Clinical Predictors for Recurrence after Curative Resection for Colorectal Cancer

Purpose: To identify clinical predictors of recurrence of colorectal cancer after curative surgical treatment. Methods: Retrospective follow-up-study of 925 consecutive patients treated with R0-resection for colorectal cancer Stage I, II and III from 1990 until 2000 with a mean follow-up of 60 ± 37 months. Predictors for cancer recurrence were identified in a pilot-sample of these patients, followed by analyses of the rest of the patients (test-sample), and finally with a concluding analyses of the entire patient group. Data were analyzed with Pearson Chi-square test (χ), Cox regression analyses and log rank test. Results: Tumor stage (Stage I: HR 0.10 (0.05; 0.19), Stage II: HR 0.31 (0.24; 0.41)) and postoperative reoperations due to complications due to other causes than anastomotic leakage (HR 2.02 (1.21; 3.36)) were significant predictors of cancer recurrence in the multivariate Cox regression model. The association between reoperations and recurrence was strongest for the patients with the best prognosis: Stage I and Stage II-cancers. Long duration of surgery, strongly associated with blood-loss and infusions of liquid and blood-products, reoperation due to anastomotic leakage as well as right colon/transversum localization were significant at a trend-level (10%). Conclusions: Tumor stage and reoperations due to postoperative complications other than anastomotic leakage are significant predictors for recurrence after curative surgery for colorectal cancer.


Introduction
Cancer recurrence occurs in about 30% of patients after curative resection for colorectal cancer [1].The main predictor for recurrence and survival is the tumor stage at the time of surgery.Previous studies have identified several possible additional factors associated with recurrence.Patient factors, like intake of coffee and alcohol, smoking, weight-loss and Glascow Prognostic score [2] [3] [4] [5] [6], treatment factors like surgical experience and technique [7] [8], tumor factors like location, differentiation, vein-, nerve-and lymphatic ingrowth [9] [10] and molecular factors [11] [12] [13] [14] have been shown to influence cancer recurrence, and adjuvant oncological therapy is often recommended.In the present study, we have taken interest in what consequences repeated or aggravated inflammatory stress will inflict on the patient and influence the risk of recurrence after potential curative resection of colorectal cancer.
We have selected clinical parameters related to repeated or exaggerated inflammatory stress, which are available at all institutions.Previous case reports have suggested that sites of inflammation and traumatized tissue are more susceptible for cancer cell deposits and growth than other sites; a phenomenon named "inflammatory oncotaxis" [15] [16] [17].The aim of the present study was to define the independent predictive value of each of these parameters: emergency resection due to bowel obstruction, tumor perforation, exaggerated blood loss/transfusions and postoperative complications in need of reoperation.So far few studies evaluated which of these potential parameters is the most predictive.
In this time period, adjuvant chemotherapy for stage III colonic cancers and preoperative radiotherapy for rectal cancers with tumors near the circumferential margin for resection was not routinely used, and very few patients (<5%) received these treatment modalities in our study.Thus, we have the opportunity to investigate the influence of clinical factors without the need to correct for these treatment modalities.The hypothesis in the present paper is that exaggerated or repeated inflammatory stress will promote cancer recurrence.

Materials and Methods
1285 consecutive unselected patients admitted with colorectal cancer at Haukeland University Hospital, Norway, from January 1990 until January 2000, were included in the present retrospective observational study.To obtain a group of patients with verified R-0-resection, patients with stage IV-tumor, non-resected patients, patients who died in hospital, R1-or R-2-resceced patients, and patients lost to follow-up were excluded, leaving 925 patients for the study (Figure 1).The patients were routinely followed-up for five years with ultrasonography  multivariate analyses, but some tumor characteristics are presented for the sake of comparison with other studies.Perioperative blood-loss, blood-transfusions, and intravenous fluid transfusions were not included as specific factors in the multivariate model, as they were all highly correlated with duration of surgery, which was selected as an indicator for all these factors (Table 1).

Statistical Analyses
Demographic and clinical characteristics are presented as mean and standard deviation (SD) or frequencies and percentages.Student's t-test or Pearson's Chi-square test (χ 2 ) were used as appropriate for comparison of groups of patients.The association between cancer recurrence and a number of potential predictors and confounders was assessed by Cox regression analysis.As there were several surgeons performing the operations, intra-surgeon correlations were assessed by intra-class correlation coefficient, however no such correlations were found and thus no adjustments were needed.
When assessing predictors, a data splitting approach was applied to avoid hypothesis fishing in the dataset [18].According to this approach, the data were split into two parts.The first part (Pilot) was used to search for significant predictors in a univariate model, and then construct a multivariate Cox regression model containing only significant and/or clinically important predictors.A formal hypothesis testing was then performed in the second part of the data set (Test).A 5% level of significance was used and only the results with p-values below this limit were accepted as significant, regardless of significance level in the pilot part.Finally, analyzes were performed on the entire data set.Proportional hazards assumption was assessed by the correlations between the ranking of individual time of metastases and Schoenfeld's residuals generated from a multivariate Cox regression model for each covariate.The results were presented as hazard ratios (HR) with the corresponding 95% confidence intervals (CI) and p-values.The effect of tumor stage and reoperations on cancer recurrence was tested by a log-rank test and illustrated using Kaplan-Meyer-plots.
The location of cancer recurrence(s) according to location of the primary tumor is shown in Table 5.The liver was the most frequent organ for cancer recurrence, accounting for 42% of all recurrence.Left sided colonic cancers did have higher rates of liver metastases, right sided colonic cancers did have higher rates of lung metastases, whereas rectal cancers had higher rates of local recurrences.*27 patients had recurrence in more than one location, three patients had tumors in more than one location, and are omitted.
The pilot dataset was employed to generate hypothesis regarding significant predictors of recurrence.In the multivariate model, age, indication for surgery (acute/planned), TNM-stage, reoperations for other causes than anastomotic leak and tumor location in the left colon were significant independent predictors (Table 6).
The test data set was used to test the hypothesis generated in the pilot set.In the multivariate model estimated on the test sample, age, gender and surgeon type were included as confounders.Indication for surgery (acute/planned) was not significant a predictor, whereas duration of surgery and tumor location (right colon/transversum) were significant on trend-level (10%) only (Table 7).
Also, pre-operative and not per-operative tumor-perforation and postoperative reoperation due to other reasons than anastomotic leakage were significant predictors.Tumor stage remained highly significant.
In the final analyses, the entire data-set was assessed, as presented in Table 8.
Indication for surgery (acute/planned), duration of surgery, TNM-stage, reoperation for complications other than anastomotic leakage and location of tumor were all significant in the multivariate model, whereas tumor perforation was not.However, the overall interpretation of the results using this methodological approach is as follows: "Indication (acute/planned)" was significant in pilot part, but non-significant in test part and thus non-significant predictor in the entire data set even though p = 0.013 in Table 8. "Tumor-perforation" was non-significant in pilot part, but left in the model as potential predictor.It became significant in test part, but with an opposite "direction" in HR.However, it was not a significant predictor in the entire data set."TNM-stage" is clearly a significant predictor (Table 8), also illustrated in Figure 2. "Duration of surgery" was not significant predictor in pilot part, but was left in the model as important predictor.It was only significant on trend level in test part.Even though it is highly significant in the entire data set, we can only conclude the significance at trend level (10%), as was localization in colon."Postoperative reoperation due to other causes than anastomotic leak was significant in both pilot and test sets.In the entire data-set,  The analyses shows that the negative influence of reoperation is most pronounced in the patient groups with the assumed best prognosis: Stage I and II.

Discussion
Previous studies have shown that postoperative complications like anastomotic    recurrence rates and reduces survival [19]- [26].Few reports have studied the effect of reoperations due to other causes.We therefore decided to separate this group from the group of patients reoperated for anastomotic leaks in the Cox model.The results showed an association between reoperation for anastomotic leaks and cancer recurrence only on trend-level (10%), which may be explained from the small number of patients in this group, and lack of statistical power.
The association between reoperation for other causes than anastomotic leaks and cancer recurrence was highly significant, which is the most interesting finding in this study.The negative effect of reoperations was most pronounced in patients with the best chances of cure according to stage: stage I and stage II.
Emergency surgery, location of the tumor, duration of surgery and tumor perforation was not significant independent predictors of recurrence.This is in contrast with previous studies, which have identified emergency surgery [27], perioperative bleeding and transfusions of blood as risk factors for recurrence [27] [28] [29] [30] [31].In one study, blood transfusions > 3 units erythrocytes increased the recurrence rates, probably through T-lymphocytes, NK-cells, CD4+/CD8+ ratio [31].The main discussion has been whether it is the bleeding itself, or the transfusion of blood that is the main cause of recurrence.However, some studies do not show negative effects of blood transfusions after colorectal surgery [32].Blood transfusions will also not influence recurrence rates after gastric cancer surgery [33].The effect of blood transfusions is thus still not settled.
Previous studies have shown that surgery for bowel obstruction and tumor perforation during surgery is a risk factor for recurrence and reduced 5 year survival [2] [10] [34] [35] [36], possibly due to performance of inadequate surgery in a life-threatening situation, or spillage of cancer cells to the abdominal cavity.
Cancer cells with the ability to grow have been demonstrated after colorectal cancer resections from luminal mucosa, rectal stump, serosal surface washings and post-dissection lavage tumor bed [37].This was not supported in the present multivariate model.
We did not find any significant overall difference in recurrence rates between colon and rectal cancers, which is in contrast to a pervious Dutch study [9].
There was also no difference in time to cancer recurrence or location of cancer recurrences in patients with or without reoperations for postoperative complications (data not shown).However, left sided colonic cancers did have higher rates of liver metastases, right sided colonic cancers did have higher rates of lung metastases, whereas rectal cancers had higher rates of local recurrences.This is in accordance with a study on 10.398 colorectal cancer survivors by Augestad et al. [38].
The statistical study design of a pilot-test-sequence instead of analyzing the whole sample was chosen to eliminate the chance of bias in selecting predictors, which is strength of the study.However, the same design also did reduce the power in the two sub-analyses (pilot and test), and have increased the risk of a type II-error, which is a weakness.This may explain why both emergency surgery and duration of surgery were significant in the overall multivariate analyses, but were interpreted as non-significant.
The retrospective method may be a limitation of the study.Seventeen patients were lost to follow-up, which may be another limitation.Other investigators have also taken interest in the importance of clinical factors on recurrence, as is discussed in the text.
The purpose of the present study was to evaluate the impact of several clinical parameters reflecting inflammatory stress on cancer recurrence simultaneously, and identify the factors of independent importance.We could demonstrate that surgical complications in need of reoperation due to other causes than anastomotic leak, was followed by higher recurrence rates, especially in the patients with the best prognosis (stage I and stage II).This is a strong argument for meticulous preoperative conditioning of the patient, and meticulous surgical technique.

Conclusion
Tumor stage and postoperative reoperations due to complications are significant predictors for recurrence after curative surgery for colorectal cancer.

Disclosure of Potential Conflict of Interest
We report no conflict of interest.

Figure 1 .
Figure 1.Patients included in the study: Consort Diagram.
other causes than anastomotic leak was a highly significant predictor for cancer recurrence.Reoperation due to anastomotic leakage was not a significant predictor in the pilot and test sets, and only significant at a trend-level (10%) in the entire dataset.The effect of postoperative reoperations on cancer recurrence is also shown for all patients (Figure3(a)) and for patients with tumor stage I (Figure 3(b)), stage II (Figure 3(c)) and stage III (Figure 3(f)).

Figure 2 .
Figure 2. Time to first cancer recurrence according to tumor stage (log rank: p < 0.001).

Figure 3 .
Figure 3. (a) Time to first cancer recurrence according postoperative reoperation (log rank: p = 0.002); (b) Time to first cancer recurrence according postoperative reoperation and tumor stage I (log rank: p = 0.001); (c) Time to first cancer recurrence according postoperative reoperation and tumor stage II (log rank: p = 0.019); (d) Time to first cancer recurrence according postoperative reoperation and tumor stage III (log rank: p = 0.717).

Table 1 .
Association between duration of surgery, peroperative bleeding, and volume of liquids-and blood-infusions.

Table 5 .
Distributions of cancer recurrence.

Table 6 .
Results of Cox regression analysis in the pilot part, N = 307.

Table 7 .
Results of multivariate Cox regression analysis in the test part, N = 591.

Table 8 .
Results of Cox regression analysis performed on the entire data set (N = 898).
1Controlled for age, gender and type of surgeon. 2 HR reported for every 1-minute difference in duration.