Treatment of colorectal cancer

Treatment of colorectal cancer Ph.D. Thesis 2005 Thomas Nahne Nickelsen Research Centre for Prevention and Health Glostrup University Hospital

CONTENTS CONTENTS...4 PREFACE...5 PUBLICATIONS...6 INTRODUCTION...7 AIMS...11 MATERIAL AND METHODS...12 Included and excluded patients...13 Outcomes...15 Statistical methods...15 RESULTS AND DISCUSSION...18 Validity and completeness of the DCCG-database (I)...18 Discussion...19 Conclusion...20 30-day mortality after colorectal cancer surgery in Denmark (II)...21 Discussion...24 Conclusion...24 Lifestyle and 30-day complications to surgery for CRC (III)...26 Discussion...27 Conclusion...28 Impact of structural Factors on outcome after surgery for CRC (IV)...28 Discussion...30 Conclusion...31 OVERALL CONCLUSION...32 PERSPECTIVES...33 SUMMARY...35 DANISH SUMMARY...37 ABBREVIATIONS...39 REFERENCES...40 APPENDICES...49 4

PREFACE The preparations to this project was made during my appointment at DSI Danish Institute for Health Services Research, and the Ph.D. thesis was accomplished during my appointment at the Research Centre for Prevention and Health, Copenhagen County, from February 2002 to February 2005. I wish to express my gratitude to everyone who helped me during this period of time. I want to thank the Danish Colorectal Cancer Group (DCCG) for data and financial support. I gratefully acknowledge following foundations for financial support: Apotekerfonden af 1991 Augustinus Fonden Læge Sofus Emil Friis og hustru Olga Doris Friis legat Fabrikant Einar Willumsens Mindelegat. Supervisors Torben Jørgensen, D.M.Sci., director, Research Centre for Prevention and Health, Copenhagen County. Ole Kronborg, D.M.Sci., professor, Surgical Department A, Odense University Hospital. 5

PUBLICATIONS This thesis is based on four scientific papers. Throughout the text these four papers will be referred to by Roman numerals. They are included in the thesis as appendix I IV. I Nickelsen,T.N., Harling,H., Kronborg,O., Bülow,S., Jorgensen,T. The completeness and quality of the DCCG database on colorectal cancer. Ugeskr Laeger 2004; 166(36): 3092-5. II T. N. Nickelsen, T. Jørgensen, O. Kronborg. 30-day mortality after surgery for Colorectal Cancer in Denmark. Colorectal Disease 2005; 7, 500-506. III T. N. Nickelsen, T. Jørgensen, O. Kronborg. Lifestyle and 30-day complications to surgery for Colorectal Cancer. Acta Oncologica, 2005; 44: 218-223.. IV T. N. Nickelsen, T. Jørgensen, O. Kronborg. The impact of structural factors on outcome after surgery for Colorectal Cancer. Submitted for publication. 6

INTRODUCTION Colorectal cancer (CRC) is defined as cancer in the colon (including the Appendix) or the rectum (0-15 cm from the anal verge). WHO estimates that 945 000 new cases occur yearly, having increased from an estimated 500.000 new cases in 1975 and 783.000 in 1990. CRC represents 12.6% / 14.1% of all incident cancers in men / women in the westernised countries compared to 7.7% / 7.9% in other countries. In the European Union it is the second most common cause of death from any cancer with a lifetime incidence of 5%. The incidence of CRC increases with age, but its incidence and mortality in general are now decreasing [1,2]. In Denmark CRC comprise approximately 3500 new patients every year. The incidence in Denmark of rectal cancer has been falling the last half century, while it has been increasing for colon cancer. There is no difference in incidence between sexes in colonic tumours, but rectal tumours are more common in the male sex. Approximately 98% of the CRCs are adenocarcinomas, the remaining are carcinoids or sarcomas. When CRC in the following are mentioned in relation to the database, the term only includes adenocarcinomas. Symptoms from CRC differ from none to changes in bowel habits, per anal bleeding, pain, and ileus. Sometimes the symptoms are unspecific like anaemia and fatigue. The disease is most often Table 1 - Survival according to stage [2] 7

diagnosed by endoscopy, but in 25% of the cases the diagnose is made during surgery initiated by ileus or perforation. As there is more than one tumour in 4-5% of the cases examination of all the colorectal segments is necessary [2-4]. Until recently the CRC was graded according to Dukes classification, grading the growth through the intestinal wall from A (not penetrating) to D (penetrating with distant metastases). Approximately 15% of the rectal tumours are Dukes A and 7% of the colonic tumours. Now the TNM classification (Tumour, Node, Metastases) is used looking separately at the tumour penetration, the nodes and the distant metastases (grading from 0 to IV). The CRC is often diagnosed in an advanced stadium because of the unspecific or weak symptoms, and the advanced stadium leads to decreased survival (Table 1)[5]. The strategy for treatment is resection of the tumour-bearing segment with draining lymph nodes and if possible removal of metastases. Approximately 80% of the patients have no macroscopic evidence of residual tumour after resection, but more than half of these develop recurrence and die of the disease. This is a result of occult viable tumour cells that have metastasised before surgery. Adjuvant treatment (chemotherapy or radiation) are therefore offered in combination with surgery in the advanced cases [2,6]. Through the nineties there was a debate on the quality of the treatment of CRC in Denmark. Statistics showed that the survival was lower compared to that in other Nordic countries [7]. This was based on central statistics in the Cancer Registry and the Central Office of Civil Registration (COCR). These registers contain no clinical data apart from diagnose, sex and age, and no adjustments were made estimating the mortality. It was therefore decided that an existing clinical database (the Rectum-Database ) should be updated and expanded to include all CRC patients, as 8

a basis for further quality improvement. The DCCG (Danish Colorectal Cancer Group), a subgroup of the Danish Surgical Society, was managing the old as well as the new database. Quality improvement uses the terms criteria, standards and quality indicators. In Danish literature the criterion is defined as the ideal situation and the standard is the realistic level of accomplishment. In American literature the differentiation is less stringent. The quality indicators are used to measure and compare the accomplished level of care. The area quality of care is classified under three categories: structure, process, and outcome of care. Structure: The resources and organisational structure of health care including staff resources and qualifications, technology, and facilities. Process: Refers to what is actually done in giving health care. Outcome: The effects of treatment on the patient s health status [8,9]. Figure 1 Diagram of the four factors with impact on the outcome of the treatment. Patient Structure Process Outcome Disease According to Sackett [10] the outcome of treatment are depending of four aspects: The patient The disease 9

The treatment (process) The organisation (structure) The theory applies well to the diagram in Figure 1. It would have been ideal to study all associations in Figure 1 to learn more about which factors that relate to the outcome, but this would exceed the limits of a Ph.D.-study. I therefore have focused on three areas. Initially I evaluate whether there are significant differences on the outcomes of the departments adjusted for case-mix. Without this difference further investigation of associations in the model (Figure 1) would seem less interesting. Two associations were chosen for analysis. Lifestyle factors are not included in prediction models or case-mix, though an increasing number of studies find it associated to mortality and morbidity [11-18]. No study however has been done nationwide and including four lifestyle factors, and only a few are including more patients [14]. The second association is between structure and outcome. This has been presumed in most accreditation companies (e.g. Joint Commission), but I have not yet identified any study systematically or nationwide analysing this. 10

AIMS The aims of this Ph.D. study were: To evaluate the validity and completeness of the DCCG-Database. To compare 30-day mortality after CRC surgery between all departments in Denmark and to evaluate whether a possible difference could be explained by differences in case-mix. To identify lifestyle factors with impact on 30-day mortality, postoperative complications in general, impaired wound healing, anastomotic leakage, deep wound infection, and thrombosis after surgery for CRC. To show whether structural factors have impact on 30-day complications or mortality after CRC surgery in Denmark, when adjusted for factors related to the disease and patient. Accordingly this thesis consists of four parts. 11

MATERIAL AND METHODS All patients operated for a first time colorectal adenocarcinoma between 1 May 2001 and 31 December 2002 were eligible. The operation should take place in one of the 53 Danish public hospitals treating CRC surgically, leaving out operations in private hospitals, estimated to be less than 1%. Foreigners operated on in Denmark are included. Data originates from four sources: the COCR, the National Patient Register (NPR), a separate questionnaire sent to all departments, and the DCCG-database. - The COCR includes an updated vital status on all Danish citizens, and has an ascertainment rate at 100%. The COCR was used to validate the 30-day mortality. - The NPR comprises information on all patients at all hospitals in Denmark and has an ascertainment rate close to 100%. The contents are primarily dates of examination, admission, operation, and departments responsible for the treatment. The data are collected for statistics in the central administration, and the only clinical information is diagnose, and type of surgery and other treatments [19]. The NPR was used to validate the completeness of the DCCG-database. - The questionnaire on structural data was made especially for this study (Appendix V) including variables relevant for the outcome, either according to other studies, accreditation companies or own experience. It was answered by 45 of 47 departments. Two departments were missing (n=193 patients), the remaining six departments (n=56 patients) was closed down during the study period. - The national clinical DCCG-database was founded 1 May 2001 by the Danish Colorectal Cancer Group (DCCG), a subgroup of the Danish Surgical Society. It includes all patients in Denmark with a first-time diagnose of colorectal adenocarcinoma treated or diagnosed in surgical departments. Its purpose is to unify procedures for diagnosis and treatment, and to improve and assure the quality. The data are provided to the database through questionnaires from the patients (Appendix VI) and the surgeons (Appendix VII) [20]. All departments are notified monthly of missing data and logic 12

errors in the questionnaire, not correctable by the DCCG-secretary. The completeness of data is validated annually using the NPR. The data reported from the patients to the DCCG-database included weight and height, comorbidity, symptoms from the tumour, alcohol and tobacco consumption, self perceived physical fitness (SPPF), and general health. The data from the surgeons were divided in four major topics in the questionnaire: diagnostics made, staging (preoperative and final), treatment (information of the operation, operative complications), and the postoperative 30 days (complications, mortality, and planned follow-up regimen). The database covers 89.2% of all patients with CRC in Denmark (2001-2). The missing data in this study are mainly caused by missing questionnaires although some missing answers occur too (errors not corrected by the departments are entered the database as missing). The questionnaire from the patients had a completeness of 58.6% (n=3550). Included and excluded patients. A total of 6058 patients were diagnosed during in the study period, according to the NPR and DCCG. Age, sex, department and mortality were known for all patients, but 10.8% (n=656) had no data from the surgeons and 41.4% (n=2508) had no data from the patient questionnaire. Only operated patients Figure 2 - Patients in- and excluded. Due to variations in the analysed variables, the number of included patients varies in the four studies All patients with colorectal adenocarcinoma in Denmark (n=6058) 656 Questionnaire from surgeon received (n=5402) 215 Operated patients (n=5187) 1703 Questionnaire from patient received (n=3484) with data from the surgeons were included in the study (n=5187) (Figure 2). All questionnaires used in this thesis were manually checked for logic errors and missing answers. The data files created by 13

the automated scanner were validated in comparison to the original paper based questionnaire through random samples and a program searching for logic errors. Any missing or erroneous data correctable without contacting the department were corrected by the author. Data not correctable or data missing in the surgeon questionnaire were returned to the departments asking for corrections. No reminders were sent and erroneous data not corrected were entered in to the database as missing. Two subgroups of patients were defined. A minimal-risk population was defined including patients fulfilling all of the following conditions: Never smoked, Body Mass Index < 30 kg/m 2 (BMI), consuming maximum 60 g alcohol/day and having a SPPF at or above average (III). A population of patients with a stoma was defined as patients having had a rectum resection, Hartmann s operation, other stoma inducing operations, or a palliative stoma (III). From 10 of the 53 departments (n=950) the DCCG-database received less than 90% of the expected surgical questionnaires according to NPR. These 10 departments underreported mortality (p=0.02) and complications (p=0.0002). The 651 excluded patients without surgical questionnaires were analysed according to age, sex, department and mortality. The oldest patients and patients surviving 30 days after the operation had a higher risk of being missing. Validity of data. The completeness and validity of the questionnaire from the surgeons were analysed for the initial eight months of the database s existence (1 May to 31 December 2001, n=2188). A random sample of 5% (n=87) of the patients were created in May 2002, and the surgeon questionnaires were filled in again with data from the patient case records. One case record was missing. The extra case record based questionnaires were handled and corrected exactly like the originals and finally compared to those. The validity was evaluated as a percentage of identical answers for each question and as a kappa value. Some questions were not analysed, as the data were not available in the case records. 14

The completeness was calculated as the part of patients registered in the DCCG-database, compared to the number of patients in the DCCG-database and the NPR for the study period (I). Patients in the DCCG-database, but not registered in the NPR (n=67), were regarded as missing in the NPR. Because the 5% sample was defined before the database was closed for analysis of completeness (spring vs. autumn 2002), the number of patients had increased and as a consequence the initial sample was now less than 5%. Outcomes. Endpoints in the study were 30-day complications or 30-day mortality. The mortality was validated in the COCR (ascertainment 100%). The complications were reported by the surgeons (Appendix VII, question 33) and were not verified systematically (apart from the random sample se above). Mortality was not included in complications. All reported complications in the Various category were entered to the database manually and if possible replaced in one of the other 19 specified complications. The complications were when possible grouped in one of four various categories (impaired wound healing (12.4%), anastomotic leakage (7.2%), deep wound infection (6.9%), and thrombosis (6.5%)). The included diagnoses are shown in Appendix III, Table 1. A total of 512 patients (9.5%) died within 30 days of the operation (n=5188) and 3491 patients had one or more complications (32.6%). Statistical methods. Summary statistics are given as mean and median (range). Simple comparisons are performed by chi 2 -test (I-IV). Kappa statistics were used to analyse the interobservational variation (I). The kappa value was calculated by the formula kappa = (p o p c ) / (1 p c ) where p o is the observed proportion of 15

agreement, and p c is the proportion of agreement expected by chance. The value can vary from +1 (perfect agreement) to 0 (agreement no greater than expected by chance), and can assume negative values when agreement is less than expected [21,22]. Logistic regression was used in article II-IV. If just one of the variables in question was missing, the patient was not included. As the analyses were made with many different variables, the number of included patients (n) varied. The lowest numbers were in analyses including data from the patient questionnaire because of its low completeness. When age was included linearity was assured. Logistic regression was used when comparing 30-day mortality rates between departments adjusted for case-mix (II). As the analysis needs a minimum of one endpoint (death) for each department some departments are missing in the presentation of radically operated patients. These departments are in a joint group of all departments with less than 25 operated patients or no mortality. Akaikes Information Criterion (AIC) was used to choose the most optimal model when taking the number of parameters in the model into account [23] (II). Evaluating the impact of lifestyle factors logistic regression was performed with the following outcome variables: 30-day mortality, complications in general and the four different categories of complications. The model was reduced by backward selection. ASA could be an intermediate factor between lifestyle and complications or death. Therefore the analyses were repeated introducing ASA separately, causing no or minimal changes in the final significant variables and estimated risks (III). Logistic regression was used in the analyses of the association between structural factors and complications or 30-day mortality distinguishing between colonic and rectal procedures. Due to the low number of deaths, univariate analyses and a reduced number of variables were introduced in some of the rectal analyses and all stoma analyses (IV). The number of operations performed by each surgeon was not available. Instead the operation-rate per surgeon in each department was used 16

as a proxy. The operations-rate was defined as the yearly number of CRC-operations performed in the department divided by the number of surgeons in the department operating without need of supervision. The attributable risk AR was calculated by the formula AR = (E (RR 1)) / (E (RR 1) + 1). E is part of exposed patients in the cohort and RR is the relative risk. Estimating AR we assume causality of the lifestyle factors and uniformity between OR and RR (III). The AR estimates a maximal effect of an intervention. Generalized Estimating Equations (GEE) were used to adjust for department-dependent factors (III- IV). All analyses were performed with 95% confidence interval. SAS version 8.02 was used for the management and analysis of the data. 17

RESULTS AND DISCUSSION. Validity and completeness of the DCCG-database (I). A total of 2598 patients were found in the DCCG-database and the NPR. In total 420 patients (16%) were missing in the DCCG-database. For the departments the average completeness was 76% (0 100%). Some of the patients identified in the NPR were not eligible for the database. This was caused by different histology (not adenocarcinomas) or failure in data concerning location of the tumour, date of initial diagnosis or operation. When corrected for these patients the completeness was 86%. The analysis for completeness was done before reminders were sent to the departments. The mean validity for the four parts of the surgeon questionnaire is shown in Table 2, and for a selection of questions in Table 3. Overall 34% of the kappa values were 0.8 or more (almost perfect) and further 20% were between 0.6 and 0.8 (substantial). Table 2 - Mean validity of the four parts of the questionnaire. Variable Percentage identical answers Kappa Diagnostics 84.3% 0.67 Treatment 93.6% 0.67 Staging 88.9% 0.65 Postoperative 90.3% 0.55 18

Table 3 - Validity for a selection of essential questions. Variable Percentage identical answers Kappa 7 ASA-class 62.9% 0.09*ª 12 Date of operation 94.1% b 13 Urgent operation 97.6% 0.86 15 Localisation of tumour 98.7%* 0.77*ª 22 Bloodloss 85.7% 0.94 28 Radicality of operation 87.5% 0.80 30a Number of glands resected 78.5% b 31 Dukes-classification 97.6% 0.94* 33 Complications 98.4%* 0.54*ª 34 Blodtranfusion 73.7% 0.45 35 Mortality 98.7% 0.90 40a Endoscopy before operation 84.3% 0.61 * Kappa is estimated for each possible marking (marking vs. no marking) the value shown is a mean. ª One or more kappa values are 0. b Not possible to calculate kappa values Discussion A premise of all quality assurance is complete and valid data, avoiding wrong assumptions initiating quality improvement programs with a wrong focus. The DCCG-database has now been 19

receiving data for more than 4 years and has become one of the largest clinical databases in Denmark. Completeness is high even compared to longer existing databases [24-36]. Estimating the completeness, we did not give the database the benefit of the doubt. The true completeness may therefore be higher. As the first two rounds of reminders based on NPR-data are completed, we now (2004) know to have a satisfying response rate, increasing the completeness further. This study defines the completeness according to patients registered in the database opposed to missing patients. Partly missing data in questionnaires received at the database are not considered. As all new questionnaires received each month are validated for logic errors and missing data asking the departments for corrections, the extend of this problem is limited. The validity was estimated through comparison of the original questionnaire received at the database and a case record based questionnaire filled in by one of three head-surgeons. Identical answers are no assurance of a true answer, but we believe the liability is greater. We compared through direct comparison and through a kappa value. Only one other study used kappa value and it had similar results to ours [22]. Other studies had comparable results using percentages (I, Table 4) [24,26-28,31,34,35,37-40]. Our analyses illustrate the limitations of the kappa value. Several kappa values were zero in spite of a high percentage of identical answers. This was caused by an empty cell in the underlying 2x2 table and was the reason we chose to include the percentage of identical answers. As the questionnaire has no variables covering identical topics, the percentage should be 100%. Conclusion The completeness and validity of the DCCG-database is considered sufficient for the examined time period. The planned quality assessment using DCCG-data will be reliable. As the procedures submitting data to the database are getting well-established, the completeness and validity is expected to rise. 20

30-day mortality after colorectal cancer surgery in Denmark (II). The 30-day mortality in 5187 operated patients was 9.9% and the mortality in 44 departments is shown in Figure 2. Baseline characteristics are shown in Table 4. The 30-day mortality was 6.1% in 3583 radically operated patients. Figure 2 Mortality in percent in 44 departments operating CRC (o=>100 operated patients) a = joint of all departments with less than 25 ptt. 25 % All operated patients (n=5187) All radically operated patients (n=3583) 20 15 10 5 0 a o o o o o o o o o o o o o o o o o o o o o o Five departments had 30-day mortality significantly different from average in all operated patients (Figure 3). When adjusted for case- mix (age, sex, urgency, tumour location, Dukes stage, ASAscore), four departments had significantly higher 30-day mortality than average, two of which also was significantly higher unadjusted (Figure 4). When only analysing the radically operated patients the numbers of significantly different departments are four unadjusted and three when adjusted for 21

case-mix. All had higher mortality than average. The overall variation between the departments was significant, except for the analysis of unadjusted radically operated patients. Table 4 - Base-line characteristics for 5187 patients operated for CRC. Variables n Sex 51.8% men 5187 Age Mean 70.2 years Median 72 (range 10-100) 5187 Excluded patients Mean 72.9 years, median 75 years 871 Dukes classification A 11.6% B 36.1% C 29.1% D 18.0% Unknown 5.2% 5129 Urgent operation 16.9% 5173 Tumour location Right colon 1 29.9% Left colon 36.4% Rectum 33.6% Unknown 0.1% 5187 ASA score I-II 68.1% III-V 22.3% Unknown 9.6% 5170 1 Includes the transverse colon 22

Figure 3 - OR for mortality unadjusted for case-mix in 44 departments (n=5187). a = all departments with less than 25 ptt. (n=49). *=significant. OR 8 7 * 6 5 4 3 2 1 0 a * * * * o o o o o o o o o o o o o o o o o o o o o o P=0.02 Figure 4 - OR for mortality adjusted for case-mix in 44 departments (n=5102). a = all departments with less than 25 ptt. or no mortality (n=48). *=sign. 8 OR 7 * 6 * 5 4 3 * * 2 a 1 0 o o o o o o o o o o o o o o o o o o o o o o P=0.01 23

Discussion Overall 30-day mortality after surgery for CRC in Denmark at 9.9% was more than in most other studies. Only one study presented a higher mortality [41]. No major limitations or differences in design were found to explain the high Danish mortality (Table 5). However our study is nationwide, including all patients in a 20-month period. None of the other studies were nationwide, but investigated a single region, area or department/surgeon. We believe that this can cause bias, as the geographical variations are not equalled out. This unintended selection bias may explain the difference in mortality rate. The variation in mortality between departments and counties in this study supports this explanation. Selected departments or counties have mortality rates as low as other studies. As illustrated in Figure 2 and 3 there was a significant variation in 30-day mortality between the operating departments. We expected this variation to disappear when adjusted for case-mix. This was not the case and the variation after adjustment remained significant. We adjusted for age, sex, urgency, tumour location, Dukes stage and ASA-score thereby compensating for departments operating more advanced cancers or patients with severe co-morbidity. Referring to Figure 1 we have adjusted for variables related the disease and the patient. Patient lifestyle variables though (smoking, alcohol intake, BMI, physical fitness), and factors relating to treatment and organisational structure are left unadjusted for because of the limited size of the patient population. These factors and factors not identified yet may have impact on the mortality too and possibly could have explained some of the variation. Conclusion The 30-day mortality after CRC surgery in Denmark is 9.9%. This seems higher than in other studies, but we have not found any nationwide comparable studies. There are significant differences 24

in the 30-day mortality between departments, even when adjusted for case-mix. Other variables not adjusted for may cause the variation. Table 5 30-day mortality in similar studies. Ref. n / country 30-day Notes mortality [42] CRC 18898 / CAN 3.8% a Only excision of segment. No recurrence. e [43] 7374 / UK 7.5% a e [44] 4562 / UK 8.4% a e [45] 3221 / UK 7.6% b e [46] 2452 / D 0.8% b e [47] 882 / UK 12.2% b e [48] 822 / TUR 4.5% b 20 non-adenocarcinoma [49] 555 / S 5% a e [50] 544 / Australia 2.9% a Only new patients [51] 501 / Ireland 4.4% b e [52] Colon 27986 / USA 4.6% a > 65 years. Only adenocarcinoma. Patients enrolled in health maintenance org. excluded (16.5%) e [53] 24166 / USA 4.6% c > 65 years. Only primary adenocarcinoma. Patients enrolled in health maintenance org. excluded (16%). Hosp/surg. spec. volume is defined otherwise exc. e [54] 22408 / USA 2.8% d Only colon resection, no recto sigmoid tumours [55] 5853 / USA 5.7% a Colectomy and primary anastomosis e [56] 2264 / D 5.0% a e 25

[57] Rectum 2815 / USA 3.3% a 65 years or older e [58] 1399 / S 3.4% a No patients older than 79 years in half material. Randomised study with radiation (2 methods) e [59] 1115 / D 2.8% a e [60] 591 / USA 3.2% a Proctectomy e [61] 1978 / F 2.5% a a Death within 30 days of surgery b In-hospital death or death within 30 days of surgery c Death within 30 days of hospitalization (proxy date for surgery) d In-hospital death e Only operated patients Lifestyle and 30-day complications to surgery for CRC (III). BMI at or above 30kg/m 2 was significantly associated with increased risk of complications in general, deep wound infection, impaired wound healing and 30-day mortality. Smoking was significantly associated with increased risk of complications in general, impaired wound healing and 30-day mortality. Former smokers endured the same increased risks except increased mortality. Drinking more than 60 g alcohol/day was significantly associated with increased risk of impaired wound healing and anastomotic leakage. No alcohol intake significantly increased the risk of mortality within 30 days. We found no interaction between alcohol and smoking. Self Perceived Physical Fitness (SPPF) below average was significantly associated with increased risk of complications in general, thrombosis and 30-day mortality. Average SPPF significantly increased the risk of thrombosis too. The risks are shown in Table 6 and did not change substantially when adjusted for other factors related to the patient, disease or treatment. The minimal-risk population consisted of 651 patients (18.7%). Significantly less of these patients had complications in general (22.0% vs. 30.0%, p<0.0001) and died within 30 days (1.4% vs. 4.5%, p=0.0002). If all patients had a minimal-risk lifestyle, assuming causality between lifestyle factors 26

and complications or mortality, there would have been 22.8% less complications and 64.0% less mortality within 30 days (attributable risks). Discussion This first nationwide study supports that high BMI, smoking and alcohol habits are risk factors for complications and 30-day mortality after colorectal cancer surgery. Furthermore the study suggests SPPF as a risk factor. Concerning smoking, BMI and alcohol habits there are theoretical and physiological explanations for the effect [14,18,62-69]. Combined with the increasing number of Table 6 Associations between various endpoints and lifestyle factors. Complications in general 1 30-day mortality 2 Impaired wound healing 3 Anastomotic leakage 4 Deep wound infection 5 Thrombosis 6 Variables OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) BMI < 30 kg/m2 1 1 1 1 1 1 = 30 kg/m2 1.60 (1.03 1.93) 1.24 (1.48-2.29) 1.34 (1.00-1.80) 1.23 (0.78-1.95) 2.81 (2.13-3.70) 1.38(0.88-2.17) SMOKING HABITS Never smoked 1 1 1 1 1 1 Ex-smoker 1.23 (1.03 1.47) 1.80 (0.87-3.73) 1.50 (1.07-2.09) 1.19 (0.73-1.92) 0.99 (0.69-1.42) 0.94(0.61-1.45) Smoker 1.35 (1.01 1.80) 2.00 (1.04-3.85) 1.43 (1.23-2.70) 1.44 (0.88-2.35) 0.94 (0.62-1.41) 0.93(0.60-1.44) ALCOHOL CONSUMPTION Nothing 1.06 (0.85 1.33) 2.13 (1.26-3.61) 0.92 (0.68-1.24) 0.81 (0.50-1.31) 0.95 (0.63-1.43) 1.01(0.68-1.50) 1 5 units/day 1 1 1 1 1 1 > 5 units/day 1.75 (0.99 3.11) 2.33 (0.80-6.74) 2.42 (1.26-4.63) 2.48 (1.07-5.77) 1.07 (0.48-2.40) 2.18(0.64-7.38) SELF PERCIEVED PHYSICAL FITNESS (SPPF) Above average 1 1 1 1 1 1 Average 1.13 (0.91 1.40) 1,28 (0.64-2,56) 1.14 (0.85 1.52) 1.04 (0.66-1.64) 1.07 (0.77-1.50) 1.82 (1.09 3.03) Below average 1.52 (1.15 2.01) 2,51 (1.34-4,69) 1.17 (0.89 1.55) 0.74 (0.40-1.39) 1.35 (0.96-1.89) 2.79 (1.73 4.51) 1 n=2605. Adjusted for age, sex, ASA and blood loss during operation. 2 n=2757. Adjusted for age, sex and ASA. 3 n=2763. Adjusted for age, sex and blood loss during operation. 4 n=1988. Adjusted for age, sex, urgency and blood loss during operation. 5 n=2944. Adjusted for age, sex and tumour location. 6 n=2944. Adjusted for age and sex. None of the disease or treatment related factors from Table 4 were significant 27

studies confirming the impact on outcome [11-18], they soon should be a mandatory part of the case-mix. SPPF is a proxy for actual physical fitness [70]. The mechanisms of the impact on outcome after surgery is various but the documentation in CRC-surgery is inadequate. Our results indicate an effect on complications in general, thrombosis and 30-day mortality, but further research is needed. The analyses included data on the disease, the treatment and patient including lifestyle. As we used logistic regression this makes the analyses very vulnerable for missing data; the patient was not included if just one variable was missing. As the completeness of the lifestyle data was 58.6%, the presented analyses include between 43.2% and 69.3% of the patients. The missing patients are older and in a higher ASA-class than the included, making it reasonable to assume, that the missing patients are more affected by their co-morbidity than those included. Thereby it is improbable that the missing patients should have less exposure to lifestyle factors than the included. We therefore believe that the missing patient will cause an underestimation of the risk and not counterbalance them. Out study only includes a subgroup of patients, but we do not believe the effect to be limited neither to CRC surgery nor the analysed complications in this study. Other studies including other subgroups of surgical patients find similar results [71]. Conclusion The risks we estimated combined with the attributable risks indicate that lifestyle factors have a substantial impact on outcome after CRC surgery. Impact of structural Factors on outcome after surgery for CRC (IV). In colonic surgery the presence of postoperative clinical guidelines (observation, mobilisation, or nutrition) was associated with a decreased risk of complications and mortality. One of three guidelines significantly decreased the risk of mortality and two of three guidelines significantly 28

decreased the risk of complications. No specific guideline accounted for the difference. No other structural variables had significantly impact on the outcome after colonic surgery. In rectal surgery a high mean surgical volume (of the surgeon) was significantly associated with less complications and an insignificant trend towards lower 30-day mortality. No other structural Table 6 Structural factors and risk of 30-day complications/mortality - Colon surgery % 30-day mortality 1 Complications 1 Variable n OR CI n OR CI Transferral of patients Yes 4.8 134 1 134 1 No 95.2 2657 1.05 0.50 2.21 2652 0.88 0.61 1.27 Specialized subdepartment No 34.4 959 1 958 Yes 65.6 1832 1.23 0.84 1.81 1828 1.07 0.80 1.43 Postoperative guidelines None 20.1 561 1 561 1 1 of 3 6.9 192 0.62 0.41 0.94 192 0.71 0.49 1.02 2 of 3 7.6 212 0.97 0.52 1.79 212 0.70 0.54 0.89 3 of 3 65.4 1826 0.87 0.58 1.31 1821 1.12 0.88 1.42 Colonoperations/surgeon per year 1 10 36.5 1020 1 119 1 11-20 40.2 1122 0.79 0.54 1.17 1121 0.87 0.63 1.20 20+ 23.3 649 1.02 0.72 1.45 646 1.22 0.93 1.60 1 Adjusted for age, sex, ASA-class, side of Colon, urgency and type of operation, and whether the operation was palliative. 29

variables were associated with complications or mortality. In the population of stoma patients the presence of a specialized nurse did not affect the mortality. The specialized nurse was significantly associated with increased risk of complications (OR 3.56 (2.29-5.53)), even when adjusted for patient, disease and structural related variables (age, sex, ASA-class, tumour location, urgency and type of operation, palliative operation, specialized subdepartment, postoperative guidelines). There was no difference in the stoma related complications in the groups with or without a specialized nurse (1.30 vs. 1.29, p= 0.99). Discussion This study found, that the only structural factor of impact on outcome after CRC surgery is the volume of the rectal surgeon. This is confirmed by other studies [72-75], as well as the missing association between volume and colonic surgery [75,76]. Other non-crc studies find an association between outcome and guidelines [77,78] we partly confirm this, as most estimates are insignificant. The increased risk of complications when a specialized stoma nurse was present we believe to be a consequence of a small reference group in the analyses (n=26). Structural demands play an important role in quality assurance and accreditation. We therefore expected more impact of the variables than we found. Part of the explanation for the missing associations may be that Denmark is a small, homogeneous country. The differences we compare are therefore too small for a possible effect on outcome to be shown. We do not believe our choice of structural variables to be irrelevant, but it is possible that other factors of impact are left out, or that we defining the variables chose definitions too unspecific to show any impact. 30

Conclusion The structural factors studied are with the exception of the rectal surgeons volume of limited impact on the outcome after CRC surgery. Table 7 Volume of rectal surgeon and risk of 30-day complications/mortality 30-day mortality 1 Complications 2 Variable n OR CI n OR CI Rectaloperations/surgeon per year 1 5 6 10 2.4 11.0 36 167 1 3 1.08 0.62 1.87 36 167 1 0.51 0.33 0.78 11 15 13.9 211 0.73 0.41 1.28 210 0.56 0.36 0.79 16 20 40.1 607 0.82 0.58 1.16 605 0.55 0.34 0.89 21+ 32.6 494 0.57 0.31 1.05 494 0.50 0.30 0.82 1515 1509 1 Adjusted for age and sex, ASA-class, urgency, and whether the operation was palliative. The analysis was univariate, as the structural variables were introduced separately. 2 Adjusted for age, sex, ASA-class, urgency and type of operation, and whether the operation was palliative. 3 Test for trend not significant 31

OVERALL CONCLUSION The completeness and data quality of the DCCG-database is sufficient for further analyses. There are significant differences in the 30-day mortality between the departments not explained by differences in case-mix. The patient s lifestyle (smoking, alcohol consumption, SPPF and BMI) has impact on the risk of mortality and morbidity. Finally it is suggested that structural factors, except for the volume of the rectal surgeons are of limited impact on the outcome. 32

PERSPECTIVES The DCCG-database was initiated because the literature suggested that mortality after surgery for CRC in Denmark was higher than in other countries. For further evaluation of this, a systematic monitoring is necessary both in Denmark and in comparable countries. Monitoring can be performed by means of clinical databases which are important tools in quality assurance. The clinical databases need to have a high rate of completion of data without any selection. It is therefore important that all departments are committed sending complete data. This can be facilitated by feedback of data to the departments, giving a sense of ownership and by respecting the enormous workload it is to collect data. When presenting results from the database it is important to remember, that the purpose is to improve the quality in general as well as locally in departments apparently enduring low quality of care. By performing a systematic registration of all patients treated for CRC it is possible both to monitor the quality and to compare departments. In Denmark we have no tradition for this and much dispute is taking place among professionals and administrators whether this comparison is fair. This thesis shows two important aspects of a clinical database. - Through clinical epidemiological studies it is possible to identify potential risk factors for treatment related morbidity and mortality. - It is possible to analyse and compare the quality of treatment between departments adjusting for variables known to have impact. These two aspects are closely related, as the risk factors found in the first can be used in the second. The risk factor should be tested in regular controlled clinical trials if possible. This leads to two new questions: how to avoid or alleviate the effect of the risk factors? how many risk factors should we include in our analyses? 33

The first problem seems to be simple although the answer may not be. An example is smoking as a risk factor. It seems obvious that the patients must cease smoking when diagnosed before the operation, but studies show that this does not affect the postoperative complications [79]. The solution therefore probably is to develop treatment regimens compensating for the increased risk, and to make prevention programs decreasing the smoking part of the population. The second problem is more complicated. In this thesis including known case-mix factors did not change the significant difference between the departments. The departments with a mortality above average without adjustment remained above average after adjustment and vice versa. It was not possible to include lifestyle data in this comparison of departments though this may have explained some of the variation. As completeness of the patient questionnaire increases, it will be possible to include lifestyle factors in the case-mix. The number of adjusting variables included is only restricted by the number of patients in the database, and the final goal could be to adjust for everything but the quality of the treatment. But this ideal goal is not without problems. If a department differentiate the treatment according to the risk profile of each patient, is it then fair to adjust for the risk factors, or is it necessary to take an eventual intervention against them into account? Which factors must be included for the comparison to be meaningful? And will it ever be fair? These questions are important in relation to comparing the quality in the departments. In the search for this answer the database is a necessary tool. Another perspective in finding risk factors is the increasing possibility to predict the outcome of an operation, thereby deciding whether to operate or not. This ability implies practical and moral problems as well as psychological issues. Setting the threshold for the risk of mortality and morbidity leading to no operation will be difficult, even without not considering the consequences. And is the patient committed to reduce the lifestyle factors increasing the risks of complications. These problems are still in the future, but the need for decisions will increase in the years to come. 34

SUMMARY Through the nineties there was a debate on the quality of the treatment of CRC (Colorectal Cancer) in Denmark. Statistics showed that the survival was lower compared to that in other Nordic countries. This lead to the foundation of the DCCG-database (Danish Colorectal Cancer Group), including all CRC patients with a first time adenocarcinoma. This systematic registration of all patients makes it possible to monitor and improve the treatment and to compare the departments. The aims of this thesis were to evaluate the validity and completeness of the DCCG-database, to compare 30-day mortality between the departments with and without adjustment for case-mix, and to identify lifestyle factors with impact on 30-day mortality and four common complications. Finally associations between structural factors in the departments and 30-day complications or mortality were studied. Initially a 5% (n=87) sample of patients from 2001 was re-entered to the database using the patient records from the departments. The validity was satisfying with mean kappa values between 0.55 and 0.67. The data quality was considered sufficient for further analyses. Data originated from the national DCCG-database, the National Patient Registry, the Central Office for Civil Registration and a questionnaire concerning structural factors sent to all departments. All patients in the database operated between 1 May 2001 and 31 December 2002 were included (n=5187) and the database had a completeness of 89.2%. The mean 30-day mortality after CRC operation was 9.9%. This was a higher rate than in most other studies, but no comparable nationwide study was found, and the difference may be explained by selection bias. Significant differences in 30-day mortality were found between the departments operating colorectal cancer (CRC) with and without adjusting for case-mix. Some of the differences might be caused by differences in case-mix factors not included, primarily lifestyle-factors. 35

One or more of the lifestyle factors (Body Mass Index at or above 30kg/m 2 (BMI), smoking, drinking more than 60 g alcohol/day, self perceived physical fitness (SPPF)) had significant impact on 30-day mortality, complications in general or one or more of four complications (impaired wound healing, anastomotic leakage, deep wound infection, and thrombosis). In a population with no exposure to smoking, BMI below 30 kg/m 2, alcohol intake below 60 gram/day and SPPF at or above average there would be 23% less complications and 64% less mortality within 30 days. Although the analyses include only 43 69% of the data caused by low completeness of the patient data, we believe the lifestyle factors to have substantial impact on the outcome after CRC. Of the analysed structural variables, only the mean surgical volume of the surgeon had significant impact on the morbidity in rectal surgery. No structural factors were substantially related to mortality and morbidity in colonic surgery. The authors were surprised to find such low impact of structural factors, as these factors are important in accreditation and quality improvement. Possible explanations to the results are that Denmark is a small country with very homogeneous structure for the treatment of CRC leaving no room for differences, or that our choice of factors was irrelevant or sub optimally defined. This thesis looks into factors with impact on the significant difference in 30-day mortality in the departments. It focuses on associations between the patient, the treatment, the organisation and the outcome. Structural factors had except from surgical volume of the rectal surgeon no substantial impact, in contrast to lifestyle factors. The thesis illustrates the possibilities to analyse and compare departments and to identify potential risk factors for mortality/morbidity by means of clinical databases. The importance of high validity and completeness is discussed. 36

DANISH SUMMARY I 1990 erne var der meget debat om kvaliteten af den danske behandling af kolorektal cancer (CRC). Mortalitetsstatistikker viste dårligere overlevelse i Danmark end i de øvrige nordiske lande. Dette ledte til grundlæ ggelsen af DCCG-databasen (Danish Colorectal Cancer Group), som registrerer alle CRC-patienter med en førstegangs adenocarcinom diagnose. Formålene med denne ph.d. tese var at validere DCCG-databasen både med hensyn til dækningsgrad og datakvalitet, at sammenligne 30-dags mortaliteten mellem alle danske afdelinger justeret og ujusteret for case-mix samt at undersøge om livsstilsfaktorer har betydning for mortalitet og fire udvalgte hyppige komplikationer. Endelig skulle sammenhængen mellem strukturelle forhold på afdelingerne og 30-dags komplikationer og død undersøges. Indledningsvist blev validiteten af data undersøgt i et randomiseret udtræk på 5% af patienter opereret i 2001. Analyserne på udtrækket viste gennemsnits kappa værdier mellem 0.55 og 0.67, og validiteten blev vurderet sufficient for de planlagte yderligere analyser. Data til analyserne stammede fra DCCG-databasen, Landspatientregisteret, CPR-registeret og et spørgeskema om strukturelle faktorer sendt til alle afdelinger. Alle patienter opereret i perioden 1. maj 2001 til 31. december 2002 blev inkluderet fra DCCG-databasen (n=5187), og dækningsgraden var 89.2%. Blandt alle opererede patienter var 30-dags mortaliteten 9.9%. Dette var højere end i de fleste andre studier, men der blev ikke fundet nogen direkte sammenlignelige landsdækkende opgørelser, og den højere danske mortalitet kan skyldes selektions bias. Der blev påvist signifikant forskel på 30-dags mortaliteten afdelingerne i mellem, både med og uden justering for case-mix. En del af forskellen kan formentligt tilskrives faktorer som ikke blev inddraget i case-mix, primært livsstilsfaktorerne. En eller flere af livsstilsfaktorerne (BMI på eller over 30kg/m 2 (Body Mass Index), rygning, alkoholforbrug, selv vurderet fysisk form (SPPF)) havde signifikant betydning for 30-dages 37

mortaliteten, komplikationer generelt og/eller en eller flere af fire hyppige komplikationer (nedsat sårheling, anastomoselækage, dyb sårinfektion og thromboser). I en population uden påvirkning af risikofaktorer (ingen rygning, BMI under 30 kg/m 2, alkohol på eller over 60 g/dag (5 genstande), SPPF på eller over middel) ville der være 23% færre komplikationer og 64% lavere 30-dags mortalitet (ætiologisk fraktion). Selv om analyserne kun inkluderer mellem 43% og 69% af patienterne på grund af lav dækningsgrad af patientspørgeskemaet, anser vi livsstilsfaktorer for at have en betydelig effekt på resultater af CRC-kirurgi. Af de analyserede strukturvariable havde kun det gennemsnitlige antal operationer pr. kirurg betydning for udfaldet af rektumkirurgi. Ingen strukturfaktorer havde nær samme betydning indenfor kolonkirurgi. Det var ikke ventet med så ringe en effekt af strukturvariablene, da de udgør en vigtig del af akkreditering og kvalitetsudvikling. En mulig forklaring er, at Danmark er et lille land med en så stor ensartethed af behandlingen på CRC-området, at forskellene er for små til at kunne analyseres. En anden mulighed er, at vores valg og definition af de analyserede faktorer ikke var optimalt. Denne tese undersøger faktorer af betydning for den signifikante forskel på 30-dags mortaliteten mellem afdelingerne. Den fokuserer på sammenhængen mellem patient, behandling, organisation og resultat. Strukturelle faktorer havde fraset antal operationer pr. rektalkirurg, kun ringe betydning, modsat livsstilsfaktorer. Tesen illustrerer mulighederne for en klinisk database med at analysere og sammenligne afdelingers resultater, samt identificere nye mulige risikofaktorer for død og komplikationer. Vigtigheden af en høj dækningsgrad og datakvalitet diskuteres. 38