer events, but lacks information about treatment. The addition of treatment information would enhance the CCR’s surveillance capacity and its analytical capacity for researchers and epidemiologists.

To address this information gap, Statistics Canada, in partnership with the Canadian Council of Cancer Registries (CCCR), undertook a study to determine the feasibility of using record linkage to add treatment information to the CCR for three cancers—breast, prostate and colorectal—in four provinces (Ontario, Manitoba, Nova Scotia, and Prince Edward Island). The study involved linking hospital data (Discharge Abstract Database [DAD] and National Ambulatory Care Reporting System [NACRS]) to the CCR. The results demonstrated the feasibility of using record linkage to add treatment data to the CCR, specifically, surgical treatment data, which are comprehensively reported in the hospital data (Carrière et al. 2015).

The Canadian Cancer Treatment Linkage Project (CCTLP) builds on that work. Using the Social Data Linkage Environment (SDLE) at Statistics Canada, the DAD and the NACRS were linked, and administrative, diagnostic and surgical treatment data were extracted and added to the CCR.

This report provides information on the record linkage process, data validation, and surgical treatment rates for six types of cancer—breast, colorectal, prostate, urinary bladder, thyroid, and lung and bronchus. The project was funded through the Cancer Data Development Initiative (CDDI) of the Canadian Partnership Against Cancer (CPAC n.d.). The linkage was approved by the Statistics Canada Executive Management Board (May 2016) (Statistics Canada n.d.a). Use of the linked data is governed by Statistics Canada’s Directive on Record Linkage (Statistics Canada n.d.c).

2 Data sources

2.1 Canadian Cancer Registry

The Canadian Cancer Registry (CCR) contains information about all cancers diagnosed in Canada, compiled from provincial and territorial cancer registries. It covers all Canadian residents, living and deceased, diagnosed with cancer since 1992, including primary (incident) cancers among patients previously diagnosed with cancer. Every calendar year, the CCR reports confirmed information about each new tumour, including tumour type and date of diagnosis, and demographic data about the patient (Statistics Canada 2008). CCR records from 1992 to 2013 were available for linkage (n = 3,126,295). 

2.2 Discharge Abstract Database

The Discharge Abstract Database (DAD) contains demographic, administrative, and coded diagnostic and intervention data for acute care, some psychiatric, chronic rehabilitation, and selected day surgery hospital discharges (CIHI 2010a, 2011a, 2012b, 2012c, 2013, 2014a, 2015a). These are reported annually by all jurisdictions, excluding Quebec, to the Canadian Institute for Health Information (CIHI) on a fiscal year basis (April 1 to March 31). The DAD registers about 3.7 million discharges per year. DAD discharges occurring between April 1, 1994, and March 31, 2015, were available for linkage (n = 77,925,269).

2.3 National Ambulatory Care Reporting System

The National Ambulatory Care Reporting System (NACRS) contains data about visits to health care facilities for ambulatory care, including community-based services, day surgery procedures, emergency department visits, diagnostic imaging, and selected clinic visits (for example, oncology care) (CIHI 2009b, 2010b, 2011b, 2011c, 2011d, 2012d, 2012e, 2014b, 2015b). At each visit, patient demographics, clinical information (diagnoses, surgical interventions), and administrative, financial and service-specific data are recorded. NACRS data are reported to CIHI on a fiscal year basis (April 1 to March 31).

NACRS data are reported most comprehensively by Ontario; less so for other provinces and territories (CIHI 2010d). Newfoundland and Labrador, and New Brunswick did not report to the NACRS for all years; Quebec does not report to the NACRS. NACRS records for April 1, 2002, through March 31, 2015, were available for linkage (n = 166,069,085).

2.4 Canadian Vital Statistics Database

The Canadian Vital Statistics (Death) Database (CVSD) compiles demographic and medical (cause of death) information annually from all provincial and territorial vital statistics registries on all deaths in Canada (Statistics Canada n.d.d). Deaths occurring from 1970 through 2012 were available for linkage (n = 8,574,561, which includes 731,953 deaths for the 2010-to-2012 period).

3 Record linkage

The linkage was conducted at Statistics Canada using the Social Data Linkage Environment (SDLE), a highly secure linkage environment facilitating the creation of linked population data files for social analysis. The linkage was conducted separately for each database in three steps: (1) data preparation, (2) record linkage, and (3) quality assessment. Given the unique nature of each database, different linking variables, methodologies, and quality assessment measures were employed (Table 1). 

3.1 Data preparation

For all four databases—CCR, DAD, NACRS, and CVSD—data preparation included a quality assessment of the linkage variables to determine the completeness and the validity of the data. This procedure identifies data errors or omissions that may impede correct linkage of a record. The choice of linkage variables has a direct impact on the efficiency of the record linkage operation. Information associated with the linkage variables must be accurately recorded, available for the vast majority (if not all) of individuals in the files to be linked, and as discriminating as possible. Each file contained a different set of linkage variables (Table 1). Exclusion criteria varied depending on the linkage strategy. 

Exclusion criteria were applied to each input data set, where applicable. Because the DAD and the NACRS were linked deterministically using only three variables, missing information for any of the three variables would make an accurate linkage impossible. 

In addition, a separate processing step was applied to the CCR to identify unique individuals. Individuals may be represented in the CCR more than once if they were diagnosed with cancer more than once. The CCR data file was unduplicated within provinces and territories, thereby facilitating linkage at the person level rather than the tumour level. This process identified 3,053,697 unique individual–province combinations. An individual diagnosed with multiple cancers in different provinces or territories would be represented more than once. To handle this situation, the CCR was linked with an N:1 correspondence, that is, more than one individual–province combination could have linked to one DAD record.

3.2 Record linkage methodology

The linkage was conducted at Statistics Canada in the SDLE. At the core of the SDLE is a Derived Record Depository (DRD), a national dynamic relational database containing only basic personal identifiers created by linking selected Statistics Canada source index files in order to produce a list of unique individuals. Each input data file (CCR, NACRS, DAD, CVSD) was separately linked to the DRD using methods appropriate to the availability of linkage variables. The following describes the linkage methodology used for each input file. The methods are summarized in Table 1.

3.3 Canadian Cancer Registry

The CCR was linked to the DRD using G-Link, a generalized record linkage system developed by Statistics Canada based on probabilistic linkage methodology developed by Ivan P. Fellegi and Alan B. Sunter. Probabilistic record linkage uses non-unique identifiers (such as name and birth date) to calculate the likelihood that records refer to the same entity (for example, individual). Probabilistic record linkage is especially valuable when the identifiers are subject to change (females’ surnames, for instance), error-prone, or frequently missing. The linkage was conducted using a range of linkage variables, including dates of birth and death, names, and geographic locations (Table 1). If CCR records contained health insurance numbers (HINs), this information was included in the DRD to facilitate linkage to the hospital data. Overall, 95.87% (n = 2,927,463) of unique individual–province identifiers in the CCR were linked to the DRD.

3.4 Discharge Abstract Database and National Ambulatory Care Reporting System

The DAD and NACRS data were linked to the DRD using a two-phase deterministic linkage. In the first phase, a linkage key was created based on sex, date of birth, and postal code for records with complete information (n = 164,649,442). That key was used to deterministically link records to the DRD. Only unique exact matches were retained (only one DRD record linked with a given key). During this phase, HINs in the DAD and the NACRS were extracted and included in the DRD as an additional unique identifier to facilitate future linkages. In the second phase, unlinked records were deterministically linked (exact match) to the DRD using only HINs.

The NACRS was linked first to the DRD. In the first phase, 78.9% (n = 129,985,322) of NACRS records with a valid key were linked to the DRD. A total of 2,895,602 links were broken, reflecting cases where a NACRS record linked to two different people on the DRD, and the conflict could not be resolved with available information. A further 22,953,303 links were created among NACRS records sharing the same HIN, resulting in a total linkage rate of 90.3%.

For the DAD data, 71.2% (n = 55,015,973) of records were linked to the DRD. Another 4,770,333 links were broken, reflecting cases where a DAD record linked to two different people on the DRD, and given the available information, the conflict could not be resolved. A further 11,230,936 links were created among DAD records sharing the same HIN, resulting in a total linkage rate of 85.0%.

3.5 Canadian Vital Statistics Database

The CVSD was linked to the DRD using probabilistic linkage. The linkage was conducted with a range of linkage variables, including dates of birth and death, names, and geographic locations. Overall, 67.1% (n = 5,749,144) of individuals were linked to the DRD. This low rate was expected, given the poor coverage of the DRD before 1980. For the study period (2010 to 2012), the linkage rate was 97.7% (714,825 divided by 731,953).

3.6 Quality assessment

Error estimation was conducted for each linkage to assess the quality of the linkage of each input file to the DRD. For the CCR, sensitivity (true linkage rate) and specificity (true non-linkage rate) were calculated by comparing the results of G-Link to a manual review of a randomly selected sample of links and non-links. The quality of the linkage was deemed high, with sensitivity and specificity rates of 97.74% and 99.36%, respectively. For the CVSD, the sensitivity was 95.4% for the 1970-to-2011 period and 98.4% for 2012. Specificity was 97.8% for the 1970-to-2011 period and 83.5% for 2012.

No manual review was conducted to determine error rates for the DAD and NACRS linkages. However, results of the second phase using HINs provide some measure of the error rate:   0.01% (n = 9,757) of DAD transactions and 0.003% (n = 5,718) of NACRS transactions were linked to different persons in the two phases.

4 Surgical treatment

Further validation was conducted to determine the fitness of the linked data for reporting surgical treatment for six cancer types—female breast, colorectal, prostate, urinary bladder, thyroid, and lung and bronchus—the leading types of new cancers in Canada (Canadian Cancer Society’s Advisory Committee on Cancer Statistics 2014). Furthermore, treatment for these cancers typically requires surgical intervention. The following describes the tumour selection process, linkage rates, and treatment rates for these cancers.

4.1 Cohort selection

A cohort of new primary malignant cancer tumours was selected, consisting of people aged 19 or older diagnosed from January 1, 2010, through December 31, 2012. For urinary bladder cancer, in situ tumours were also included. International Classification of Diseases for Oncology, Third Edition (ICD-O-3) (Fritz et al. 2000) codes were used to define the tumour cohort; these were grouped using Surveillance, Epidemiology, and End Results (SEER) Program grouping definitions (Horner et al. n.d.) (Table 2). Histology for all cancer types excluded: mesothelioma (M-9050 to M-9055), Kaposi sarcoma (M-9140), and hematopoietic and lymphoid neoplasms (M-9590 to M-9992).

To determine a single primary tumour for each individual, the International Agency for Research on Cancer (IARC) rules (International Agency for Research on Cancer et al. 2004) for multiple primary tumours were applied to the CCR. In general, application of these rules removes subsequent tumours of the same type and histology. This modified CCR file forms the basis of the data released by Statistics Canada to the public (Statistics Canada n.d.e) and is accessible to researchers in the Research Data Centres (RDCs). The file is called the IARC Tabulation Master File (TMF) (Statistics Canada 2008).

Individuals may be represented in the IARC TMF more than once if, for example, they were diagnosed with more than one tumour of the same type but of a different histology. To assign surgical treatment at the individual level, it was necessary to ensure that only one tumour of a given type was included for each cancer patient. Hence, a further review of the IARC TMF was conducted to remove tumour records in cases where multiple tumours of the same type were identified for the same patient occurring one year before and/or one year after the date of diagnosis of the primary tumour. This would remove cases of multiple tumours of the same type but different histology, for example. 

Reported treatment rates were not age- or sex-adjusted. However in addition to overall surgical treatment rates, rates were produced by sex and by age group (19 to 49, 50 to 69, and 70 or older).

4.2 Treatment codes

A comprehensive list of potential surgical treatments was developed for each cancer type, based on published sources, including the National Comprehensive Cancer Network Clinical Practice Guidelines in Oncology (NCCN Guidelines) (NCCN 2002, 2013a, 2013b, 2014, 2015a, 2015b, 2015c, 2016) and Facility Oncology Registry Data Standards (FORDS) (Commission on Cancer 2002). The lists pertinent to breast, colorectal and prostate cancers had previously been reviewed by members of the feasibility national advisory committee, technical experts at the provincial cancer agencies, and clinical experts when required (Carrière et al. 2015). After consultations, a final set of surgical treatments was selected. For the three additional cancer types, the initial list of treatments was reviewed, and consultations were held with clinical experts and classification experts from CIHI. Appendix B contains the list of surgical interventions included for each cancer type.

The Canadian Classification of Health Interventions (CCI), versions 2009 and 2012, (CIHI 2009a, 2012a) were used to define the surgical intervention in the DAD and NACRS. All intervention fields in the DAD (20) and NACRS (10) records were used to identify the surgical interventions associated with each cancer type. This was done independently for each surgical treatment code because multiple treatments in a single hospital admission are captured as separate treatment events.

4.3 Follow-up period

Surgical treatments occurring within one year after or 31 days before the tumour date of diagnosis recorded on the CCR were considered. The admission date recorded in the DAD and the NACRS was used to determine the eligibility of interventions contained in the record. 

4.4 Reporting facility type

The DAD and the NACRS represent different frames of hospital services that are expected to have an impact on reporting treatment rates. The DAD includes all discharges from all acute care facilities for all territories and provinces, except Quebec, and represents about 75% of all acute separations for Canada (CIHI 2012c). For the reference period of this analysis (fiscal years 2009/2010 through 2013/2014), health service facilities reported same-day surgery visits to the DAD and/or to the NACRS depending on the year and jurisdiction (CIHI 2009b, 2010b, 2011b, 2011c, 2011d, 2012d, 2012e, 2014b, 2015b). Around 2.4 million day-surgery visits are submitted to CIHI annually—35% are sent to the DAD, and 65%, to the NACRS (CIHI 2012d).

The NACRS includes a broader range of services: emergency room visits, day surgery, oncology clinics, Cancer Care Ontario for oncology care, and other types of ambulatory care (for example, renal dialysis clinics) (CIHI 2009b, 2010b, 2011b, 2011c, 2011d, 2012d, 2012e, 2014b, 2015b). The data for provinces that report surgical events to the NACRS offer a greater opportunity to link tumours to surgical treatments (for instance, emergency departments, oncology clinics) than is available for provinces not reporting visits for the same range of services to the NACRS. Consequently, overall treatment rates are expected to be higher in jurisdictions with wider ranges of surgical event coverage. Table 3 displays information on coverage, by reference year, for all jurisdictions. CIHI offers guidelines to prevent double-counting of day-surgery events between the DAD and the NACRS (CIHI 2009b).

All in-scope records that contained one or more of the selected surgical interventions that had linked to the six types of cancers in the cohort were considered for analysis of treatment rates; no exclusions based on reporting facility type were applied. Surgeries to treat these cancers can occur outside hospital settings; for example, at specific cancer centres (Winnipeg Regional Health Authority n.d.), in practitioners’ offices, and at private clinics. Surgeries performed in those settings were not included in this analysis.

4.5 Surgical treatment rates

Surgical treatment rates for each cancer type are reported by province, and year. Numerators are the number of tumours having at least one occurrence of the selected surgical intervention during the follow-up period. Denominators are the total number of tumours in the CCTLP Tumour Cohort (CCTLP-TC).

5 Results

5.1 Cohort selection

Table 4 presents the selection process for the CCTLP-TC. Overall, 225,330 single primary cancer tumours were selected, representing 97.4% of tumours of the same site reported in the CCR tumour file. 

The linked tumour cohort median patient age for female breast cancer ranged from 59 to 65 across provinces; for colorectal, from 68 to 73; for prostate, from 65 to 69; for lung and bronchus, from 69 to 72; for urinary bladder, from 66 to 76; and for thyroid, from 47 to 57 (data not shown). 

Because cancer outcomes vary for men and women, distributions by sex for four cancer sites were considered. In the linked tumour cohort and across jurisdictions, men accounted for about three-quarters (75% to 76%) of urinary bladder tumours; just over half of colorectal (54% to 55%) and lung and bronchus tumours (51% to 52%); and one-fifth to one-quarter (22% to 24%) of thyroid tumours (data not shown).

5.2 Linkage results for the Canadian Cancer Treatment Linkage Project-Tumour Cohort

Overall, 99% of cancer tumours were linked to the DRD, rendering them eligible to link to a hospital record. Rates were consistent across provinces and territories (where reportable), but were lower for lung and bronchus tumours diagnosed in Nunavut (Table A.1).

Three-year linkage rates for the DAD and NACRS (2009/2010 to 2012/2013) files that were used to identify surgical interventions were greater than 90%. Linkage rates varied across provinces; the lowest levels were reported for the Northwest Territories and Nunavut (Table A.2).

5.3 Treatment rates

Tables 5 to 10 show the percentage of tumours receiving at least one type of surgical intervention during the follow-up period for each type of cancer.  The results are presented for all years of data combined. An examination of rates by single years revealed consistent patterns across years (data not shown).

The majority (88%) of female breast cancer tumours received a surgical intervention, with rates ranging from 85% in Manitoba to 92% in Prince Edward Island (Table 5). Rates in the territories were more variable (84% to 93%), owing to smaller numbers of cases. Surgical rates varied by patient age, with the highest among women younger than 70 (Table A.3).

Similarly, the majority (83%) of colorectal cancer tumours received a surgical intervention, with rates ranging from 81% to 82% in Ontario and Manitoba to 87% in Newfoundland and Labrador and in British Columbia (Table 6). Because of the smaller numbers of cases, rates for the territories were variable. Colorectal surgical rates were highest at ages 50 to 69 (Table A.4).

Overall, about a third (31%) of prostate cancer tumours received a surgical intervention. Rates ranged from 17% in Prince Edward Island to 35% in Newfoundland and Labrador and Nova Scotia (Table 7). Annual surgical rates varied considerably (10% to 22%) in Prince Edward Island, a result of the relatively small number of cases in that province (data not shown). In all jurisdictions, the highest rate of surgical intervention was at ages 19 to 49 (Table A.5).

Of the six selected cancer sites, lung and bronchus tumours had the lowest surgical rates—overall, 19% received at least one of surgical intervention. Rates ranged from 14% in Prince Edward Island to 25% in New Brunswick (Table 8). In all jurisdictions, rates were highest at ages 19 to 49 (Table A.6).

A large majority (91%) of bladder tumours received surgical treatment; rates were high in all jurisdictions, ranging from 88% in Manitoba and Ontario to 94% in Newfoundland and Labrador (Table 9). The highest rates were at ages 50 to 69 (Table A.7).

Most thyroid tumours received at least one surgical treatment. The overall rate was 93%, ranging from 91% in Prince Edward Island, Manitoba and British Columbia to 98% in Newfoundland and Labrador (Table 10).  Thyroid surgical patients tended to be younger than those who had surgery on other cancer sites; the lowest rates were among patients aged 70 or older (Table A.8).

6 Discussion

This study demonstrates how record linkage can be used to add surgical treatment information to a national cancer registry. By means of the SDLE platform at Statistics Canada, most records in the CCR, the DAD and the NACRS were linked. The surgical treatment rates for the six selected cancers derived from the linked data reflected expected values. Because they are based on the majority of tumours of the selected types diagnosed between 2010 and 2012, the results are deemed unbiased and representative of the surgical treatment experience of cancer patients in Canada.

The linkage rates achieved for each input file were 90% or better; the rate was 99% for the selected cancer tumours diagnosed from 2010 to 2012 among adult patients. For the DAD, rates were highest for the most recent years of data and for records representing patients aged 19 or older.

An advantage of conducting the linkage at the national level is the ability to capture surgical treatments occurring outside patients’ province or territory of residence. Generally, national-level linkage rates equal or exceed rates from previous linkage projects (Carrière et al. 2015; Rotermann et al. 2014, 2015). Nonetheless, some regional variation in linkage rates was apparent for the DAD. Continued efforts are required to improve DAD linkage rates, specifically, for selected regions, including the Northwest Territories and Nunavut, to ensure comparability.

DAD linkage rates were higher when HINs were available. However, not all cancer registries currently report HINs to the CCR. Continued use of HINs as a linkage variable would increase the pan-Canadian linkage rate, but differentially affect rates for reporting and non-reporting provinces and territories. Requesting HINs is part of the CCR annual cancer data call; comprehensive submission from all provincial and territorial cancer registries would help to resolve this issue. In addition, an assessment of the quality of HIN reporting to the CCR should be undertaken before it is assigned a primary role in linkage with the DAD or other datasets.

Surgical treatment rates derived from the linked CCR–DAD–NACRs data are generally at expected levels and comparable to published information on surgical rates. As anticipated, surgical rates varied by cancer site and were higher for breast, colorectal, urinary bladder, and thyroid cancers, compared with lung cancer, for which survival outcomes are poor, and prostate cancer, for which ” active surveillance” may be the preferred approach (Dragomir, Cury and Aprikian 2014).

The results of this study indicated that breast-conserving surgery and mastectomy were the most prevalent types of surgical treatment for breast cancer. Other published sources have identified these surgeries as integral to breast cancer treatment (Urbach, Simunovic and Schultz 2008; Quan et al. 2008; CIHI and CPAC 2012; Turner et al. 2007). Combined breast-conserving surgery and/or mastectomy rates resemble those previously reported for Ontario and Manitoba (Quan et al. 2008; CIHI and CPAC 2012; Turner et al. 2007). Furthermore, disaggregated breast surgical rates used for validation (data not shown) showed similarities to published results for cancer system performance reports (CPAC 2012, 2016). Compared with rates calculated for four provinces during the 2005-to-2008 period (Carrière et al. 2015), the breast surgery rates in this study are slightly lower or higher, possibly because of different linkage approaches. This demonstrates that use of the SDLE platform yielded higher rates for breast surgeries for two provinces than had been obtained via direct linkage using only HINs reported to the CCR in the earlier feasibility study.

The present analyses revealed differences in breast surgery by patient age (data not shown). Therefore, some variation in treatment rates across jurisdictions or in comparison with other reports is due, in part, to differences in the age distribution of this cohort. Further analysis is required to assess treatment rates for all six cancers by patient characteristics and cancer stage.

Surgery rates for colorectal cancer in Ontario were similar to published findings (Carrière et al. 2015; Nenshi et al. 2008). Surgical treatment rates were highest at ages 19 to 49, consistent with results noted in the earlier feasibility report (Carrière et al. 2015).

A decade ago, an analysis of U.S. data found that 93.4% of nearly 54,000 thyroid cancers (histologies taken together) received surgical treatment (thyroidectomy and/or lymph node sampling and/or dissection) (Hundahl et al. 1998). According to the present study, 93% of thyroid tumours received at least one surgical treatment, primarily thyroidectomy.

Patients with urinary bladder cancer frequently experience recurrence (Kassouf et al. 2010), and with the prevalence of this cancer being 10 times its incidence (Kassouf et al. 2010), the likelihood of at least one surgical treatment was expected to be high. In fact, rates consistently exceeded 91%. Treatment rates were not calculated by tumour stage; however, this likely would impact rates for surgery. A retrospective review using Alberta Cancer Registry data from 2007 to 2011 reported that overall, 27.8% of high-grade T1 bladder cancer experienced early repeat resection, and that by 2011, the rate had increased to 37.8% (Gotto, Shea-Budgell, and Ruether 2016). For future analyses, the record linkage approach in this study would enable measurement of changes in surgical patterns across time.

The utility of linked data about surgery for cancer depends in part on the accuracy and comprehensiveness of hospital data. Evidence suggests that cancer registry information about surgical treatment is more complete than information in the DAD (Turner et al. 2007). Consequently, this analysis may underestimate treatment rates. As well, DAD and NACRS coding standards may limit the degree to which the data can be used to report specific surgical interventions. For example, previous research found lower-than-expected rates of lymph node removal for breast and prostate cancer (Carrière et al. 2015). This was attributed, in part, to the fact that multiple axillary lymph node procedures are not always recorded separately in the DAD when radical mastectomy and prostatectomy are performed. Therefore, obtaining comprehensive or absolute counts of lymph node interventions is not feasible for all years of DAD and NACRS data. Mandatory reporting guidelines may have addressed this issue in more recent years of DAD and NACRS data. Further analyses of these newly linked data are required to determine the accuracy of reporting more specific surgical interventions.

Finally, results of this and previous studies have demonstrated the feasibility of using hospital data (such as data from the DAD and NACRS) linked to cancer registry data to derive surgical treatment rates. This approach is appropriate when interventions occur primarily in hospital or clinic settings that report to one of the two national hospital data sources. It may not be appropriate for some types of cancers (such as skin cancer) for which surgical interventions may occur in physicians’ offices. Linkage to physician claims data would be required to capture this information. 

7 Limitations

Although the linkage rates were considered to be robust for both the DAD and NACRS, the 8% of non-linking cases potentially represent missed surgical treatments. Furthermore, the feasibility of using record linkage to capture surgical interventions to report on treatment for childhood cancers warrants further investigation, as overall linkage rates for DAD records related to children were generally lower (data not shown).

The current study was based on the majority of tumours (97.4%) of the selected cancer types, excluding  cases where: (1) more than one primary tumour in the same organ was reported to the CCR; (2) the diagnosis dates were within 365 days of each other; and (3) the tumour record did not link in the SDLE. The overall impact was a loss of 2.6% of tumours reported to the CCR. Colorectal cases were most affected, with a loss of 5.4%, followed by female breast with a loss of 2.9%. The impact of multiple tumours may be more pronounced for other cancer sites. Given the overall high linkage rate for the CCR, most exclusions were based on criteria 1 and 2. The challenge presented by multiple tumours is proper attribution of a surgical intervention. The results of this study cannot be generalized to cases with multiple tumours. Future work should focus on determining the feasibility of using record linkage to correctly assign surgical information to the appropriate tumour in such cases.

Except for urinary bladder cancer, this study did not include in situ tumours. Future studies should attempt to include them, and thereby, determine the feasibility of using linked data to report on treatment rates for these tumours.

Surgical treatments that may have occurred outside of hospital settings, for example, prostate surgery in physician’s office or surgery at the Winnipeg Breast Health Centre (Winnipeg Regional Health Authority n.d.) were not captured in the data used in this study. Therefore, surgical treatment rates are slightly underestimated.

8 Conclusions

Results of this study demonstrate the feasibility of using record linkage to bring together information in cancer registries with surgical intervention information in hospital data. The use of Statistics Canada’s linkage environment, SDLE, is a viable, cost-effective method of adding surgical treatment data to the CCR, and thereby, enhancing the capacity to report on a key treatment modality at the national level. Future work should focus on continued improvement of linkage rates, specifically, for hospital data; the feasibility of extending this approach to cases representing multiple tumours, younger patients, and other types of cancers; and the quality of surgical data. 

Appendix A – Additional tables on linkage results

Appendix B – Surgical treatments for cancer primary sites

References

Canadian Cancer Society’s Advisory Committee on Cancer Statistics. 2014. Canadian Cancer Statistics 2014. Toronto: Canadian Cancer Society.

Carrière, G.M., C. Sanmartin, R. Trudeau, P. Murison, D. Turner, M.J. King, K. Vriends, R. Woods, R. Loucini, B. Wagar, and G. Lockwood. 2015. The feasibility of adding treatment data to the Canadian Cancer Registry using record linkage. Health Research Working Paper Series, no. 9. Statistics Canada Catalogue no. 82-622-X. Ottawa: Statistics Canada.

CIHI (Canadian Institute for Health Information). 2009a. Canadian Classification of Health Interventions (CCI): Volume Three – Tabular List. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2009b. National Ambulatory Care Reporting System Manual, 2009–2010. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2010a. Data Quality Documentation, Discharge Abstract Database, 2009–2010: Executive Summary (Revised February 2011). Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2010b. National Ambulatory Care Reporting System Manual for 2010-2011. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2011a. Data Quality Documentation for External Users: Discharge Abstract Database, 2010–2011. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2011b. Data Quality Documentation for External Users: National Ambulatory Care Reporting System, 2010–2011. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2011c. Data Quality Documentation, National Ambulatory Care Reporting System, 2009–2010: Executive Summary (Revised March 2011). Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2011d. National Ambulatory Care Reporting System Manual for 2011-2012. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2012a. Canadian Classification of Health Interventions (CCI): Volume Three – Tabular List. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2012b. Data Quality Documentation, Discharge Abstract Database—Current-Year Information, 2011–2012. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2012c. Data Quality Documentation, Discharge Abstract Database—Multi-Year Information. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2012d. Data Quality Documentation, National Ambulatory Care Reporting System—Multi-Year Information. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2012e. National Ambulatory Care Reporting System Manual for 2012-2013. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2013. Data Quality Documentation, Discharge Abstract Database—Current-Year Information, 2012–2013. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2014a. Data Quality Documentation, Discharge Abstract Database—Current-Year Information, 2013–2014. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2014b. Data Quality Documentation, National Ambulatory Care Reporting System — Current-Year Information, 2013–2014. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2015a. Data Quality Documentation, Discharge Abstract Database—Current-Year Information, 2014–2015. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information). 2015b. Data Quality Documentation, National Ambulatory Care Reporting System — Current-Year Information, 2014–2015. Ottawa: Canadian Institute for Health Information.

CIHI (Canadian Institute for Health Information) and CPAC (Canadian Partnership Against Cancer). 2012. Breast Cancer Surgery in Canada, 2007–2008 to 2009–2010. Ottawa: Canadian Institute for Health Information.

Commission on Cancer. 2002. “Appendix B: Site-Specific Surgery Codes.” In FORDS: Facility Oncology Registry Data Standards. Revised for 2011, p. 247–287. Chicago: American College of Surgeons.

CPAC (“Canadian Partnership Against Cancer”). n.d. Canadian Partnership Against Cancer Corporation. Available at: http://www.partnershipagainstcancer.ca/ (accessed August 17, 2017).

CPAC (Canadian Partnership Against Cancer). 2012.  The 2012 Cancer System Performance Report. Toronto: Canadian Partnership Against Cancer.

CPAC (Canadian Partnership Against Cancer). 2016. The 2016 Cancer System Performance Report. Toronto: Canadian Partnership Against Cancer.

Dragomir A., F.L. Cury, and A.G. Aprikian. 2014. “Active Surveillance for low-risk prostate cancer compared with immediate treatment: a Canadian cost comparison.” Canadian Medical Association Journal Open 2 (2): E60-E68. doi:10.9778/cmajo.20130037.

Fritz, A., C. Percy, A. Jack, K. Shanmugaratnam, L. Sobin, D.M. Parkin, and S. Whelan, eds. 2000. International Classification of Diseases for Oncology, Third Edition. Geneva, Switzerland: World Health Organization.

Gotto, G.T., M.A. Shea-Budgell, and J.D. Ruether. 2016. “Low compliance with guidelines for re-staging in high-grade T1 bladder cancer and the potential impact on patient outcomes in the province of Alberta.” Canadian Urological Association Journal 10 (1-2): 33-38.

Horner, M.J., L.A.G. Ries, M. Krapcho, N. Neyman, R. Aminou, N. Howlader, S.F. Altekruse, E.J. Feuer, L. Huang, A. Mariotto, B.A. Miller, D.L. Lewis, M.P. Eisner, D.G. Stinchcomb, and B.K. Edwards, eds. n.d. SEER Cancer Statistics Review, 1975-2006. Bethseda, Maryland: National Cancer Institute. Available at: http://seer.cancer.gov/csr/1975_2006/ (accessed August 17, 2017). Based on November 2008 SEER data submission, posted on the web site, 2009 (archived).

Hundahl, S.A., I.D. Fleming, A.M. Fremgen, and H.R. Menck. 1998 “A National Cancer Data Base report on 53,856 cases of thyroid carcinoma treated in the U.S., 1985-1995.” Cancer 83 (12): 2638–2648. doi: 10.1002/(SICI)1097-0142(19981215)83:12<2638::AID-CNCR31>3.0.CO;2-1

International Agency for Research on Cancer, World Health Organization, International Association of Cancer Registries, and European Network of Cancer Registries. 2004. International Rules for Multiple Primary Cancers (ICD-O Third Edition). Lyon: International Agency for Research on Cancer. Internal Report No. 2004/02.

Kassouf, W., A.M. Kamat, A. Zlotta, B.H. Bochner, R. Moore, A. So, J. Izawa, R.A. Rendon, L. Lacombe, and A.G. Aprikian. 2010. “Canadian guidelines for treatment of on-muscle invasive bladder cancer: a focus on intravesical therapy.” Canadian Urological Association Journal 4 (3): 168–171.

NCCN (National Comprehensive Cancer Network). 2002. “NCCN Guidelines for treatment of cancer by site: Breast.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

NCCN (National Comprehensive Cancer Newtork). 2013a. “NCCN Guidelines for treatment of cancer by site: Colon Version 3.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp(accessed August 23, 2017).

NCCN (National Comprehensive Cancer Network). 2013b. “NCCN Guidelines for treatment of cancer by site: Rectum Version 4.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

NCCN (National Comprehensive Cancer Network). 2014. “NCCN Guidelines for treatment of cancer by site: Prostate Version 1.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

NCCN (National Comprehensive Cancer Network). 2015a. “NCCN Guidelines for treatment of cancer by site: Bladder Cancer Version 2.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

NCCN (National Comprehensive Cancer Network). 2015b. “NCCN Guidelines for treatment of cancer by site: Non-Small Cell Lung Cancer Version 7.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

NCCN (National Comprehensive Cancer Network). 2015c. “NCCN Guidelines for treatment of cancer by site: Thyroid Cancer Version 2.2.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

NCCN (National Comprehensive Cancer Network). 2016. “NCCN Guidelines for treatment of cancer by site: Small-Cell Lung Cancer Version 1.” NCCN Guidelines. Archived content. Updated content available (by subscription) at: https://www.nccn.org/professionals/physician_gls/f_guidelines.asp (accessed August 23, 2017).

Nenshi, R., N. Baxter, E. Kennedy, S.E. Schultz, N. Gunraj, A.S. Wilton, M. Simunovic, and D.R. Urbach. 2008. “Surgery for colorectal cancer.” In Cancer Surgery in Ontario: ICES Atlas, ed. D.R. Urbach, M. Simunovic, and S.E. Schultz, Chapter 3, p. 53–96. Toronto: Institute for Clinical Evaluative Sciences.

Quan, M.L., N. Hodgson, Przybysz, R., N. Gunraj, S.E. Schultz, N. Baxter, D.R. Urbach, and M. Simunovic. 2008. “Surgery for breast cancer.” In Cancer Surgery in Ontario: ICES Atlas, ed. D.R. Urbach, M. Simunovic, and S.E. Schultz. p. 7–28. Toronto: Institute for Clinical Evaluative Sciences.

Rotermann, M., C. Sanmartin, R. Trudeau, and H. St-Jean. 2015. “Linking 2006 Census and hospital data in Canada.” Health Reports 26 (10): 10–20. Statistics Canada Catalogue no. 82-003-X.

Rotermann, M., C. Sanmartin, G.M. Carriere, R. Trudeau, H. St-Jean, A. Saïdi, A. Reicker, A. Ntwari, and E. Hortop. 2014. “Two approaches to linking census and hospital data.” Health Reports 25 (10): 3–14. Statistics Canada Catalogue no. 82-003-X.

Statistics Canada. n.d.a. Approved microdata linkages. Last updated July 27, 2017. Available at: http://www.statcan.gc.ca/eng/record/summ  (accessed August 16, 2017).

Statistics Canada. n.d.b. Canadian Cancer Registry. Available at: http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3207&lang=en&db=imdb&adm=8&dis=2 (accessed July 2, 2013).

Statistics Canada. n.d.c. Directive on Microdata Linkage. Last updated July 25, 2017. Available at: http://www.statcan.gc.ca/eng/record/policy4-1 (accessed August 16, 2017).

Statistics Canada. n.d.d. Vital Statistics - Death Database (CVSD). Detailed information for 2012. Last updated December 9, 2015.  Available at: http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&Id=257641 (accessed August 16, 2017).

Statistics Canada. n.d.e. Table 103-0550 New cases of primary cancer (based on the July 2016 CCR tabulation file), by cancer type, age group and sex, Canada, provinces and territories, annual (table). CANSIM (database). Last updated January 5, 2018. Available at: http://www5.statcan.gc.ca/cansim/a26?lang=eng&id=1030550&p2=33 (accessed January 24, 2018).

Statistics Canada. 2008. Canadian Cancer Registry System Guide, 2007 Edition. Statistics Canada Catalogue no. 82-225-X. Ottawa: Health Statistics Division, Statistics Canada.

Statistics Canada. 2011. Cancer Incidence in Canada 2008 and 2009. Statistics Canada Catalogue no. 82-231-X. Ottawa: Statistics Canada.

Turner, D., K.J. Hildebrand, K. Fradette, and S. Latosinsky. 2007. “Same question, different data source, different answers? Data source agreement for different surgical procedures on women with breast cancer.” HealthCare Policy 3 (1): 46–54.

Urbach, D.R., M. Simunovic, and S.E. Schultz. 2008. “Technical Appendix.” In Cancer Surgery in Ontario: ICES Atlas, ed. D.R. Urbach, M. Simunovic, and S.E. Schultz, p. 207–212. Toronto: Institute for Clinical Evaluative Sciences.

Winnipeg Regional Health Authority. n.d. Breast Health Centre. Available at: http://www.wrha.mb.ca/community/bhc/Breast-Health-Centre.php (accessed August 17, 2017).