THE PROCESSES OF CARE AFTER COLORECTAL CANCER SURGERY IN ONTARIO. Jensen Tan. A thesis submitted in conformity with the requirements

Transcription

1 THE PROCESSES OF CARE AFTER COLORECTAL CANCER SURGERY IN ONTARIO By Jensen Tan A thesis submitted in conformity with the requirements for the degree of Master of Science (Clinical Epidemiology) Graduate Department of Health Policy, Management and Evaluation University of Toronto Copyright by Jensen Tan 2008

2 Abstract THE PROCESSES OF CARE AFTER COLORECTAL CANCER SURGERY IN ONTARIO Jensen Tan Master of Science 2008 Graduate Department of Health Policy, Management and Evaluation University of Toronto Introduction: Colorectal cancer (CRC) is common in Ontario. This study described the processes of care following CRC resection, and identified CRC relapse from administrative data. Methods: CRC patients aged from with a colorectal resection were identified from the Ontario Cancer Registry. Linked discharge abstracts and physician billings were examined for physician visits, body imaging and endoscopy over the 5 year follow-up period. Administrative codes suggesting disease relapse were compared with patient charts. Results: Overall, 12,804 patients were identified and 8,804 had no evidence of relapse. Most (96.2%) patients had general practitioner follow-up, while 49.3% had medical oncology and 80.4% had general surgery follow-up. Greater than 90% of patients received endoscopy, while only 68.7% of patients received body imaging. Detecting disease relapse was 87.5% sensitive and 93.0% specific. Conclusions: There is potential for improving post-resectional follow-up in CRC patients. It is possible to detect relapse through administrative databases. ii

3 Acknowledgments First I would like to thank Vivian for her support during my research time and Alyssa for being the perfect child. Sincere thanks to my committee members for the time they put into my entire journey: Alex Kiss for his statistical expertise and insight; Steve Gallinger for his content knowledge; Nancy Baxter for her methodologic expertise and attention to detail; David Hodgson for his guidance and support; and last but not least Calvin Law for his mentorship, friendship and building my confidence. To my colleagues for their support and constructive criticisms Nick Daneman, Girish Kulkarni, Laura Rosella, Anand Govindarajan, Karen Devon and Nicole Look Hong. Finally, I acknowledge my PS3 and Guitar Hero 3 for maintaining my dexterity while away from the clinical stream. iii

4 Table of Contents List of Tables vi List of Figures viii List of Appendices ix Introduction 1 Colorectal cancer epidemiology 1 Natural history and management of colorectal cancer 1 Colorectal cancer relapse and management 5 Post-resectional surveillance for colorectal cancer 9 Variation in care after colorectal cancer resection 13 Population based studies on CRC relapse and management 14 Rationale 16 Objectives 17 Methods 18 Data sources 18 Selection Criteria 20 Demographic and Patient related variables 22 Definition of high risk for recurrence 22 Definition of disease relapse 22 Descriptive statistics 23 Preliminary analyses 24 Objective #1: Visits and tests 25 Objective#2: Treatment for disease relapse 26 Secondary Objective: Comparison with primary chart reviewed cohort 28 Data Analysis 29 Results 30 Inception cohort 30 Exclusions 30 Baseline characteristics 31 Preliminary Analyses 36 Objective #1: Visits and tests 37 Objective #2: Treatment for relapse 47 Secondary Objective: Comparison with primary chart reviewed cohort 55 iv

5 Discussion 58 Cohort formation 58 Visits and tests group #1 59 Endoscopic follow-up 61 Body imaging follow-up 62 Visits and tests group #2 64 CRC relapse and treatment 64 Secondary objective: Comparison with chart reviewed cohort 69 Limitations 71 Summary and future directions 74 Appendix 75 References 80 v

6 List of Tables Table A. Tumor-Node-Metastasis definitions, AJCC 6 th ed. Table B. Tumor-Node-Metastasis (TNM) Staging for Colorectal Cancer, AJCC 6 th ed. Table C. Summary of eight randomized trials comparing surveillance strategies after CRC resection. Table 1: Types of CRC by ICD9 diagnosis code Table 2: Number of patients meeting each exclusion criteria Table 3: Pair-wise summary of first year exclusion criteria Table 4: Excluded patients Direct age-sex standardization to Ontario population 1996 Table 5: Baseline characteristics of excluded cohort Table 6: Characteristics of study cohort, eligible for postoperative surveillance Table 7: Total visits and tests over follow-up years 2-5 in patients alive at 5 years without evidence of disease relapse Table 8: Mean number of MD visits or tests per 6 months of follow-up Table 9: Model #1. Univariate and multivariable analysis for having at least one endoscopic examination in follow-up years 1-5, among patients alive and no evidence of relapse at 5 years, who did not receive a total colectomy Table 10: Model #2. Univariate and multivariable analysis for having at least one body imaging modality in follow-up years 2-5, among patients alive and no evidence of relapse at 5 years, stratified by high and low risk primary Table 11: Relapse based on site and high/low risk status of primary, by year of relapse Table 12: Relapse rates and treatments for relapse, by income quintiles Table 13: Relapse rates and treatments for relapse, by LHIN of residence Table 14: LHIN of institution performing liver and lung resections vi

7 Table 15. Model #3. Univariate and multivariable analysis for receiving surgical (lung or liver) resection for CRC relapse, among patients who had evidence of disease relapse in follow-up period Table 16: 2x2 Frequency table for eligibility for follow-up using administrative data compared to the reference standard of primary chart review Table 17: 2x2 Frequency table for relapse detection using administrative data compared to the reference standard of primary chart review Table 18: Location of relapse and surgery for false negatives (relapsed by chart review, but not detected in administrative data) Table 19: 2x2 Frequency table for detection of relapse over all follow-up years 1-5 using administrative data, compared to the reference standard of primary chart review vii

8 List of Figures Figure 1.Algorithm for classification of disease relapse Figure 2. Schematic timeline representing censoring period for measuring test and MD visit frequency in relapsed patients Figure 3. Number of medical oncology follow-up visits over follow-up years 2-5 for low and high risk CRC patients, among those patients with no evidence of relapse Figure 4. Number of general surgeon follow-up visits over follow-up years 2-5 for low/high risk colon and rectal patients, among those patients with no evidence of relapse Figure 5. Number of endoscopy tests in follow-up years 1-5 among patients with no evidence of relapse, who did not receive a total colectomy. Figure 6. Number of body imaging tests in follow-up years 2-5 for patient s with no evidence of relapse Figure 7. Sensitivity analysis: Forest plot of odds ratios and 95% confidence intervals of body imaging frequency, depending on censoring interval prior to first evidence of relapse viii

9 List of Appendices Table A.1 OHIP fee codes used to identify colorectal resectional surgery Table A.2 CIHI procedure codes for colorectal resectional surgery Table A.3 ICD-9 codes for secondary disease Table A.4 Surgery and biopsy codes for lung and liver, OHIP claims and CIHI procedure codes Table A.5 Chemotherapy codes, OHIP claims Table A.6 Physician consultation codes, OHIP claims Table A.7 Imaging and endoscopic modalities, OHIP claims ix

10 Introduction Colorectal Cancer Epidemiology Colorectal cancer (CRC) represents the second highest cancer related cause of mortality in Canada (1). In a report from the National Cancer Institute of Canada (1), it is estimated that for the year 2007, there were approximately 20,800 new cases diagnosed and 8,700 estimated deaths. In Ontario alone, the age standardized incidence rate for CRC is among the highest in the world (2); with 60 cases per 100,000 for men and 41 cases per 100,000 for women (1). The incidence of CRC has been observed to increase with age, with the highest incidence among the age group (1). The anatomic distribution also appears to change with increasing age. As age advances, proximal cancers increase in incidence relative to that of distal (rectal) cancers. However, the cause of shift is not known. A statistically significant reduction of the overall mortality rate from CRC patients in the population of Ontario has been observed (3). For males the CRC mortality rate decreased by 1.3% and for females by 1.7% over 1992 to This decrease in mortality rate is thought to be due to improved awareness in screening, improvements in treatment (particularly with the introduction of newer chemotherapy agents), and increased attention to quality indicators in surgery such as lymph node retrieval. These health services delivered to a patient are known as processes of care (4), which may in turn affect the health status of patients, or outcomes (4). In addition to improving patient outcomes by conferring the appropriate processes of care for the treatment of the primary tumour, there is also a potential for improvement in outcomes by optimizing processes of care after treatment of the primary tumour. Processes of care following CRC surgery can include modalities such as physician visits, and performing surveillance tests such as body imaging to detect disease relapse. To assess the processes of care after CRC surgery, the management and various available treatments for CRC must first be understood. Natural history and management of colorectal cancer Histologically, the vast majority (99%) of CRC are adenocarcinomas. Adenocarcinoma of the colon and rectum originate from benign, adenomatous polyps (5) (although not all adenomatous polyps will develop into cancer). This polyp to carcinoma sequence has been well described, and has become the recognized process of development for most carcinomas. Colonic polyps may be discovered during endoscopy, and are treated by local excision. Sessile, or flat, polyps occasionally are problematic for endoscopic removal, and a segmental colectomy is required to fully excise the lesion. When left untreated or undetected, the 1

11 2 adenoma-carcinoma sequence takes approximately ten years. Hereditary forms of CRC exist where patients may present in their 20s, but the vast majority of carcinomas are sporadic, where patients present in the 6 th decade of life. The gold standard for colonic evaluation is colonoscopy. This allows for biopsy and histological diagnosis of CRC, in addition to completely surveying the entire colon for any synchronous polyps or primary carcinomas. However, in patients that present with obstructing lesions, complete colonic assessment is not possible, and therefore complete colonoscopy is recommended following surgical resection of the primary tumour. Surgical resection The primary curative modality for colorectal cancer is surgical resection. The surgical procedure of choice is dependent on the location of the primary tumor, the presence of other polyps or primary tumours, and whether or not it is considered safe to restore bowel continuity. Commonly, proximal (right-sided) and transverse colon tumours are resected with a right hemicolectomy, while left sided tumors and sigmoid tumors receive a left hemicolectomy or sigmoidectomy, respectively. Proximal rectal tumors may be treated with a low-anterior resection along with a sphincter preserving reconstruction, while most distal rectal tumors are treated with an abdominoperineal resection with a permanent endcolostomy. Important surgical considerations for resection are the establishment of clear resection margins, to ensure no microscopic traces of the primary tumour are left, and the adequate sampling of lymph nodes. As lymph nodes are typically the first tissues to be involved in the development of metastatic disease, adequate lymph node sampling allows for proper staging of the carcinoma. Staging Staging of CRC is important for several reasons. The stage of the tumour allows prognostication of the patient, as higher (or more advanced) stage is associated with a higher risk of disease relapse, thus affecting patient survival. Staging of the primary tumour also helps direct subsequent therapy, as patients with higher stages are recommended to have adjuvant chemotherapy to reduce the risk of disease relapse. Table A and B summarizes the American Joint Committee on Cancer 6 th edition tumor-node-metastasis (TNM) staging of colon cancer (6). This common staging system is based on the depth of invasion of the bowel wall, extent of regional lymph node involvement, and the presence of distant metastatic disease. With increasing stage of disease, the 5 year overall survival decreases from greater than 90% to less than 10% (7).

12 3 Table A. Tumor-Node-Metastasis definitions, AJCC 6 th ed. Primary Tumor (T) TX Primary tumor cannot be assessed Tis Carcinoma in situ T1 Tumor invades submucosa T2 Tumor invades muscularis propria T3 Tumor penetrates muscularis propria and invades subserosa T4 Tumor directly invades other organs or structures or perforates visceral peritoneum Nodal status (N) NX Regional lymph nodes cannot be assessed N0 No metastases in regional lymph nodes N1 Metastases in one to three regional lymph nodes N2 Metastases in four or more regional lymph nodes Distant Metastases (M) MX Presence or absence of distant metastases cannot be determined M0 No distant metastases detected M1 Distant metastases detected AJCC, American Joint Committee on Cancer * Adapted from Greene et al. (6) Table B. Tumor-Node-Metastasis (TNM) Staging for Colorectal Cancer, AJCC 6 th ed. Stage TNM Classification Five year overall survival (%) I T1-2, N0, M0 >90 IIa T3, N0, M IIb T4, N0, M0 IIIa T1-2, N1, M IIIb IIIc T3-4, N1, M0 T (any), N2, M0 IV T (any), N(any), M1 5-7 AJCC, American Joint Committee on Cancer * Adapted from Greene et al. (6) Chemotherapy An important component in the treatment of colorectal cancer is chemotherapy. Due to the risk of developing disease relapse in stage III cancer, current guidelines (8) recommend adjuvant chemotherapy in this setting to improve survival. The foundation of adjuvant chemotherapy agents is flurouracil (5FU). The benefit of 5FU was demonstrated in the 1989 North Central Cancer Treatment Group (NCCTG) trial (9) and 1990 Eastern Cooperative Oncology Group (ECOG) Trial, where significant reductions in recurrence and mortality was observed with administration of 5FU based chemotherapy. Therefore in

13 4 1990, National Cancer Institute consensus guidelines recommended, as the standard of care, 5FU based adjuvant chemotherapy for all patients with resected stage III colon or rectal cancer (10). With regards to chemotherapy for stage II colon cancer, the evidence remains controversial (11). Though all stage II patients are node negative, there is a range of local tumor characteristics (from little bowel wall infiltration to extensive tumors with extramural spread), and differences in adequacy of lymph node assessment and accuracy of nodal staging (12) that may render a patient at higher risk of disease relapse. One further consideration is that although survival may be statistically improved with the administration of adjuvant chemotherapy, the clinical or absolute improvement may be minimal because of the overall lower risk of recurrence compared to stage III disease. Therefore the benefit of adjuvant therapy in this setting may not exceed the potential harms of therapy such as toxicity. In light of this, the Cancer Care Ontario guidelines (8) recommend 5FU-leucovorin for stage III colon cancer, once a day for 5 days, repeating every 28 days, for a usual total of 6 cycles. In total, this amounts to 30 doses of chemotherapy over approximately 6 months. In addition, initiation of chemotherapy is to begin within 5 weeks of resection of the primary tumour. Adjuvant chemotherapy is currently not recommended routinely for patients with stage II colon cancer, although it is acknowledged that it should be considered for patients with stage II disease who have high risk features. Considerations for rectal cancer Due to the anatomic location of rectal lesions, lack of a mesentery, and proximity to other structures such as the bladder, uterus, prostate and sacrum, rectal cancers have a much higher rate of locoregional recurrence than colonic primaries. Preoperatively, the location of the lesion must be precisely defined in relationship with the anal sphincters, as well as the extent of the penetration into the bowel wall and adjacent lymph nodes in the perirectal fat. Therefore, with the goals of downstaging an advanced tumor to increase resectability, and reducing local recurrence rates, preoperative radiation therapy has gained an important role in rectal cancer. The major trial that first suggested the potential for preoperative radiotherapy was the Swedish rectal cancer trial in 1997(13), which was followed in 2004 with the definitive German Rectal Cancer Study Group trial (14). There is an important difference to note in the recommendations for adjuvant chemotherapy between colon and rectal cancer. In contrast to colon cancer, where all stage III and some stage II cancers are recommended adjuvant chemotherapy, all stage II and stage III rectal cancers are recommended adjuvant chemotherapy (10, 15). This is due to the aforementioned anatomic considerations and resultant increase in risk of disease relapse.

14 5 Colon cancer relapse and management Despite the fact that approximately two thirds of CRC patients present with non metastatic (stage I-III) disease, it is estimated that up to half of these patients will develop disease recurrence (16-20). The liver is by far the most common intra-abdominal site of recurrent disease, likely via hematogenous spread through the portal system. It has been estimated that the liver may be involved in up to 71% of patients with disease recurrence (21) and is most frequently the first site of recurrence (20). The most common extrabdominal site for disease relapse is the lung (20, 21). Other commonly observed sites of recurrence include locoregional recurrence (anastomotic, mesenteric, or a second colorectal primary) (20-24), peritoneal, bone, CNS, or adrenal (21). Chemotherapy for CRC relapse The most widely used chemotherapeutic agent for CRC is 5FU, and this also holds true for relapsed (or advanced) disease (25, 26). Combined with the modulating effects of leukovorin, and optimized dosing schedules, response rates of up to 23% and median survivals from 11 to 13 months have been reported for patients with advanced disease (27, 28). Beginning in the mid 1990 s, irinotecan and oxaliplatin were found to have activity against advanced CRC. The impact of irinotecan as a second line therapy was demonstrated in two large randomized phase III studies, including patients who failed initial treatment with 5FU/LV (29, 30). In 1998, Cunningham et al. (29) randomly assigned metastatic CRC patients, who had progression of their disease after 6 months of 5FU treatment, to a course of mg/m 2 irinotecan every three weeks along with supportive therapy or supportive care alone, in a 2:1 ratio. The irinotecan arm had 189 patients while the supportive care only arm had 90 patients. After a median follow-up of 13 months, the irinotecan arm had a significantly higher one year survival (36.2% vs. 13.8%), as well as significantly better quality of life scores. Also in 1998, Rougier et al. (30) randomized 267 metastatic CRC patients who failed first line 5FU treatment to receive either irinotecan ( mg/m 2 every three weeks) versus continuous 5FU infusion. Median survival favored the irinotecan arm (10.8 months vs. 8.5 months). The efficacy of irinotecan or oxaliplatin combined with standard 5FU as first-line therapy for advanced disease has also been assessed in several phase III trials, where median survivals of 14.8 to 21.5 months have been reported (27, 31-35). In addition to new chemotherapeutic agents, the efficacy of using targeted therapy for advanced CRC is being explored. Cetuximab is an antibody targeted to epidermal growth factor receptor (EGFR), which is upregulated in 60-80% of colorectal cancers. Saltz et al. (36) reported a 10% tumour shrinkage was observed when 60 patients were treated with antibody monotherapy. The synergistic relationship between

15 6 cetuximab and irinotecan was confirmed by Cunningham et al (37), where 329 patients with advanced CRC refractory to irinotecan/5fu were treated with cetuximab alone or in combination with irinotecan/5fu/lv (IFL). Median survival improved from 6.9 to 8.6 months with the combination therapy, when compared to antibody monotherapy. Bevacizumab is an antibody against vascular endothelial growth factor (VEGF), thus inhibiting the angiogenesis and vascularity of tumors, potentially improving the delivery of chemotherapy. In combination with IFL, bevacizumab improved median survival from 15.6 to 20.3 months (38). However, despite the improvements in survival from both new chemotherapy agents and targeted agents, the prognosis for all patients with advanced CRC treated with only these therapies remains poor; five-year survivors are seldomly reported. Hepatic resection for CRC metastases Among patients who experience disease relapse, it is estimated that up to half of these patients have the liver as the only site of relapse (39). Of the patients who develop CRC liver metastases, it is estimated that 10-30% may be suitable for hepatic resection (39-42). Over the past 3 decades, a body of research has assessed the survival benefit of hepatic resection for isolated CRC metastases. Historically, until the 1980s, CRC metastases to the liver were often left untreated. Data from this time period demonstrated the poor survival associated with this disease stage, with median survivals of five to ten months (42, 43). Five-year survivors were extremely rare (39, 42). Even in studies examining patients with limited liver metastases who could have potentially been candidates for liver resection, five year survival rates of 2% to 8% were observed (42, 43). Early reports on the success of hepatic resection were met with skepticism (44). In the past, liver resections were also perceived to be associated with a high degree of morbidity and mortality, and the potential benefit for resection of metastatic disease was reserved for carefully selected patients (45). In addition, it was argued that the benefits of resection were biased as there were no adequate controls for comparison (44). However, in the following years, three major factors have improved the outcomes of resection of hepatic metastases. Patient selection has improved through advances in medical imaging. Perioperative and anaesthetic care has also improved through better understanding of physiology, and finally, surgical technique has improved through better understanding of hepatic anatomy (42). It is now recognized that CRC liver metastases are not only potentially curable, but are also optimally managed in a multidisciplinary setting including liver surgeons, medical oncologists, radiologists and

16 7 interventional radiologists. As a result, reported 5 year survival following hepatic resection for CRC metastases have slightly risen from 25-37% in the 1990 s (46-50) to 40% and above since 2002 (41, 51). A 10-year survival rate of about 20% has also been reported for these patients in at least four series with long-term follow-up (41, 46-48), as compared to a negligible 10-year survival rate in patients with liver metastases not treated with resection. The eligibility for patients to potentially receive a hepatic resection for CRC metastases is also increasing, as more aggressive approaches are adopted, such as two stage hepatectomies (52-55) and repeat hepatectomies (39, 56-58). Although there has not been a randomized trial examining the benefit liver resection for CRC metastases, these recent results are clearly superior compared with historical controls, and those with no treatment or received systemic chemotherapy only. Furthermore, it is highly unlikely that a trial comparing surgery versus chemotherapy versus no treatment would ever be conducted, due to lack of clinical equipoise. Recently, the 10 year experience at the University of Toronto for hepatic resection of CRC metastases was reported (41). All patients undergoing liver resection for metastatic CRC were identified over a 10 year period starting in A total of 423 hepatectomies were performed in 395 patients. The primary outcome was overall survival, while the secondary outcomes were disease free survival, perioperative morbidity and mortality. The various hepatic resections ranged from wedge (non-anatomic) resections, to major resections of greater than 4 anatomic segments. Nearly one-third (32%) of patients presented with stage IV disease. Notable results included a median OS of 53 months, along with 1, 5, and 10 year OS of 93%, 47% and 28%, respectively. The median DFS was 19 months, with a 1, 5, and 10 year DFS of 64%, 27% and 22%, respectively. Factors identified as having a poor prognosis included advanced age (>60 years), multiple metastases, lesions greater than 5cm, and a positive margin after hepatic resection. These data were in agreement with other single institution data, and also demonstrates a substantial improvement in survival as compared to a report from an earlier period ( ) at the University of Toronto, where the 5 year OS was 34% (59). A number of factors may explain the improved survival rates, including improvements in surgical technique, critical care, the recognition of other adjunctive modalities such as radiofrequency ablation (60, 61) and peri-operative chemotherapy which may provide an additional survival benefit (62). Recently, the EORTC Intergroup randomized phase II study (EPOC) evaluated the benefit of peri-operative oxaliplatin based chemotherapy for patients with potentially resectable liver metastases, by randomizing 364 patients to either chemotherapy plus surgery or surgery alone. It was reported that peri-operative chemotherapy and surgery for patients with resectable liver metastases confers a statistically significant 7.1% absolute increase in three year progression-free survival (63).

17 8 Since only 10-30% of CRC metastases to the liver may be initially resectable, attention had turned to chemotherapy regimens to improve resectability rates. With recent flurouracil, oxaliplatin, or irinotecan based chemotherapy regimens, previously unresectable hepatic metastases have been successfully downstaged to allow resection (64-68). The survival benefit of such resections have shown to be superior to chemotherapy alone, with 5 year overall survivals of 35-40% (65, 66). With the increasing use of targeted therapies such as cetuximab and bevacizumab, resectability rates are continuing to rise along with a larger proportion of patients eligible to receive potentially curative surgical therapy. Pulmonary resection for CRC metastases Despite being involved in up to 20% of CRC relapses (20, 69), pulmonary relapses are often asymptomatic. The most frequent symptoms of cough and hemoptysis occur in 15-20% of patients, and are usually secondary to the proximity of the lesion to major airways. Other symptoms may also include pneumonia, discomfort, neoplasia or paraneoplastic symptoms (69). Therefore, most pulmonary relapses are detected incidentally on chest imaging. An important consideration is that the differential diagnosis for lung nodule in a patient with a history of CRC may also include a benign neoplasm, granuloma, hamartoma, or even a primary lung cancer (69). Thus, a tissue diagnosis via biopsy is required prior to treatment. This is in contrast to liver relapses, where characteristic CT and ultrasound findings, along with serial imaging, can establish a high index of suspicion for diagnosis. Historically, it has been estimated that a small proportion (1-2%) of CRC relapses are both isolated to the lung and are amenable to surgical resection. The first reports suggesting pulmonary resection for metastatic disease were dated as far as 1927 and 1939 (70, 71). The first reported resection of the lung for CRC metastasis was in 1944 by Blalock (72). As in the case with isolated liver metastases, there have not been any prospectively randomized studies comparing surgical resection to medical management, however there is a growing body of evidence supporting a survival benefit for surgical resection of isolated lung recurrences (73-84). In recent years, the 5 year overall survival rates after resection of CRC lung metastases are from 30-60% (73-77, 85-87). A recent retrospective series by Yedibela et al. (77) reviewed a single institution experience of resecting 153 patients with pulmonary CRC metastases. The range of surgical procedures included wedge resection, segmental resection, lobectomy, bilobectomy, or pneumonectomy. The 5-year overall survival was 37%, with a median survival time of 39 months. When the analysis was performed on patients with a curative operation (R0, microscopically free of residual disease), the 5-year overall survival was 39% with a median survival time of 43 months. Although not universally agreed upon, poor

18 9 prognostic factors associated with pulmonary CRC relapse resection may include multiple lesions, a rising CEA, higher stage of the primary tumor, and lymphatic involvement of the relapsed lesion (76-78, 82, 85). There have also been several small series examining the survival following the resection of both hepatic and pulmonary CRC relapses in selected patients (88-91). Similarly, in these patients, it has been shown that resections of both organs are safe, and confer a potential survival benefit. In a recent report documenting the experience at the University of Toronto (89), 39 patients underwent both lung and liver resections for metastatic CRC from 1992 to Eleven of these patients had synchronously identified metastases and underwent staged resections of the primary CRC and metastases. The remainder underwent sequential resections of the colon, lung and liver. The median survival following the final resection of metastasis (either in the liver or the lung) was 42.2 months. Furthermore, it was observed that the median overall survival from the initial diagnosis of colon cancer was nearly ten years (117 months). This was a relatively small series of patients, however the results are in agreement with other studies; 5 year overall survival following resection of both lung and liver recurrences ranged from 30-60%, with median survivals from the time of the last resection of metastasis of months (88-91). Post-resectional surveillance for Colorectal Cancer Due to the advent of newer chemotherapy such as oxaliplatin or irinotecan-based regimens, targeted molecular therapies such as cetuximab and bevacizumab and the availability of potentially curative surgery in the treatment of disease relapse, effort has been directed to establish the role surveillance after resection of the primary tumour. The goal of surveillance is three-fold: to monitor for the development of second primary tumors of the colon and rectum, to detect tumor recurrence at a stage where potentially curative surgical therapy may be used, and to provide psychosocial support to the patient (19). As approximately 80% of disease relapse occurs within the first 3 years after the primary CRC resection, and over 90% within 5 years (92, 93), surveillance during this period is particularly important. However, randomized trials of surveillance strategies have had conflicting results likely due to the heterogeneity of follow-up strategies employed by the trials, and there is controversy as to the optimum strategy to follow. As a result, there have been conflicting recommendations on how follow-up should be carried out (16, 94-98). In addition, practice patterns and physician attitudes towards CRC follow-up are also variable (19). These variations may stem from the fact that many relapses are first manifested by new symptoms by patients (99), and perhaps the yield of any extra surveillance tests are low for finding additional treatable recurrences. Modalities for surveillance commonly include periodic history taking

19 10 and physical examinations via physician follow-up visits, combinations of laboratory tests (eg. CEA, liver function, complete blood counts, fecal occult blood), diagnostic imaging (ultrasound, computed tomography, magnetic resonance imaging), or endoscopic surveillance (colonoscopy, sigmoidoscopy). Since 1990, there have been at least seven studies comparing different high intensity versus low intensity surveillance strategies following curative CRC resection (99-105). The findings from these studies are summarized in Table C.

20 11 Table C. Summary of seven randomized trials comparing surveillance strategies after CRC resection. Trial Patients/Intervention Summary Kjeldsen et al. (101) Makela et al. (102) Ohlsson et al. (99) Pietra et al. (105) Rodriguez et al. (103) Secco et al. (104) Schoemaker et al. (100) Control (n=307): Clinical assessment, Hb, ESR, LFTs, CXR, FOB, colonoscopy +/- DCBE at 5,10,15 years Intervention (n=290): Same tests as control, except q6mo x 3 years, then at 4, 5, 7.5, 10,12.5, 15 years Control (n=54): Clinical assessment q3mo x 2 years, then q6mo x 3 years, including CBC, FOBT, CEA, CXR. Rigid sigmoidoscopy for rectal/sigmoid patients, barium enema q12 mo. Intervention (n=52): Same frequency and testing as control arm, except q12 mo colonoscopy, flex sig. for rectal/sigmoid patients q3mo, liver US q6 mo, abdominal CT q12mo Control (n=54): FOB q3mo x 3 yrs then q12 mo; clinical assessment when symptoms arise. Intervention (n=53): Clinical assessment, rigid sigmoidoscopy, CEA, LFTs, FOBT, CXR q3mo x 2 years, then q6 mo Endoscopy assessment of anastomosis (flex sig, colonoscopy) at 9, 21, 42 mo Complete colonoscopy at 3, 15, 30, 60mo CT pelvis at 3, 6, 15,18, 24 mo (if APR) Control (n=103): Clinical assessment, CEA, liver US q6 mo x 1 yr, then q12 mo. CXR and colonoscopy q12 mo. Intervention (n=104): Clinical assessment, CEA, liver US q3mo x 2 yr then q6 mo x 3 yr, then q12mo. CXR, abdominal CT, colonoscopy q12mo. Control (n=132): Clinical assessment, CBC, LFTs, CEA q3mo x 2 yr then q6mo x 3 yr Intervention (n=127): Clinical assessment, and lab tests as control arm Abdominal CT or US q6mo x2yr, then q12mo CXR q12 mo, colonoscopy q12 mo Control (n=145): Assessment by telephone q6mo. Clinical assessment by family physician q12 mo or when symptoms arise. Intervention (n=192): High risk pts: Clinical assessment and CEA q3mo x 2 yrs, q4 mo x 1 yr, then q6 mo x 2 yr Abdominal and pelvic US q6mo x 3 yr then q12 mo x2 yr. Rigid sigmoidoscopy and CXR q12 mo for rectal pts. Low risk pts: Clinical assessment and CEA q6mo x 2 yrs, then q12mo x 3 yrs. Abdominal and pelvic US q6mo x 2 yr then q12 mo x 3yr. Rigid sigmoidoscopy q12mo x 2 yr then q2yr and CXR q12mo for rectal pts. Control (n=158): Clinical assessment, CBC, LFTS, CEA, FOBT q3mo x 2 yrs, q6mo x 5 yrs Intervention (n=167): Clinical assessment and lab tests as per control arm, plus CXR q12mo, CT liver q12 mo, colonoscopy q12mo. Recurrence rates equal among both arms (26%), but those in intervention arm were more frequently asymptomatic, diagnosed 9 months earlier, with more having surgery for curative intent. However, overall and disease specific survival was similar in both arms. Recurrence rates equal among both arms (41% overall). Intensively followed patients were diagnosed with recurrence 5 months earlier. Endoscopy and US were beneficial but CT failed to improve diagnostic rate. However, cumulative 5 year survival was similar between both arms. Recurrence rates equal among both arms (32-33%). No patient underwent surgery for distant metastases. Cumulative 5 year survival rates for the control arm was 67% and the intervention arm 75%, which was not statistically significant. Recurrence detection rates in both arms were not statistically different (20% and 26%). Distant metastases were resected at similar rates in both arms. 5 year cumulative (53.3% vs 73.1%) and disease free survival (53.3% vs. 68.2%) was statistically different among both arms. Overall, the data supported an intense follow-up regimen, at least for rectal cancer patients due to the higher rates of local recurrence. There was no statistical difference in the rate of tumor recurrence in either arm, time to relapse, or type of recurrence. There was a higher proportion of resectable tumor recurrence in the intervention arm, and remained significant after adjusting for potential confounders. Overall, there was no difference in overall survival, but a survival benefit was observed for those with stage II disease and rectal tumors. Recurrence rates were similar among both arms (52.6%, 57.2%). Patients in the intervention arm at high risk had a significantly higher number of curative reoperations. Actuarial 5 year survival of all patients in intervention arm was significantly better than those in the control arm There was no statistically significant difference in mortality rates between the two follow-up schedules, overall, by stage, or by tumor location. Although more patients had asymptomatic liver recurrences detected with intense surveillance, a similar number of patients in both arms received a liver resection.

21 12 Recently, six (99-103, 105) of the aforementioned seven trials were analyzed in a meta-analysis (92). Although limited by the heterogeneity of the included trials, their inclusion provided improved power to detect statistical differences among different CRC surveillance modalities. This analysis observed an overall survival benefit at 5 years for more intensive follow-up (OR 0.73, 95%CI ), although as expected, the number of recurrences in both arms were similar (OR 0.91, 95%CI ). There was a mortality benefit for performing more tests compared to fewer tests (OR 0.64, 95%CI ), and liver imaging (OR 0.64, 95%CI ). In the more intensively followed arm, there was also a significantly higher proportion of patients who had curative surgical procedures attempted (OR 2.41, 95%CI ). The weighted mean differences for the time to recurrence was also significantly smaller in the intensive surveillance arm, with an observed reduction of months (95%CI ). However the conclusion of this meta-analysis emphasized that it was not possible to determine from the data the best combination or frequency of physician visits and surveillance test that would yield the optimum follow-up strategy, nor was it possible to evaluate costs or potential harms of intensifying follow-up. Despite these limitations, in the context of the controversy surrounding whether there is any clinical benefit in intensifying CRC surveillance, this is among the best evidence to date in support of intensifying follow-up. The current guidelines issued by Cancer Care Ontario (94, 106) are also based on a meta-analysis of similar studies as the Cochrane review (99-102, 104, 105). The findings of this study were similar: there was a significant improvement in survival with more intensive follow-up strategies (relative risk ratio 0.80, 95%CI ), and despite a similar number of recurrences in both arms, asymptomatic recurrences and reoperations for cure of recurrences were more common in the more intensively followed arm. At the conclusion of the study, it was recommended that for patients at high risk of disease recurrence (stage IIb, III), clinical assessment should be performed at least every 6 months (or when symptoms occur) for the first 3 years of follow-up. At the time of these physician visits, patients may have blood CEA, liver imaging, or chest radiography. For patients at lower risk of recurrence (stage I, IIa), or with comorbidities that preclude surgery for metastatic disease, physician visits should be performed at least yearly, or when symptoms occur. With regards to colonoscopy, all patients should receive a follow-up colonoscopy within 6 months of curative resection (if a preoperative colonoscopy was not performed), or else repeated every 3-5 years if no adenomas are found. It is noted that these guidelines do not mandate liver imaging, but suggest that they may be a part of the follow-up regimen. In contrast, follow-up guidelines from the American Society of Clinical Oncology [ASCO] (96) mandate an annual CT scan for the first three years following primary CRC resection.

22 13 There has also been one randomized controlled trial comparing general practice versus surgical-based follow-up for patients with colon cancer (107). A total of 203 patients were randomized to either a general practitioner or surgeon-led follow-up, with identical guidelines given to both groups. It was acknowledged that clinicians in either setting were not compelled to adhere to the guideline, and crossover between the two arms was allowed. The recommended follow-up regimen included 3 monthly visits for the first 2 years, followed by visits 6 monthly. In addition to history and physical examination, diagnostic tests included annual FOBT and colonoscopy every 3 years. Primary outcomes were quality of life, depression and anxiety, and satisfaction. Secondary outcomes were the number and type of investigations, number and time to detection of recurrences, and deaths from all causes after 2 years into the study. There was no significant difference in quality of life, depression or anxiety, or satisfaction between the two arms. However, patients in the GP-led arm were more likely to have FOBT, while those in the surgeon arm were significantly more likely to have more imaging (ultrasound) or endoscopic (colonoscopy) surveillance. The study was not powered to detect differences in death or recurrence rates, and despite a lower time to recurrence and death rate in the surgeon led group, there was no statistically significant difference between the two arms. Variation in care after colorectal cancer resection With respect to surveillance following curative colorectal resection, variations and disparities have been identified. Rolnick et al. (108) examined a cohort of 881 non-metastatic resected CRC patients in the United States, and showed that African-American and elderly patients were less likely to receive colonic endoscopy modalities at 1, 3 and 5 years following primary CRC resection. Similar results were found in a study of SEER-Medicare patients, where elderly African Americans were less likely to receive colonic surveillance after primary CRC resection, despite adjusting for sociodemographical, hospital, and clinical characteristics. Cooper et al. (109) examined the patterns of postoperative endoscopic follow-up among SEER-Medicare patients, and showed that only 51% of patients received an endoscopic exam in the follow-up period, along with geographic variations in practice patterns. In another study by Cooper (110), further tests were examined among SEER-Medicare CRC patients, including liver enzymes, chest x-rays, and CT scans. It was observed that rates of testing were lower with increasing age, less comorbidities, and there was significantly variability among different SEER regions in the use of these tests, ranging from 1.5 to 3.6-fold. Attitudes towards CRC follow-up among Ontario physicians were examined in a recent survey by Earle et al (19). In this study, all Canadian radiation oncologists, medical oncologists and surgeons specializing in CRC were assessed for their follow-up recommendations for a hypothetical patient (a 50 year old healthy

23 14 man with stage III disease after a curative CRC resection). Out of the 160 physicians who completed the survey, the majority recommended clinical assessment every three to four months for the first two followup years. The recommended follow-up frequency decreased with each year until year 5, where 45% recommended 6 monthly and 45% recommended yearly visits. Beyond 5 years of follow-up the majority recommended no follow-up visits. Less than one third of the physicians recommended body surveillance by ultrasound imaging for any follow-up year, and less than 10% recommending abdominal CT in any follow-up year (including year 1). In contrast, about 90% of physicians recommended bowel surveillance with colonoscopy in follow-up year one, decreasing to 65% of physicians recommending it in year 5. There were other notable results from this study. The majority of physicians surveyed (64%) stated that they routinely discharged patients to their primary care physician for follow-up, although a similar proportion (66%) agreed that alternating follow-up between primary care physician and specialists was appropriate. Regarding attitudes towards follow-up, 65% thought that finding local recurrences improved survival, while 50% felt that specialists were more efficient (eg. less likely to order unnecessary tests) at providing follow-up care than primary care physicians. Population-based studies on CRC relapse and management By examining population based data, one may be able to examine treatments and outcomes in the absence of the bias introduced by single center series. In addition, examination of a large number of patients in a population allows examination of non-clinical, health services related variables which may have an effect on treatment and outcome. These variables may include geographic region, socioeconomic status, ethnic origin, proximity to academic center, or urban versus rural place of residence. Although clinical details are often not available because such studies often examine administrative data, population-based studies can be a rich source of descriptive information for a population of interest, and can generate numerous hypotheses to base future studies upon. Several groups have attempted to describe CRC relapse, and its management on a population level, especially with respect to the use of hepatic resection. Leporrier et al (111) examined patients with CRC from the Digestive Cancer Registry of Calvados, France, from 1994 through Of 1,315 patients examined, 358 developed hepatic metastases, where 17.3% received a surgical resection, 40.2% were treated with palliative chemotherapy, while the remaining 42.5% were treated symptomatically. For the overall group, median survival was 10.7 months, while the 5 year OS was 14%. When analyzed separately, the group with hepatic metastases who were surgically resected had a median survival of 44.7

24 15 months, those treated with palliative chemotherapy had a median survival of 13.5 months, and those treated symptomatically had a median survival of 4.5 months. A logistic regression analysis was performed to attempt to identify factors associated with the use of surgical resection for hepatic metastases. Significant clinical parameters negatively associated with hepatic resection included age greater than 75 (OR 0.14, 95%CI ), 2-3 hepatic metastases (OR 0.28, 95%CI ) or >3 hepatic metastases (OR 0.02, 95%CI ), or size of largest hepatic metastasis > 50mm (OR 0.47, 95%CI ). Non-clinical parameters evaluated included the place of treatment (academic vs nonacademic center), place of residence (urban vs rural) and diagnosis period ( vs ), however the only significant factor with a positive association with hepatic resection was recent period of diagnosis (OR 2.03, 95%CI ), likely reflecting the increasing recognition of its potential survival benefit. Other non-clinical, population level factors such as socioeconomic status and geographic region were not assessed. Manfredi et al. (112) also examined a population-based cancer registry in France from 1976 to 2000, examining the incidence, treatment and prognosis of CRC liver metastases. Of the 13,463 patients with CRC, 14.5% had hepatic metastases discovered during diagnostic workup or during the course of treatment, with 76.8% confined to the liver. Resection for cure was performed in 6.3% of these patients. For 3,655 patients with follow-up data, 12.8% developed liver metastases over the 5 years following diagnosis, and 16.9% of patients were resected for cure. The proportion of patients treated with palliative chemotherapy increased with time, while those treated symptomatically decreased. Multivariable analyses to determine factors associated with resection for cure or survival did not include any nonclinical factors such as geographic region, socioeconomic status, or treatment at academic center. There has been one population based study that suggests the potential role for imaging, follow-up routines and multidisciplinary management to improve outcome in patients with CRC hepatic metastases. Sjovall et al (113) examined 2,280 CRC patients in Sweden from 1996 through 1999 for their treatment and outcome, and also analyzed their imaging to evaluate resectability. Hepatic metastases were discovered in 537 patients (24%), where 266 patients had disease isolated to the liver. However, a hepatic resection was performed in only 21 patients (4% of those with liver metastases). By re-evaluation of liver imaging of these patients, it was determined that less than 50% of immediately potentially resectable patients received a resection. By using reported rates of salvageability with chemotherapy for non-resectable patients combined with eligible patients who did not receive resection, it was estimated that up to 17% of those with liver metastases could have been treated with potentially curative surgery. This finding suggests that in addition to the clinical factors associated with surgical resection that were explored in

25 16 other series, there may also be other non clinical health services related factors that may be associated with the use of surgery in the management of relapsed disease.

26 Rationale For the population of Ontario, the processes of care and outcomes of patients who have received curative resection for CRC have not been previously described. These data are important in order to establish the current practice patterns in Ontario, as well as the outcomes that result from them. Such a description may reveal potentially modifiable or addressable processes of care that a patient receives following CRC surgery. For example, there may be variations among different Ontario regions, or patients in different socioeconomic groups for receiving potentially curative surgery for disease relapse, but there is no data existing to determine whether or not such variation may be appropriate (eg. if a LHIN had an unusually large proportion of high risk patients, that LHIN may appropriately have a higher rate of potentially curative surgery). As Ontario has recently regionalized health care into Local Health Integration Networks (114), it is particularly important that individual needs or deficiencies in regions are identified such that further research may be performed to elucidate the details of the cause and what is needed to address them. An initial but detailed descriptive study of the Ontario population would help establish future research priorities for processes of care in CRC follow-up. A detailed descriptive study may also provide information about the potential denominator of patients who may be eligible for surgical treatment of relapse. By comparing the proportion of patients receiving surgery to expected proportions from the literature, potential differences can be identified and investigated. Previous population-based studies have not attempted to examine the relationship between processes of care following primary CRC resection and the treatment of relapse. Preliminary analytical models may be constructed from descriptive data that may provide further evidence to support guidelines for surveillance, and generate further hypotheses to explore. Finally, there is a need to develop a validated method for identifying CRC relapse in administrative data. As administrative data in Ontario is readily available and population based, the potential for using this algorithm in future studies is substantial. This may include future studies that evaluate measures of quality, or that assess population health effects from knowledge translation initiatives. 17

27 Objectives Primary objectives 1) To describe the processes of care that patients receive in the follow-up period, including MD visits and surveillance tests 2) To describe the proportion of patients that had evidence of disease relapse, who receive surgical treatment in the management of relapse, with stratification by LHIN and income quintile Secondary objectives 1) To explore potential factors associated with the receipt of body imaging or endoscopy in the follow-up period after CRC resection 2) To explore potential factors associated with the use of surgery in the management of relapsed disease 3) To assess the sensitivity and specificity of classification of disease relapse and high risk disease in the study cohort by using a primary chart reviewed cohort of CRC patients Hypotheses It is hypothesized that greater than 80% of Ontario CRC patients will receive surveillance colonoscopy at least once in their 5 year follow-up period, as well as at least one body imaging modality. Most patients will receive follow-up from their general surgeons on average once per year. For disease relapse, it is hypothesized that there will be no significant association between LHIN or income quintile and the rate of relapse. However, there will be statistically significant differences in the proportion of patients who receive surgical treatment for relapse, across LHINs and income quintiles. Among the exploratory analyses, it is hypothesized that income quintile will be a determinant for the receipt of both body imaging and endoscopy. The main predictors for the use of surgery will be age and comorbidity, as well as the frequency of body imaging that the patient received in follow-up. It is hypothesized that the classification of relapse used in the study will yield both a specificity and sensitivity rate of greater than 80%. Furthermore, the classification of high risk disease will capture greater than 90% of stage III patients and 30% of stage II patients. 18

28 Methods A retrospective cohort study was performed at the Institute for Clinical Evaluative Sciences (ICES). The sources information included the Ontario Cancer Registry (OCR) and the following linked administrative databases: the Canadian Institutes of Health Information Discharge Abstract Database (CIHI-DAD), the Ontario Health Insurance Plan (OHIP), the ICES physician database (IPDB), the Registered Persons Database (RPDB). Research ethics board (REB) approval for this study was obtained from the Sunnybrook Health Sciences Center and the University of Toronto. For secondary analysis of the externally collected, chart reviewed cohort of CRC patients from Princess Margaret Hospital and Credit Valley Hospital, REB approval was obtained from the University Health Network and Credit Valley Hospital and permission of Dr. Lillian Siu. Data sources The Ontario Cancer Registry The OCR collects information on all incident cancers in Ontario. The OCR is operated by the Ontario Cancer Treatment and Research Foundation (OCTRF), which was established in 1943 by the Cancer Act. Via subsequent amendments to the Cancer Act, hospitals are required by law to report incident diagnoses of cancer to this registry, where information such as tumor characteristics and some patient demographics are recorded (115, 116). There are four major data sources from which the OCR abstracts information (115, 116). The first source is by hospital discharge abstract summaries with a diagnosis of cancer. These hospital abstracts are summarized, forwarded and recorded into CIHI-DAD via the Ministry of Health. From this point the record is forwarded to the OCR. These hospital abstracts may contain up to 16 diagnoses, from which one may be that of cancer. The second source of information is via pathology reports. Starting in 1973, all pathology laboratories in Ontario were required to submit reports where cancer was mentioned, and compliance reached 100% in These reports are submitted directly to the OCR, where the data is coded and abstracted. The third major source of data is death certificate information. This is achieved via special arrangement with the Office of the Registrar General of Ontario, where all deaths are reported and the underlying cause of death is determined. When the underlying cause is determined to be cancer, the case is reported to the OCR. Additional information may be sought by the OCR (data requested from health institutions, etc) to complete the record, if corresponding cancer information cannot be found in any of the other data sources (ie. the case is a death-certificate-only (DCO) case). The fourth major source of data is direct reporting from regional cancer centers (RCC s) and the Princess Margaret Hospital (PMH). These cancer centers record tumor registry data and forward 19

29 20 them directly to the OCR for data abstraction. The records in OCR are indexed by an encrypted OHIP number (IKN, ICES key number), each representing an individual patient. Administrative databases The CIHI-DAD is an administrative database consisting of discharge abstracts of each hospital stay for every patient in Ontario, regardless of disposition (patient went home, died, transferred to another hospital). The main data elements include patient demographics (eg. sex, date of birth, postal code, county and residence code), clinical data (eg. diagnoses, procedures performed), and administrative data (eg. institution number, admission category, length of stay, disposition). The OHIP database consists of all physician billings submitted to OHIP. Information in an OHIP record includes the date of service, fee code (as per the OHIP Schedule of Benefits), and the billing physician s number (corresponding to the IPDB). The IPDB is a database updated yearly, consisting of a unique number for every physician in Ontario. Each IPDB record also contains the physician s self-reported main specialty. Finally, the RPDB consists of basic demographic information about anyone who has ever received an Ontario health card number. This data is supplied by the Ministry of Health and is enriched with other in-house data at ICES. There is one record for each health card number issued, and consists of date of birth, sex, geographic information on the place of residence, and if applicable, the date of death. All databases are indexed by an ICES key number (IKN), a unique, encrypted identifier for individual patient records that allows for linkage across the various data sources. Validity The use of administrative databases results in a large sample size, and often provides statistical power where single institution series cannot. Furthermore, the bias introduced by selection of patients by single centers is reduced. However, prior to using the data, it must first be ascertained whether these administrative databases are dependably and accurately coded. By confirming the validity of the data used, a higher degree of validity may be conferred to any conclusions drawn from the study of this data. The completeness of detection of cancer, or case ascertainment, of the OCR has been evaluated as described by Robles et al. (117). Two methods were used to estimate completeness of registration: mathematical modeling using three data sources, and using two data sources with capture-recapture methods. It was found that the two methods were similar in their estimates of case ascertainment, where at most there was a 7% difference in estimate by cancer site. Overall, case ascertainment was greater than 90% for each of the 16 cancer sites.

30 21 Notably, variables describing the tumor stage (primary tumor size, extent, nodal status, and metastasis) are not directly recorded into the OCR database. Staging of a tumor is very important clinically, as this is what guides subsequent treatment of the tumor, and is used for prognostication. These parameters are also important from a health services perspective, in order to aid researchers in determining whether or not the patient received appropriate treatment. Although these data are potentially available, they can only be accessed from the individual archived pathology reports that were submitted on cancer diagnosis. With respect to CIHI-DAD, the validity of this database has been assessed in a recent reabstraction study (118) as well as a previous inventory of seven reabstraction studies (119). Overall, less than three percent of Ontario health records are missing demographic data. The agreement between the database and the hospital chart for admission and discharge date, sex, birth date, admission source, and discharge destination is greater than 97%. For principal procedure code there is 88%-96% agreement, and 100% for death. However, there is less agreement when examining diagnoses, with 81% agreement for most responsible diagnosis, and a greater than 60% false negative rate for the coding of specific comorbid conditions and complications. These figures suggest that using these data will allow a high sensitivity and specificity for detecting principal procedures such as surgical resection. However, with a high false negative rate for coding specific comorbid conditions, there is a risk of under-detection of diagnoses such as metastatic cancer. To the knowledge of the author, there are no validation studies of the other administrative databases used. Selection Criteria Patients aged 18 to 80 years, with an ICD-9 diagnosis of colon or rectal cancer ( , , 154.0, 154.1), and an ICD-0-2 histology code of adenocarcinoma (8140, 8141, 8143, 8144, 8145, 8147, 8210, 8211, 8220, 8221, 8260, 8261, 8262, 8263, 8430, 8440, 8480, 8481, 8490, 8510, 8550, 8551, 8560, 8562, 8570, 8571, 8572, 8573, 8574, 8575, 8576) were identified from the OCR from January 1, 1996 through November 30, Patients with a diagnosis of any other neoplasm at any time based on OCR records were excluded in order to remove the confounding effect of receipt of surgery or chemotherapy for any tumors other than the colon cancer. Patients who had colorectal resectional surgery were identified by searching for linked OHIP billings and CIHI-DAD procedures (Appendix Tables A.1, A.2), starting from 14 days prior to the diagnosis date. This 14 day window allowed the identification of the index colorectal resection for patients who did not have a preoperative diagnosis of CRC (eg. those who presented with an obstruction, perforation). The service date for the OHIP claim for resection took precedence for the CIHI admission date, in the case where both an OHIP claim and CIHI record for resection was found. Either one OHIP claim or one CIHI code for CRC resection was considered

31 22 sufficient for classifying a patient as having received a primary CRC resection. Patients who did not receive a colorectal resection, or who received a colorectal resection greater than 120 days following diagnosis were excluded. These two exclusions were applied because the population of interest was those who received a potentially curative colorectal resection for CRC; patients undergoing resection more than 4 months from diagnosis were not likely for curative intent. As the 90 th percentile for Ontario wait times in those patients with gastrointestinal cancer (including colorectal cancer) was less than 70 days (120), this 120 day cutoff was conservative. In order to identify a cohort of patients likely resected for cure and potentially disease-free following primary colorectal resection, a series of exclusion criteria were applied based on data over the first year after resection. This one year interval from the CRC resection permitted patients to complete multidisciplinary treatment for the primary tumor (ie. the administration of a course of adjuvant chemotherapy). Therefore, patients who had evidence of relapse in the first year were also excluded based on these criteria. The exclusion criteria applied during the first year following CRC resection were: Death within the first year after colon resection A diagnosis of advanced (secondary) disease code (Table A.3) within CIHI diagnosis codes in hospital admissions Evidence of early relapse of disease via lung or liver procedure (resection, destruction or biopsy) (Table A.4) A palliative care consult (OHIP fee codes A945, C945, W982, C982, W882, C882, W872, W972, K023, K998, K996) First claim for chemotherapy (Table A.5) greater than 120 days following primary colon resection in unpublished data (M. Kryzanowska, verbal comm.), over 90% of patients will initiate a course of adjuvant chemotherapy within 4 months of the primary colon resection. Initiation of chemotherapy later than this (but within the first year) is atypical and suggests treatment of early disease relapse Days between first and last chemotherapy claim exceeding 270 days, between days 0 and 395 after CRC resection. The usual duration of a course of adjuvant chemotherapy is 6 months. Allowing time for breaks in treatment due to side effects or delays, a duration of chemotherapy greater than 270 days likely represents ongoing treatment for metastatic or relapse of disease. For

32 23 this exclusion criterion, the first month of the second follow-up year was examined to accommodate the definitions of late start and prolonged duration. Finally, patients with missing LHIN of residence, or income quintile were excluded. Those patients residing in Southeast LHIN were also excluded due to the alternate funding plan employed by this LHIN, which affected the completeness of OHIP billings. Demographic and patient related variables Patient variables collected included sex, age at diagnosis of CRC and comorbidities at the time of CRC resection. The comorbidity of a patient was expressed as a Charlson score (121), with the Deyo modification (122) for use with ICD-9-CM diagnosis codes. The Charlson score is a weighted score based on 16 diagnostic categories, and was designed to predict 1 year mortality. Although originally designed based on a cohort of breast cancer patients, the Charlson score has been validated for CRC patients as well (123) as a predictor for mortality. The Charlson comorbidity was calculated excluding the index admission for CRC resection. Hospital admissions for two years prior to the index admission were examined for primary and secondary diagnosis codes. The overall Charlson score was classified as indicating low (0 to 1), medium (2) and high (greater than 2) comorbidity. Since every patient in this study had cancer, the two diagnostic categories referring to neoplasm or metastatic disease were not included in the calculation of the Charlson score. Patients who relapsed also had a Charlson score calculated from the date of relapse (minus the index date for relapse, with a look-back period of two years). Patients were classified as having received a total colectomy if their original CRC resection was associated with CCP code 57.6, or OHIP fee codes S169, S170, S172-S174. Demographic variables collected included the LHIN of the patient s residence, as determined through the RPDB, and the LHIN of the institution performing the colon resection. The mean neighborhood income quintile was derived from the RPDB, postal codes and census tract information (census year 2001). Rural status was based on the StatsCan definition, defined as residence in a community size of less than 10,000, based on the 2001 census. Definition of high risk for recurrence The receipt of adjuvant chemotherapy (at least one claim within 120 days of CRC resection) was used to represent disease that was at high risk for relapse. The risk for relapse is usually represented by disease stage, where higher disease stage was associated with a higher likelihood of disease relapse. However, there was no variable representing tumor stage within OCR for the time period of this study. Since

33 24 adjuvant chemotherapy is recommended for stage III and selected stage II patients, this was chosen as a proxy for stage in the absence of this variable in the administrative data. Definition of disease relapse As there was not an explicit variable within the administrative databases that represents disease relapse, the event of disease relapse was represented by several variables: the appearance of a diagnosis code for advanced disease (Table A.3), a liver or lung procedure (resection, destruction or biopsy) (Table A.2), the new administration of chemotherapy (A.5), or new consultation to palliative care (OHIP fee codes A945, C945, W982, C982, W882, C882, W872, W972, K023, K998, K996) (Figure 1). Lung or liver destruction that were not clearly definitive resections or that were biopsies were classified as other procedure. The date of relapse was defined as the earliest date of any of the aforementioned events. Surgical resection for relapse was defined as a formal lung or liver resection for disease relapse (Table A.4). Patients who died with a cancer related cause of death code (ICD ) from the OCR were classified as having disease relapse, but with an unknown date of relapse. Figure 1. Algorithm for classification of disease relapse Descriptive statistics Prior to applying exclusion criteria, the type of CRC of each patient in the inception cohort was described via ICD-9 code. Descriptive statistics were generated for the excluded cohort and the final study cohort. Continuous data were expressed as means and standard deviations, while categorical variables were expressed as counts and percentages. When applicable, continuous data were compared using the

34 25 Student s t-test, while categorical data were compared using the chi-square test. To better understand why patients were excluded in their first year, counts of patients having at least two exclusion criteria were summarized in a table of pair-wise combinations of exclusion criteria. The exclusion rates by LHIN were also examined by direct standardization against the 1996 Ontario population by age (18-50, 51-60, 61-70, 71-80) and sex. Univariate descriptive statistics included patient demographics (age, sex), primary cancer type (colon or rectum), high or low risk primary (represented by adjuvant chemotherapy), Charlson comorbidity classification, LHIN, income quintile, and the time period of the study ( , ). The study cohort was divided into four groups: 1) those who were alive and had no evidence of relapse at the end of the follow-up period, 2) those who had evidence of relapse within the follow-up period, 3) those who had no evidence of relapse but died within the follow-up period, and 4) those whose only evidence of relapse was a cancer related cause of death (ie. classified as relapse but the date of relapse is unknown). Only the former two groups were used in the analysis of visits and tests, because patients in the latter two groups either had an incomplete disease-free follow-up time, or an unknown date of relapse. Preliminary analyses Following the application of the patient selection criteria, the final study cohort of patients were likely disease free and eligible for post operative follow-up in years two through five (the follow-up period ). A series of preliminary analyses were performed to establish data consistency. For all colon, liver and lung resections, dates between OHIP claims and CIHI procedures were assessed for agreement. When the documentation of a resection was from an OHIP claim only, all CIHI admissions 7 days prior and 7 days following the claim was searched to possibly document the setting in which the OHIP claim was made. Likewise, when the only documentation of the procedure was via CIHI procedure code, all OHIP claims within 7 days of the admission were examined to document the activity surrounding the admission. The diagnosis codes for all hospital admissions with lung and liver resections were also evaluated to determine whether the resection was performed for metastatic disease. For patients who received adjuvant chemotherapy, the mode, median and interquartile range of the interval from the date CRC resection to the date of the first dose of chemotherapy was reported. The mode and median number of chemotherapy doses were reported, with each dose identified as an OHIP claim for chemotherapy on a different date. The mode and median duration of chemotherapy was

35 26 reported, with duration defined as the interval between the first and last dates of chemotherapy OHIP claims. Objective #1: Visits and tests For physician visits, the specialty of the physician was determined by the specific consultation code that was billed (general practitioner, general surgery, radiation oncologist, Table A.6). Medical oncologists bill consultations using internal medicine fee codes, and thus could not be readily identified due to the non-specificity of this designation. Instead, those physicians who had both a history of billing for chemotherapy in the data set (January 1996 through March 2007), had a designation of Medical Oncology, Hematology, or GP/FP in the IPDB and billed using internal medicine, GP/FP or hematology consult codes were classified as medical oncologists (Table A.6). For group #1 (who were alive and had no evidence of relapse in the follow-up period), the count of MD visits (general practitioner, general surgeon, medical oncologist, radiation oncologist, see Table A.6) was categorized (None, 1, 2, 3-5, >5). Endoscopy testing (sigmoidoscopy, colonoscopy, Table A.7), body imaging modalities (CT abdomen/pelvis, ultrasound abdomen/pelvis, Table A.7), CT thorax, and MRI abdomen/pelvis was similarly categorized. When reporting the number of patients having >5 visits or tests, the median and range of the number of visits or tests for that group was also reported. The counts of endoscopic modalities included those performed within the first year of follow-up, as adherence to current guidelines may result in only one colonoscopy in the five year follow-up period. Patients who received a total colectomy were excluded from endoscopy counts. The polypectomy rate per endoscopy was calculated by taking the quotient of the total number of polypectomy fee codes (Z570, Z571) by the total number of endoscopy, among patients alive with no evidence of relapse at 5 years (group #1), who did not receive a total colectomy. This rate represents endoscopies with >=1 polypectomy, as additional polypectomies are billed under different fee codes. For the same group of patients, the rate of complete colonoscopy was determined by searching for a colonoscopy code (Z555) accompanied by an additional code representing assessment of the caecum (E747) or terminal ileum (E705). Due to the varying amounts of time in follow-up for patients that relapsed within the follow-up period (group #2), visits and tests were expressed as rates, per 6 months of follow-up time. A censoring period (figure 2) was used to exclude any increased activity prior to the relapse event which may artificially

36 27 inflate the frequency of tests. Due to the possibility of an increase in testing or MD visits prior to relapse (for example, multiple MD visits and CT scans prior to a liver resection), the analyses were performed using a 60 or 90 day censoring period. Since endoscopy counts included the first 12 months after CRC resection, the denominator in endoscopy rates included this time interval. For either denominator of follow-up time, the frequency or rate of general surgeon visit and endoscopy was stratified by colon or rectal primary, and the frequency or rate of radiation oncologist visit was expressed for rectal primary (there is almost no role for radiation in the treatment of colonic primaries). All categories were also stratified by high or low risk primary tumour. Exploratory analyses for visits and tests (Secondary objective #1) To examine potential factors associated with the increased likelihood of receiving at least one endoscopy modality in follow-up years 1 through 5, a multivariable logistic regression model was constructed (Model #1). The population for this analysis was group #1 (those who were alive and had no evidence of relapse at the end of the follow-up period), but did not receive a total colectomy. The dependent variable was the binary variable of receiving 1 or more endoscopies in the follow-up period (including year 1), versus none. Covariates used included age at CRC diagnosis (continuous variable), sex, mean neighborhood income quintile, colon versus rectal primary, rural or urban, Charlson comorbidity classification, and period of diagnosis ( vs ). Similarly, factors associated with the increased likelihood of receiving one or more body imaging modalities was assessed using multivariable logistic regression (Model #2). This analysis was performed on group #1. The dependent variable was the binary outcome of receiving one or more body imaging modalities (CT, ultrasound) versus none. Covariates included age at diagnosis (continuous), sex, mean neighborhood income quintile, Charlson comorbidity classification, colon versus rectal primary, rural or urban, colon versus rectal primary, period of diagnosis, medical oncologist and general surgeon followup. The latter two variables were classified as zero follow-up versus one or more visits in years 2-5. Interactions between high/low risk primary tumour and general surgery or medical oncology follow-up was assessed. If any of these interactions were significant, the model would be run with stratification by high or low risk primary tumor. If not significant, the final model was run excluding these interaction terms.

37 28 Objective #2: Treatment for disease relapse To describe group #2 (relapsed during the follow-up period), the count of patients who relapsed by follow-up year stratified by colon/rectum and high/low risk primary tumour was summarized. Secondly, while stratifying by year of relapse, relapse events among this group over the follow-up period was reported as counts and proportions. The counts and proportion of relapsed patients who received either palliative (chemotherapy) or potentially curative (lung or liver resection) treatment was then summarized by stratifying by LHIN or mean neighborhood income quintile. As a measure of difference between highest and lowest LHIN or income quintile, the extremal quotient (EQ) was used. The EQ calculated by taking the quotient of the highest to the lowest proportion. Differences between LHINs and income quintiles were also assessed using the chi-square test. The two-sided Cochran-Armitage trend test was used to assess for increasing or decreasing trends among income quintiles due to the ordinal classification of this variable. To characterize possible regionalization of lung and liver resections, the institution number associated with the lung or liver resection was obtained from either the CIHI-DAD or OHIP record of the procedure, and linked with its LHIN. The LHINs where all patients with either lung or liver resection received their procedure were summarized by counts and proportions. For survival, the vital status and date of death was obtained from the RPDB. Overall survival (OS) was calculated from the date of liver or lung resection, or from the date of relapse until their date of death or end of the available dataset (March 31, 2007). Survival was modeled using the Kaplan-Meier method (124), reporting median and 5 year OS. Differences in survival between groups were assessed using the log-rank test (125). Survival was assessed among those with evidence of CRC relapse in the 5 year follow-up period, for those patients receiving liver resection versus none, lung resection versus none, and receipt of chemotherapy versus none. Exploratory analysis for relapse treatment (Secondary objective #2) The primary outcome of surgical treatment of relapse (lung or liver resection) was examined using multivariable logistic regression (Model #3). This analysis was performed on group #2 (evidence of relapse in the follow-up period). Covariates in this analysis included age at relapse (continuous), sex, Charlson comorbidity classification at the time of relapse, income quintile, rural or urban, colon or rectal primary, high or low risk primary, period of diagnosis, general surgery visit frequency, medical oncology

38 29 visit frequency, and body imaging frequency. The latter three variables were expressed as frequencies per 6 months of follow-up, as patients who relapsed had varying follow-up times. As described above, there was a potential for an increased frequency in tests and MD visits in the period prior to the first evidence of relapse. Therefore the initial analysis was performed only examining events up to 60 days prior to relapse. To assess sensitivity of results to the length of the censoring period, the analysis was repeated censoring up to 0, 30, 90, and 120 days prior to the relapse date (Figure 2). Figure 2. Schematic timeline representing censoring period for measuring test and MD visit frequency in relapsed patients In all logistic regressions, multicollinearity among covariates was assessed. A statistically significant correlation coefficient greater than or equal to 0.8 was considered to indicate the presence of multicollinearity, and in such cases only one member of a correlated set was retained for the multivariable analysis. Secondary Objective: Comparison with primary chart reviewed cohort To determine the accuracy of the algorithm used to identify disease relapse in administrative data, a cohort of patients reviewed by primary chart extraction was acquired (with permission from Dr. Lillian Siu). This cohort consisted of stage II and III CRC patients treated during the time period of our study (from 1999 through 2001) who were followed up at either an academic hospital (Princess Margaret Hospital) or in a community setting (Credit Valley Hospital). Probabilistic linkage of individual level data to the administrative databases occurred through a multi-step sequential process using patient name,

39 30 date of birth and diagnosis date of CRC if available. Of 529 patients, there were 6 duplicates and 9 nonmatches to the administrative data, and one invalid IKN. A total of 513 (333 eligible for follow-up, 180 ineligible due to progressive or recurrent disease on restaging at less than one year of follow-up) patients were matched to a corresponding valid, unique encrypted identifier in the administrative data. These patients were then matched with their corresponding records among the study cohort of patients, prior to applying year-one exclusions. Variables used included date of diagnosis, clinical stage, pathologic stage, relapse status (yes/no), location of relapse, and treatment for relapse. Sensitivity was calculated as the quotient of true positives by the sum of true positives and false negatives. Specificity was calculated as the quotient of true negatives by the sum of true negatives and false positives. Positive predictive value was calculated as the quotient of true positives by the sum of true positives and false positives. Negative predictive value was calculated as the quotient of true negatives by the sum of true negatives and false negatives. The sensitivity, specificity, positive predictive value and negative predictive value of classifying patients as eligible for follow-up was calculated by a 2x2 table summary. For any false positives (those patients who actually relapsed in their first year, but misclassified as being eligible for follow-up), it was determined whether or not they were eventually classified as having relapsed. In addition, the interval from the date of resection to the date of relapse detection was described by the range and median value. For the false negatives (patients eligible for follow-up by chart review but classified as ineligible by the present study), exclusion criteria were examined for the cause of misclassification. The group of patients who were appropriately classified as eligible for follow-up by the administrative data was then examined for the sensitivity, specificity, positive predictive value and negative predictive value of detecting disease relapse in the follow-up period, by a 2x2 table summary. For false positives (classified as relapsed but did not by primary chart review) relapse criteria were examined for the cause of misclassification. For false negatives (recorded as relapsed in the primary chart review but not by the present study), the year of relapse, location of relapse, and where or not the relapse was surgically resected was reported. The overall group of patients (the group of patients excluded due to first year events, as well as those deemed eligible for follow-up at year 2) was then summarized by a 2x2 summary table for the accuracy of detecting disease relapse in the follow-up years 1-5. The sensitivity, specificity, positive predictive value and negative predictive value was calculated. Finally, the accuracy of using adjuvant chemotherapy as a proxy for high risk disease was evaluated by stratifying the cohort by colon or rectal primary, then comparing receipt of adjuvant chemotherapy to the pathologic stage recorded in the chart reviewed data. If pathologic stage was not available, clinical stage

40 31 was used. This stratification was used to reflect clinical practice guidelines (10) where both stage II and III rectal cancers are recommended to receive adjuvant chemotherapy. Data analysis All programming and statistics was performed using SAS for Unix (Cary, NC). All odds ratios were reported with 95% confidence intervals. P-values of less than 0.05 were considered to be statistically significant. All tests were two-sided.

41 Results Inception cohort A total of 23,914 patients were identified from the OCR from January 1, 1996 through November 30, These patients had a diagnosis of colorectal cancer with a histology of adenocarcinoma, aged 18-80, and had no other primary cancers recorded in the OCR. After examination of linked data from CIHI procedure codes and OHIP billing codes, 20,635 patients received a colorectal cancer resection from 14 days prior, to 120 days after the OCR diagnosis date. Of the different types of CRC, classified by ICD-9 diagnosis code, cancers of the rectosigmoid junction and rectum comprised of 29.3% of these patients. For the colonic primaries, the most common ICD-9 diagnosis codes were for tumors of the sigmoid colon (20.9%), cecum (13.0%), ascending colon (12.5%), and not otherwise specified (11.3%) (Table 1). Table 1: Types of CRC by ICD9 diagnosis code ICD9 code Definition N % Hepatic flexure % Transverse colon % Descending colon % Sigmoid colon % Cecum % Ascending colon % Splenic flexure % NEC % Not otherwise specified % Rectosigmoid junction % Rectum % Exclusions To create the inception cohort of patients eligible for postoperative surveillance, patients were excluded based on events at presentation or in the first year following primary CRC resection suggesting advanced (metastatic) disease. These exclusions are summarized in Table 2 (numbers are not mutually exclusive). The most common reason for exclusion was diagnosis of advanced disease (4609, 58.9%), followed by death (3055, 39.3%), prolonged chemotherapy (1013, 22.7%), palliative care consult (1646, 21.0%), liver procedure (1590, 20.3%), and late start of chemotherapy (804, 18.0%). Lung procedures accounted for a minority of exclusions (113, 1.4%). To understand the proportion of patients who were excluded based on fulfilling at least two exclusion criteria, counts of patients by pair-wise combinations of exclusion criteria were tabulated (Table 3). For 32

42 33 each of the criteria, with the exception lung procedures (42%), more than 50% of patients with any exclusion criteria also had a diagnosis of advanced disease in the first year. The proportion of excluded patients from each LHIN from the denominator of resected CRC patients was evaluated, standardizing each LHIN proportion by age and sex against the reference population of Ontario With the exception of the Southeast LHIN where 100% of patients were excluded (due to alternate funding plan, see Methods, page 21), the overall crude exclusion rate was 34.9%, with crude rates from each LHIN ranging from 31.0% to 37.9%. Following adjustment by age and sex, the overall directstandardized rates of exclusion was 34.7%, with LHIN rates ranging from 29.9 to Excluding the Southeast LHIN, no LHIN had a significantly different rate of exclusion than others (Table 4). Baseline characteristics - excluded cohort Baseline characteristics were also described for the excluded group of patients in Table 5. For the excluded patients, the mean age at CRC diagnosis was 65.4 years. The group was predominantly male (57%). Over 75% of these excluded patients had a medium or high Charlson comorbidity classification. The distribution of patients ranged from 18.5%-21.8% among the five income quintiles. A small number of patients had missing income quintile (306) or missing LHIN (4) and these patients were eliminated from future analyses. Following application of the exclusion criteria, 7,831 patients were eliminated while 12,804 patients were deemed likely disease-free at the start of the follow-up period (post primary CRC resection year 2 through year 5), and eligible for postoperative surveillance. Baseline characteristics study cohort There were 12,804 patients in the study cohort, eligible for follow-up at the start of the follow-up period, and baseline characteristics are described in Table 6. The mean age of the inception cohort was 65.3 years, with 7110 (55.5%) male patients. This cohort consisted of four groups of patients. Group #1 (alive and had no evidence of disease at the end of the follow-up period) consisted of the majority of patients (8804, 68.8%). The next largest was group #2 (who had evidence of CRC relapse in the follow-up period, 3316, 25.9%). The final two groups comprised of those who died within the 5 year follow-up period yet had no evidence of relapse or those whose only evidence of relapse was a cancer-related cause of death in

43 34 the follow-up period. More than one-third of the entire cohort (5203, 40.6%) had a primary CRC that was high risk for recurrence (as defined in Methods). The majority of these patients had a low to medium Charlson comorbidity classification prior to primary CRC resection (79%). Distribution across income quintiles ranged from 17.9% to 21.5%. Slightly more patients were diagnosed in 1999 to 2001 than 1996 to 1998 (53% vs. 47%). Table 2: Number of patients meeting each exclusion criteria N = 7,831 N % Advanced disease in year % Liver procedure (incl. biopsy) in year % Lung procedure (incl.biopsy) in year % Late start of chemotherapy (>120d) % Prolonged course of chemotherapy (>270d) % Palliative care consult % Died in year % Table 3: Pair-wise summary of first year exclusion criteria* Advanced disease (n=4609) Liver procedure (n=1590) Lung procedure (n=113) Late start of Chemotherapy (n=804) Long course of Chemotherapy (n=1013) Palliative care consult (n=1646) Died In 1 st yr Pall. care Long chemo Late chemo Lung procedure Liver procedure Advanced disease Died in first year (n=3055) *represents the number of excluded patients having both criteria from row/column; numbers are not mutually exclusive

44 35 Table 4. Excluded patients Direct age-sex standardization to Ontario population 1996 LHIN Excluded All CRC Crude Rate Standardized 95%CI patients (%) rate (%) Erie St. Clair South West Waterloo Wellington Hamilton Niagara Haldimand Brant Central West Mississauga Halton Toronto Central Central Central East Champlain North Simcoe Muskoka North East North West Overall Southeast* Total** * patients from Southeast LHIN excluded from standardization **patients with missing LHIN information excluded from standardization

45 36 Table 5: Baseline characteristics of excluded cohort N = 7,831 N % Age at CRC diagnosis 65.4 (mean) 10.5 (std. dev.) Male Female % 43.0% Charlson score Low (0-1) Medium (2) High (>2) % 36.6% 42.1% Income quintile 1 (poorest) (richest) Missing Local Health Integration Network Erie St. Clair South West Waterloo Wellington Hamilton Niagara Haldimand Brant Central West Mississauga Halton Toronto Central Central Central East Southeast Champlain North Simcoe Muskoka North east North West Missing % 21.8% 21.2% 18.7% 18.5% 4% 5.5% 7.5% 4.4% 12.5% 3.6% 5.9% 8.1% 9.3% 11.2% 12.7% 7.8% 3.6% 5.8% 2.1% 0.06%

46 37 Table 6: Characteristics of study cohort, eligible for postoperative surveillance N= 12,804 N % Age at CRC diagnosis 65.3 (mean) 10.5 (std. dev.) Male Female % 44.5% Colon high risk Colon low risk Rectum high risk Rectum low risk % 43.6% 15.8% 15.8% Alive and no evidence of relapse at 5 years Evidence of relapse in 5 years Died but no evidence of relapse at 5 years Cancer related cause of death only Charlson score Low (0-1) Medium (2) High (>2) Local Health Integration Network Erie St. Clair South West Waterloo Wellington Hamilton Niagara Haldimand Brant Central West Mississauga Halton Toronto Central Central Central East Champlain North Simcoe Muskoka Northeast Northwest Income quintile 1 (poorest) (richest) Period of diagnosis % 25.9% 4.3% 1.1% 21.6% 57.4% 20.9% 6.2% 9.3% 5.7% 12.8% 4.0% 6.7% 8.8% 11.2% 11.3% 9.8% 3.8% 7.9% 2.4% 17.9% 21.5% 20.3% 19.4% 20.9% 47.1% 52.9%

47 38 Preliminary analyses The OHIP billings and CIHI procedure codes for colon resection, liver resection, and lung resection were examined for agreement in procedure dates. Among primary CRC resections, there was an 85% agreement between dates of OHIP claims and CIHI procedure codes. For the study cohort, there was a 90% agreement in dates for liver resections, and 86% for lung resections. Furthermore, the primary diagnosis code was examined in all patient CIHI admissions where a liver or lung resection was documented, at any point following the primary CRC resection. For all admissions documenting a procedure of liver resection or destruction from the time of primary CRC resection to five years afterward, 95% of the ICD-9 or ICD-10 codes for primary diagnosis corresponded to either CRC (153.x, 154.0, 154.1, C18.0, C18.2-C19.0, C20.0), or secondary neoplasm to the liver (197.7, C787). For admissions documenting a lung resection or destruction in the same time frame, 92% of all ICD-9 or ICD- 10 primary diagnosis codes corresponded to CRC, secondary neoplasm to the lung (197.0, C780) or malignant neoplasm of the lung (162.X, C343-C349). The receipt of adjuvant chemotherapy was characterized for the inception cohort. The most common interval to the first course of adjuvant chemotherapy was 41 days, with a median of 47 days and an interquartile range of days. For the number of doses of chemotherapy, the mode was 30, with a median of 25 doses. For duration of chemotherapy, the mode was 144 days, with a median of147 days.

48 39 Objective #1 Visits and tests To describe the processes of care that patients receive in the follow-up period, including MD visits and surveillance tests Physician visits and tests Group #1 In the cohort of patients who were both alive and had no evidence of relapse at the end of the follow-up period (Group #1, n=8,804), the frequency of physician visits, imaging, and endoscopy was examined over the entire follow-up period. The overall summaries of visit and tests are listed in Table 7. A minority of patients (3.8%) did not see their general practitioner in the entire follow-up period and the vast majority visited their general practitioner greater than 5 times in this period (87.3%, median 21, range 6-213). For medical oncology follow-up, a marked difference was observed for patients with a high risk versus low risk primary tumour. For low risk patients, the majority (71.3%) did not see a medical oncologist in follow-up, but most high risk patients saw their medical oncologist greater than 5 times (55.1%, median 8, range 6-86) (Figure 3). The patterns of follow-up for general surgeon visits were similar among patients with a rectal or colonic primary, where a slightly higher frequency of follow-up visits was observed in those patients who had high risk primary tumours. In colon cancer patients, 39.2% of high risk patients (median 8 visits, range 6-27) compared with 33.3% of low risk patients (median 8, range 6-38) had greater than 5 physician visits. In rectal cancer patients, 42.3% of high risk patients (median 8, range 6-53) and 38.2% of low risk patients (median 8, range 6-47) had greater than 5 visits (Figure 4). Radiation oncology follow-up of rectal patients appeared to be higher in frequency among those with a high risk primary tumour, where more than 35% of high patients had 3 or more visits over the follow-up period, while 88.1% of low risk patients did not see a radiation oncologist at all. The number of patients in the subgroup who did not undergo a total colectomy was 8,486. In both patients who had rectal or colonic primaries, more than two thirds of patients received between 2 and 5 endoscopic examinations over all five years after primary CRC resection. A small number in each group (416 or 7.1% of colon patients, 172 or 6.5% of rectal patients) did not receive any endoscopic evaluation at all (Figure 5a). A summary of the counts of complete colonoscopies over follow-up years 1-5 in this population is summarized in Table 5b. The frequency of body imaging modalities appeared to be different among patients with high risk and low risk primary tumours. More than one third (38.7%) of the low-risk cohort did not undergo CT imaging or

49 40 ultrasound in the follow-up period. In contrast, more than half of the high risk group (51.2%) received at least 3 imaging tests over this time period (Figure 6). The imaging modalities of thorax CT and MR of the abdomen or pelvis were rarely used, where for each modality greater than 89% of patients had zero tests over follow-up years 2-5.

50 41 Table 7: Total visits and tests over follow-up years 2-5 in patients alive at 5 years without evidence of disease relapse N=8804 None >5 MD Visits General Practitioner High risk (N=3293) Low risk (N=5511) 336 (3.8%) 104 (3.2%) 232 (4.2%) 152 (1.7%) 54 (1.6%) 98 (1.8%) 144 (1.6%) 61 (1.9%) 83 (1.5%) 487 (5.5%) 218 (6.6%) 269 (4.9%) Test Medical Oncologist High risk Low risk General Surgeon (colon, N = 6107) High risk (N=2034) Low risk (N=4073) General Surgeon (rectal, N=2697) High risk (N = 1259) Low risk (N = 1438) Radiation Oncologist (rectal only, N=2697) High risk (1259) Low risk (N=1438) Endoscopy * (colon, N = 5833) High risk (N=1926) Low risk (N=3907) Endoscopy * (rectum, N = 2653) High risk (N=1242) Low risk (N=1411) Body imaging ** (N=8804) High risk Low risk Thorax CT (N=8804) High risk Low risk Body MR (N=8804) High risk Low risk 4464 (50.7%) 536 (16.3%) 3928 (71.3%) 1197 (19.6%) 360 (17.7%) 837 (20.5%) 446 (16.5%) 190 (15.1%) 256 (17.8%) 1925 (71.4%) 658 (52.3%) 1267 (88.1%) 416 (7.1%) 53 (2.8%) 363 (9.3%) 172 (6.5%) 56 (4.5%) 116 (8.2%) 2760 (31.3%) 625 (19.0%) 2135 (38.7%) 7892 (89.6%) 2816 (85.5%) 5076 (92.1%) 8544 (97.0%) 3138 (95.3%) 5406 (98.1%) 518 (5.9%) 174 (5.3%) 344 (6.2%) 575 (9.4%) 181 (8.9%) 394 (9.7%) 235 (8.7%) 114 (9.1%) 121 (8.4%) 115 (4.2%) 85 (6.8%) 30 (2.1%) 725 (12.4%) 177 (9.2%) 548 (13.9%) 324 (12.2%) 145 (11.7%) 179 (12.7%) 1693 (19.2%) 505 (15.3%) 1188 (21.6%) 598 (6.8%) 295 (9.0%) 303 (5.5%) <5 (<0.02%) 0 (0%) <5 (<0.02%) 304 (3.5%) 140 (4.3%) 164 (3.0%) 597 (9.8%) 177 (8.7%) 420 (10.3%) 243 (9.0%) 103 (8.2%) 140 (9.7%) 83 (3.1%) 65 (5.2%) 18 (1.3%) 1528 (26.2%) 539 (28.0%) 989 (25.3%) 574 (21.6%) 290 (23.3%) 284 (20.1%) 1258 (14.3%) 479 (14.5%) 779 (14.1%) 171 (1.9%) 94 (2.9%) 77 (1.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 1010 (11.5%) 627 (19.0%) 383 (6.9%) 1583 (25.9%) 519 (25.5%) 1064 (26.1%) 692 (25.7%) 320 (25.4%) 372 (25.9%) 290 (10.8%) 234 (18.6%) 56 (3.9%) 2783 (47.7%) 1017 (52.8%) 1766 (45.2%) 1210 (45.6%) 593 (47.7%) 617 (43.7%) 2211 (25.1%) 1105 (33.6%) 1106 (20.1%) 134 (1.5%) 80 (2.4%) 54 (1.0%) 252 (2.9%) 151 (4.6%) 101 (1.8%) Percentages reflect row totals * colonoscopy, flexible and rigid sigmoidoscopy; includes year 1 of follow-up but excludes patients with total colectomy ** ultrasound abdomen, CT abdomen/pelvis 7685 (87.3%) 2856 (86.7%) 4829 (87.6%) 2508 (28.5%) 1816 (55.1%) 692 (12.6%) 2155 (35.3%) 797 (39.2%) 1358 (33.3%) 1081 (40.1%) 532 (42.3%) 549 (38.2%) 284 (10.5%) 217 (17.2%) 67 (4.7%) 381 (6.5%) 140 (7.3%) 241 (6.2%) 373 (14.1%) 158 (12.7%) 215 (15.2%) 882 (10.0%) 579 (17.6%) 303 (5.5%) <13 (<0.3%) 8 (0.2%) <5 (<0.1%) 7 (<0.3%) <5 (<0.2%) <5 (<0.02%)

51 42 Figure 3. Number of medical oncology follow-up visits over follow-up years 2-5 for low and high risk CRC patients, among those patients with no evidence of relapse at the end of the follow-up period. Figure 4. Number of general surgeon follow-up visits over follow-up years 2-5 for low/high risk colon and rectal patients, among those patients with no evidence of relapse at the end of the followup period.