Clinical determinants of survival in patients with 5-fluorouracil- based treatment for metastatic colorectal cancer: results of a multivariate analysis of 3825 patients

C.-H. Köhne1,+, D. Cunningham2, F. Di Costanzo3, B. Glimelius4, G. Blijham5, E. Aranda6, W. Scheithauer7, P. Rougier8, M. Palmer9, J. Wils10, B. Baron11, F. Pignatti11, P. Schöffski12, S. Micheel13 and H. Hecker12

1Medizinische Klinik und Poliklinik I, Universitätsklinikum Carl Gustav Carus der TU-Dresden, Dresden, Germany; 2The Cancer Research Company Section of Medicine and the GI Unit, The Royal Marsden Hospital, Sutton, UK; 3Department of Internal Medicine and Oncology, Azienda Ospedaliera S. Maria, Terni, Italy; 4University Hospital, Department of Oncology, Uppsala, Sweden; 5Academisch Ziekenhuis Utrecht, Department of Internal Medicine, Utrecht, The Netherlands; 6Hospital Universitario ‘Reina Sofia’, Cordoba, Spain; 7University Hospital, Department of Internal Medicine, Vienna, Austria; 8Hôpital Ambroise Paré, Service de Hepato-Gastroent. Concologie Digestive, Boulogne Cedex, France; 9ASTRA-ZENECA Pharmaceuticals, Macclesfield, UK; 10Laurentius Hospital, Roermond, The Netherlands; 11EORTC Data Center, Brussels, Belgium; 12Medical School Hannover, Department of Internal Medicine, Hannover; 13Robert-Rössle-Clinic Charité, Campus Berlin-Buch, Berlin, Germany

Received 6 July 2001; accepted 27 August 2001.


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Background

Patients with metastatic colorectal cancer are usually offered systemic chemotherapy as palliative treatment. A multivariate analysis was performed in order to identify predictors and their constellation that allow a valid prediction of the outcome in patients treated with 5-fluorouracil (5-FU)-based therapy.

Patients and methods

A total of 3825 patients treated with 5-FU within 19 prospective randomised and three phase II trials were separated into learning (n = 2549) and validation (n = 1276) samples. Data were analysed by tree analysis using the recursive partition and amalgamation method (RECPAM). A predictor could only enter the RECPAM analysis if the number of patients with missing values was <33.3% within a node, and the minimal node size was set to 50 patients. Twenty-three potential predictors were grouped into subsets of laboratory variables (11 parameters), tumour-related variables (seven parameters) and clinical variables (five parameters). In the first step, tree analysis was performed separately for each predictor subset. The selected prognostic parameters of the resulting partial models (the ‘winners’) were entered into the general model. The classification rule from the data of the learning set was applied to the independent validation set.

Results

Winners of the subgroup analysis for laboratory variables were: platelets >=400 x 109/l, alkaline phosphatase >=300 U/l, white blood cell (WBC) count >=10 x 109/l and haemoglobin <11 x 109/l, and all predicted a worse outcome. Negative predictors within the subgroup of tumour parameters were: number of tumour sites more than one or more than two, presence of liver metastases or peritoneal carcinomatosis, which predicted a worse outcome. Furthermore, presence of lung metastases, a primary rectal cancer and presence of lymph node metastases all predicted a better outcome in the multivariate setting. Among the clinical parameters only performance status of ECOG 0 or 1 predicted better outcome. In the final regression tree, three risk groups could be identified: low risk group (n = 1111) with a median survival of 15 months for patients with ECOG 0/1 and only one tumour site; intermediate risk group (n = 904) with a median survival of 10.7 months for patients with ECOG 0/1 and more than one tumour site and alkaline phosphatase <300 U/l or patients with ECOG >1, WBC count <10 x 109/l and only one tumour site; high risk group (n = 534) with a median survival of 6.1 months for patients with ECOG 0/1 and more than one tumour site and alkaline phosphatase of >=300 U/l or patients with ECOG >1 and more than one tumour site or WBC count >10 x 109/l. The median survival times for the good, intermediate and high risk groups in the validation sample were 14.7, 10.5 and 6.4 months, respectively.

Conclusions

Patients can be divided into at least three risk groups depending on the four baseline clinical parameters: performance status, WBC count, alkaline phosphatase and number of metastatic sites. Any molecular or biological marker should be validated against these clinical parameters and decisions for more or less intensive treatments may be studied separately in these three risk groups. Also, clinical trials should be stratified according to the three risk groups.

Key words: colorectal neoplasm, metastatic disease, multivariate analysis, prognosis


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Cancer of the large intestinum is one of the leading causes of cancer deaths in elderly men and women [1]. Approximately 50% of patients will eventually die due to metastatic spread of their disease. Compared with patients receiving best supportive care only, chemotherapy prolongs the median survival of patients with metastatic disease [2]. Therefore, palliative chemotherapy is offered to these patients to ameliorate symptoms, preserve the quality of life and improve survival.

Clinical trials, although using similar patient selection criteria, often display a surprising heterogeneity in survival rates [3], which is usually explained by differences in patient characteristics or prognostic factors. Patients entering randomised trials are usually stratified according to their performance status, but in addition other prognostic factors or their constellation may have the potential to determine the survival of patients to a greater extent than any promising antineoplastic agent or drug combination.

A variety of clinical parameters such as performance status [4], elevated lactate dehydrogenase, white blood cell (WBC) count [5], serum albumin [4], elevated liver transaminases [6], level of haemoglobin [7] or platelets, pathological grading [8] or localisation of the primary [9], or tumour markers like carcinoembryonic antigen (CEA) [8] have been identified as prognostic markers in some studies that seldom included more than 400 patients. There is no consensus and general acceptance about the importance of various prognostic factors. We collected a large database of patients with metastatic colorectal cancer who were eligible for clinical trials and treated with a 5-fluorouracil (5-FU)-based regimen, in order to identify predictors and their constellation that would classify patients according to their predictor values into a few clearly separated groups with different risks.


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Data collection
European study coordinators of large clinical trials were contacted and source data of 3825 patients from 19 randomised and three phase II studies were collected [7, 10–29]. The trials recruited patients from 1982 until 1996 (Table 1) and investigated different 5-FU-based schedules. Data for each study were checked for consistency after transformation to the main database.


View this table:
[in this window]
[in a new window]
 
Table 1. Trials included in 10 analyses
 
Three trials compared 5-FU/leukovorin against raltitrexed [27–29], but only patients who received the 5-FU-containing arm were selected for the analysis. All trials are listed in Table 1. The variables were grouped into laboratory, tumour or clinical parameters. The cut points for laboratory parameters were chosen as above or lower normal limits and arbitrarily at other values (Table 2).


View this table:
[in this window]
[in a new window]
 
Table 2. Predictors, cut off points, percentage of missing data
 
Statistical analysis
Tree analysis. A tree analysis approach was used to analyse the relations between the predictor variables and the outcome, the total survival. Recursive partition and amalgamation (RECPAM) is especially designed for this situation [30]. In this methodology the patient population is recursively partitioned into subgroups, the ‘nodes’. Each split is defined by the value of one selected predictor. At each step, among all candidates the predictor with the highest power to discriminate between good and bad survival is chosen. As a result, each terminal node is characterised by a set of predictors and their values and is simultaneously associated with a distinct survival curve. Terminal nodes derived from different parent nodes may have similar survival curves and are finally amalgamated. The resulting clusters should define classes of patients with clearly separated survival curves.

We preferred RECPAM over the classical Cox’s regression [31] approach for the following reasons:

(i) RECPAM is directly focussed on the derivation and characterisation of clearly separable risk classes rather than on fitting regression equations to the data.

(ii) Interactions between the predictors, if present in the data, are entered automatically into the model.

(iii) Missing values of active predictors are replaced by surrogate variables in a special way, newly adapted for each node. Thus all patients may contribute to the derivation of the model.

(iv) Internal validation is incorporated into the model by applying pruning techniques and the Akaike Information Criterion (AIC) as an approximate cross-validation procedure. It was applied to select the tree with the smallest AIC as the ‘best honest tree’, but also to select the simplest acceptable tree by applying the ‘elbow rule’ to the curve of AIC versus number of nodes within the pruning procedure [30]. Thus, prediction bias due to over-fitting is generally avoided.

External validation. In order to add an external validation to the analyses, we split the population randomly into a learning (n = 2541) and a validation (n = 1276) set. The RECPAM prediction model was derived using only the learning set and it was then applied to the independent validation set. Thus the predicted survival curves within the validation clusters could be compared with the observed curves.

Details of the RECPAM approach. The prediction model was derived in two steps: first a RECPAM subtree was derived for each of the above-mentioned subset of predictors (clinical, tumour-related and laboratory parameters). In the second step, only those parameters that were selected for the subtrees (the ‘winners’) were chosen as candidates for the general model. By this restriction the possibility of over-fitting was further diminished. In addition, the results of the subtree models are of interest in their own right.

The RECPAM method itself was applied with the following restrictions: the minimal node size was set to 50 patients throughout the tree, and at each node only those variables with <33% missing values were accepted for the next split.


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Univariate analyses
Table 3 displays the results of the univariate analyses for the whole cohort of patients. Due to the large number of patients, statistical significance is not equivalent to clinical relevance. Nevertheless, notable are those parameters that, despite of the large number of patients, were not found to be significant: age and whether the patients had received adjuvant chemotherapy. However, the number of patients who had received adjuvant chemotherapy was low (n = 277) and information about the time-span between the adjuvant chemotherapy and initiation of treatment for metastatic disease is unknown.


View this table:
[in this window]
[in a new window]
 
Table 3. Results of the univariate analyses
 
Multivariate analysis for subgroups of predictors
Among the clinical variables only the ECOG performance status (0, 1 versus >1) remained an independent predictor for survival in the multivariate setting, and patients with a better performance status had a better survival.

For the tumour-related parameters the number of tumour sites of more than one and also of more than two, and presence of liver metastases or peritoneal carcinomatosis predicted a worse patient outcome. In contrast, in well-defined subpopulations, presence of lung or lymph node metastases and rectal cancer as the primary tumour were all associated with a better outcome. Among the laboratory parameters, platelets >400 x 109/l, alkaline phosphatase of >300 U/l, WBC count >10 x 109/l and haemoglobin <11 g/dl all predicted an inferior survival probability. The winners of each subgroup analysis were entered into the general predictive model.

General predictive model
The general model without amalgamation is displayed in Figure 1 and Table 4. With the knowledge of eight different parameters the individual prognosis of a patient can be estimated, ranging from a median of 4 months up to >16 months. The strongest predictor for survival is the performance status. The number of involved tumour sites is separating the cohort learning sample at seven of 15 possible nodes, thus indicating the importance of tumour load for the prognosis of patients with metastatic colorectal cancer. The WBC count (cut point at 10 x 109/l), location of the primary tumour, level of alkaline phosphatase (cut point of 300 U/l), platelets (cut point of 400 x 109/l), haemoglobin (cut point of 11 x 109/l), and presence of peritoneal involvement all entered the general model and influence the prognosis at various nodes in various constellations with other parameters.



View larger version (43K):
[in this window]
[in a new window]
 
Figure 1. General prediction model without amalgamation. Numbers in circles are number of patients at that node. Split to the left denotes higher risk criterion. Numbers in boxes represent median survival in months (upper number), number of patients at end node (middle number) and node number (bottom number), referring to Table 4.

 

View this table:
[in this window]
[in a new window]
 
Table 4. Median survival time of patients in different nodes with 95% confidence interval (number of nodes is referring to Figure 1)
 
Figure 1 also suggests that certain terminal nodes have a similar outcome, with one group of poor survival ranging from 4 to ~7 months, one intermediate group ranging from ~8 to 10 months, and one lower risk group with range of survival from 12 to 16 months. Therefore, in a further step, the evaluation was focussed on the selection of the simplest acceptable tree (by applying the above mentioned elbow rule to the AIC criterion). This was followed by the amalgamation procedure that was limited to result in only three prognostic groups that could clearly separate the patient prognosis (Figure 2). With the knowledge of the four parameters—performance status, number of involved tumour sites, WBC count and alkaline phosphatase—the patients are separated into three clearly distinguishable risk groups. Low risk group (n = 1111, median survival 15 months): patients with performance status of ECOG 1 or 0 and only one tumour site. Intermediate risk group (n = 904, median survival 10.7 months), consisting of two subsets: (i) patients with a good performance status (ECOG <1), but more than one tumour site plus an alkaline phosphatase level of <300 U/l; (ii) patients with a poor performance status, plus a low WBC count and only one tumour site. High risk group (n = 534, median survival 6.1 months), who either have a good performance status but more than one tumour site and a high alkaline phosphatase level, or a poor performance status plus a high WBC count, or a poor performance status plus a low WBC count and more than two tumour sites.



View larger version (34K):
[in this window]
[in a new window]
 
Figure 2. General prediction model for three risk groups. Numbers in circles are number of patients. Split to the left denotes higher risk criterion. Numbers in boxes are median survival in months (upper number) and number of patients at end node (bottom number). Numbers in shaded boxes are number of patients in parentheses, median survival of learning sample and median survival of validation sample in months, 95% CI of median in parentheses.

 
In patients with a good performance status but two or more tumour sites, the level of alkaline phosphatase predicted a median survival of either 6.1 or 10.7 months. Also, in patients with a poor performance status but a normal WBC count, the number of metastatic sites predicts a median survival of 6.1 months as opposed to 10.7 months.

Figure 3 shows the survival curves of the three different risk groups for the learning and validation samples. In both samples the three risk groups are clearly separated with a high level of concordance. Also, the survival curves for the good, intermediate and high risk groups of the validation set lie within the 95% confidence limits of the learning set for the first 24 months at least. The P value for the three risk goups in the validation set is <0.0001.



View larger version (28K):
[in this window]
[in a new window]
 
Figure 3. Survival curves for the three risk groups of the learning set (n = 2549) and validation set (n = 1276). The dashed lines denotes the risk groups of the learning set; the dotted lines are the upper and lower limits of the 95% confidence limits of the learning set for the three risk groups; the solid lines denote the three risk groups of the validation set.

 

    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
To our knowledge this multivariate analysis presents the largest study to identify clinical prognostic factors for patients with metastatic colorectal cancer treated with 5-FU-based chemotherapy. We used 23 variables, of which some entered the model at different cut off points. The statistical methodology was able to identify those cut off points with the best predictive value and also potential interactions between prognostic factors. Based on a total of eight clinical parameters—performance status, number of involved tumour sites, WBC count, alkaline phosphatase, platelets, haemoglobin, presence of peritoneal carcinomatosis and location of primary tumour—the median survival for an individual patient may be estimated, ranging from 4 to 16.3 months. The evaluation of the eight parameters usually belongs to the clinical routine before a patient receives chemotherapy. In order to generate a more simplified model, patient risk groups with similar outcomes were combined and three clearly separated risk groups were defined. With the use of only four clinical parameters, i.e. performance status, level of WBC count, alkaline phosphatase and number of involved tumour sites, a high risk group with a median survival of 6.1 months, an intermediate risk group with a median survival of 10.7 months and a low risk group with a median survival of 15 months were identified. About 21% of the patients in our validation set belonged to the high risk, 35% into the intermediate and 44% into the low risk group. Whether this actually reflects the population regularly seen in oncology clinics or rather a typical population entering clinical trials is unknown.

Earlier attempts to identify prognostic factors identified variables consistent to our findings. Most investigators re-ported the performance status [5, 8, 32–34] as a major prognos- tic factor, but the WBC count (10 x 109/l) [5, 35, 36], haemoglobin [6], alkaline phosphatase [5] and site of the primary tumour (i.e. colon or rectum) [9] were also identified by some.

Our general model, however, has some limitations. In approximately two-thirds of cases we had no information on weight loss, level of lactate dehydrogenase (LDH), serum protein, albumin or pathological tumour grading. In addition, in ~50% of patients information on bilirubin, transaminases, CEA levels and tumour-related symptoms were missing. Serum levels of LDH, CEA, bilirubin transaminases and albumin and weight loss were all identified as prognostic factors in smaller cohorts of patients [5, 8, 32, 37–39]. The RECPAM analysis uses surrogate markers in case of missing values in certain patients. This is a useful tool in order not to exclude too many patients with missing values or alternatively not too exclude too many variables from the analysis. We determined that no variables should have >33.3% of patients with missing values within a node that separates the population by this variable. Hence, all variables with a moderate (50%) or high (75%) rate of missing value still had a chance to enter the model; however, the likelihood was lower compared with the chances of other variables, unless they had a marked impact on prognosis.

Any multivariate analysis faces the problem that only those variables that have been put into the model come out as important predictors. It is possible that other potential parameters or constellation of variables may better describe and separate the prognosis of patients with metastatic colorectal cancer. Nevertheless, our learning sample clearly identifies three separated risk groups validated in the validation set. Whether other variables including those with a high rate of missing values might add any information to our model remains speculative and at present less important, as we were able to clearly separate patient risk groups with the available data. The difference in median survival in the three groups is also of definite clinical relevance.

Our model was based on patients who received several forms of 5-FU-based treatment, i.e. bolus 5-FU, 5-FU with modulation or infusional 5-FU regimens. Whether this model also applies to patients receiving 5-FU in combination with CPT-11 or oxaliplatin as their first-line treatment is unknown. In other words, it may by speculated that a more effective treatment may alter the predictive value of prognostic factors. Interestingly, the number of involved organ sites, performance status and WBC count have all been identified as important prognostic factors in a recent randomised trial with the use of CPT-11 and 5-FU as first-line treatment [39]. Also, two studies with the use of 5-FU in combination with oxaliplatin found the number of involved organ sites, performance status and alkaline phosphatase as independent prognostic factors consistent with our findings [40]. In addition, similar results have been reported in two studies for patients receiving CPT-11 as second-line treatment after 5-FU failure [41, 42]. In a multivariate Cox’s regression analysis performance status, haemoglobin, WBC count and alkaline phosphatase were identified in both studies and number of involved tumour sites, aspartate aminotransferase and right-sided primary tumours of the colon in at least one of these two trials. In summary, the prognostic factors identified within our model appear to be a consistent finding, irrespective of the type of cytostatic treatment and whether patients receive first- or second-line therapy.

What are the potential implications of our findings?
First, our three risk groups could serve as a useful stratification tool for clinical trials. Four parameters separate patient survival by a median of ~5 months. Indeed, very often multivariate analysis performed in controlled randomised trials demonstrates a larger influence on survival for certain prognostic factors than for the treatment that was under investigation [19, 39, 43]. This is of special importance as most clinical trials demonstrate only a small difference, if any at all, in median survival, which seldom exceeds 2–3 months.

Secondly, to use the four variables for a more precise description of patients entering non-randomised trials also appears useful. Patients entering phase II trials may often represent a more favourable subgroup. Thus, a median survival of 12 months in a patient cohort that belongs to the high risk category may be very promising, but more disappointing for a low risk group.

Thirdly, our model allows for an estimate of expected individual patient survival, and every fifth patient in our population had a a median survival of only ~6 months, despite receiving 5-FU-based treatment. Prospective trials should answer the question of whether this high risk group of patients might benefit from a more intensive first-line treatment. Preliminary retrospective data, presented by Knight et al. [44], indicate the opposite, namely that patients with a favourable prognostic profile, rather than those with a poor one identified by levels of LDH, did benefit more from intensified treatment. Nevertheless, only a prospective randomised trial would help to identify the groups of patients that benefit most from intensified combination treatment with CPT-11 or oxaliplatin added to modulated 5-FU.

Fourthly, the level of haemoglobin has also been identified by the Nordic Group as a risk factor. Recently, trials using erythropoietin to increase the haemoglobin level suggest the possibility of an improved outcome for patients with haematological malignancies and solid tumours receiving erythropoietin in addition to chemotherapy [45, 46]. Our investigation implies that further study of the effects of erythropoietin in patients with metastatic colorectal cancer and low haemoglobin levels is necessary.

Finally, markers identified as predictors of response to chemotherapy such as thymidilate synthetase or thymidine phosphorylase [47, 48], or molecular prognostic parameters for survival, such as proteins involved in the cell cycle regulation and apoptosis [49], should be validated against our simple model, as all of them need sophisticated and expensive laboratory investigations and thus are less easily obtained compared with our simple clinical parameters.

In summary, patients with metastatic colorectal cancer are very heterogeneous in respect to survival after a standard cytostatic drug treatment. Patients entering randomised trials for metastatic colorectal cancer may be stratified by at least four simple parameters. Also, the EORTC GI group will initiate studies for high risk, intermediate and low risk patients separately. Our model may serve as a point of reference for any other models using different parameters.


    Footnotes
 
+ Correspondence to: Medizinische Klinik und Poliklinik I, Universitätsklinikum Carl Gustav Carus der TU-Dresden, Fetscher Strasse 74, 01307 Dresden, Germany. Tel: +49-351-458-4780; Fax: +49-351-458-5859; E-mail: koehne@mk1.med.tu-dresden.de Back


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
1. Ries LA, Wingo PA, Miller DS et al. The annual report to the nation on the status of cancer, 1973–1997, with a special section on colorectal cancer. Cancer 2000; 88: 2398–2424.

2. Jonker DJ, Maroun JA, Kocha W. Survival benefit of chemotherapy in metastatic colorectal cancer: a meta-analysis of randomized controlled trials. Br J Cancer 2000; 82: 1789–1794.[ISI][Medline]

3. Köhne CH, Kretzschmar A, Wils J. First-line chemotherapy for colorectal carcinoma—we are making progress. Onkologie 1998; 21: 280–289.[ISI]

4. Steinberg SM, Barkin JS, Kaplan RS, Stablein DM. Prognostic indicators of colon tumors. The Gastrointestinal Tumor Study Group experience. Cancer 1986; 57: 1866–1870.[ISI][Medline]

5. Kemeny N, Braun DW Jr. Prognostic factors in advanced colorectal carcinoma. Importance of lactic dehydrogenase level, performance status, and white blood cell count. Am J Med 1983; 74: 786–794.[ISI][Medline]

6. Graf W, Bergstrom R, Pahlman L, Glimelius B. Appraisal of a model for prediction of prognosis in advanced colorectal cancer. Eur J Cancer 1994; 30A: 453–457.

7. Graf W, Glimelius B, Pahlman L, Bergstrom R. Determinants of prognosis in advanced colorectal cancer. Eur J Cancer 1991; 27: 1119–1123.[ISI][Medline]

8. Webb A, Scott-Mackie P, Cunningham D et al. The prognostic value of CEA, beta HCG, AFP, CA125, CA19-9 and C-erb B-2, beta HCG immunohistochemistry in advanced colorectal cancer. Ann Oncol 1995; 6: 581–587.[Abstract]

9. Thirion P, Cassano A, O’Dwyer PJ et al. Meta-analysis of alpha-interferon plus 5-fluorouracil in advanced colorectal cancer. Proc Am Soc Clin Oncol 2000; 19: 244a (Abstr 945).

10. Köhne CH, Schöffski P, Wilke H et al. Effective biomodulation by leucovorin of high-dose infusion fluorouracil given as a weekly 24-hour infusion: results of a randomized trial in patients with advanced colorectal cancer. J Clin Oncol 1998; 16: 418–426.[Abstract]

11. Köhne CH, Wilke H, Hiddemann W et al. Phase II evaluation of 5-fluorouracil plus folinic acid and alpha 2b-interferon in metastatic colorectal cancer. Oncology 1997; 54: 96–101.[ISI][Medline]

12. Köhne CH, Hiddemann W, Schuller J et al. Failure of orally administered dipyridamole to enhance the antineoplastic activity of fluorouracil in combination with leucovorin in patients with advanced colorectal cancer: a prospective randomized trial. J Clin Oncol 1995; 13: 1201–1208.[Abstract]

13. Scheithauer W, Kornek G, Marczell A et al. Fluorouracil plus racemic leucovorin versus fluorouracil combined with the pure 1-isomer of leucovorin for the treatment of advanced colorectal cancer: a randomized phase III study. J Clin Oncol 1997; 15: 908–914.[Abstract]

14. Aranda E, Diaz-Rubio E, Cervantes A et al. Randomized trial comparing monthly low-dose leucovorin and fluorouracil bolus with weekly high-dose 48-hour continuous-infusion fluorouracil for advanced colorectal cancer: a Spanish Cooperative Group for Gastrointestinal Tumor Therapy (TTD) study. Ann Oncol 1998; 9: 727–731.[Abstract]

15. Rougier P, Paillot B, LaPlanche A et al. 5-Fluorouracil (5-FU) continuous intravenous infusion compared with bolus administration. Final results of a randomised trial in metastatic colorectal cancer. Eur J Cancer 1997; 33: 1789–1793.[ISI][Medline]

16. Hill M, Norman A, Cunningham D et al. Royal Marsden phase III trial of fluorouracil with or without interferon alfa-2b in advanced colorectal cancer. J Clin Oncol 1995; 13: 1297–1302.[Abstract]

17. Hill M, Norman A, Cunningham D et al. Impact of protracted venous infusion fluorouracil with or without interferon alfa-2b on tumor response, survival, and quality of life in advanced colorectal cancer. J Clin Oncol 1995; 13: 2317–2323.[Abstract]

18. Ross P, Norman A, Cunningham D et al. A prospective randomised trial of protracted venous infusion 5-fluorouracil with or without mitomycin C in advanced colorectal cancer. Ann Oncol 1997; 8: 995–1001.[Abstract]

19. Blijham G, Wagener T, Wils J et al. Modulation of high-dose infusional fluorouracil by low-dose methotrexate in patients with advanced or metastatic colorectal cancer: final results of a randomized European Organization for Research and Treatment of Cancer Study. J Clin Oncol 1996; 14: 2266–2273.[Abstract]

20. Blijham G, Wagener T, Wils J et al. Double modulation of high dose 5-fluorouracil (FU) with low dose PALA and methotrexate (MTX): results of a randomized EORTC study. Proc Am Soc Clin Oncol 1997; 16: 267a (Abstr 948).

21. Di Costanzo F, Bartolucci R, Calabresi F et al. Fluorouracil-alone versus high-dose folinic acid and fluorouracil in advanced colorectal cancer: a randomized trial of the Italian Oncology Group for Clinical Research (GOIRC). Ann Oncol 1992; 3: 371–376.[Abstract]

22. Di Costanzo F, Gasperoni S, Malacarne P et al. High-dose folinic acid and 5-fluorouracil alone or combined with hydroxyurea in advanced colorectal cancer: a randomized trial of the Italian Oncology Group for Clinical Research. Am J Clin Oncol 1998; 21: 369–375.[ISI][Medline]

23. Di Costanzo F, El-Taani H, Marzola M et al. Hydroxyurea (HU), high dose folinic acid (l-FA) and 5-FU vs. HU, 5FU and interferon-alpha-2b (IFN) in advanced colorectal cancer (ACRC): A randomized trial of the Italian Oncology Group for Clinical Research (GOIRC). Proc Am Soc Clin Oncol 1995; 14: 208a (Abstr 961).

24. Glimelius B. Biochemical modulation of 5-fluorouracil: a randomized comparison of sequential methotrexate, 5-fluorouracil and leucovorin versus sequential 5-fluorouracil and leucovorin in patients with advanced symptomatic colorectal cancer. The Nordic Gastrointestinal Tumor Adjuvant Therapy Group. Ann Oncol 1993; 4: 235–240.[Abstract]

25. Anonymous. Superiority of sequential methotrexate, fluorouracil, and leucovorin to fluorouracil alone in advanced symptomatic colorectal carcinoma: a randomized trial. The Nordic Gastrointestinal Tumor Adjuvant Therapy Group. J Clin Oncol 1989; 7: 1437–1446.[Abstract]

26. Anonymous. Expectancy or primary chemotherapy in patients with advanced asymptomatic colorectal cancer: a randomized trial. Nordic Gastrointestinal Tumor Adjuvant Therapy Group. J Clin Oncol 1992; 10: 904–911.[Abstract]

27. Cunningham D, Zalcberg JR, Rath U et al. ‘Tomudex’ (ZD1694): results of a randomised trial in advanced colorectal cancer demonstrate efficacy and reduced mucositis and leucopenia. The ‘Tomudex’ Colorectal Cancer Study Group. Eur J Cancer 1995; 31A: 1945–1954.

28. Cocconi G, Cunningham D, Van Cutsem E et al. Open, randomized, multicenter trial of raltitrexed versus fluorouracil plus high-dose leucovorin in patients with advanced colorectal cancer. Tomudex Colorectal Cancer Study Group. J Clin Oncol 1998; 16: 2943–2952.[Abstract]

29. Pazdur R, Vincent M. Raltitrexed (Tomudex) versus 5-fluorouracil and leucovorin (5-FU+LV) in patients with advanced colorectal cancer (ACC): results of a randomized, multicenter, North American trial. Proc Am Soc Clin Oncol 1997; 16: 228a (Abstr 801).

30. Ciampi A, Negassa A, Lou Z. Tree-structured prediction for censored survival data and the Cox model. J Clin Epidemiol 1995; 48: 675–689.[ISI][Medline]

31. Cox DR. Regression models and life-tables. J R Stat Soc 1972; B34: 187–202.[ISI]

32. Steinberg J, Erlichman C, Gadalla T et al. Prognostic factors in patients with metastatic colorectal cancer receiving 5-fluorouracil and folinic acid. Eur J Cancer 1992; 28A: 1817–1820.

33. Graf W, Pahlman L, Bergstrom R, Glimelius B. The relationship between an objective response to chemotherapy and survival in advanced colorectal cancer. Br J Cancer 1994; 70: 559–563.[ISI][Medline]

34. Lavin P, Mittelman A, Douglass H et al. Survival and response to chemotherapy for advanced colorectal adenocarcinoma. Cancer 1980; 46: 1536–1543.[ISI][Medline]

35. Edler L, Heim ME, Quintero C et al. Prognostic factors of advanced colorectal cancer patients. Eur J Cancer Clin Oncol 1986; 22: 1231–1237.[ISI][Medline]

36. Köhne CH, Hecker H, Schöffski P et al. Weekly high dose infusional (CI) 5-FU plus folinic acid (FA) has major impact on survival in patients with advanced colorectal cancer. Results of a multivariate analysis using RECPAM. Proc Am Soc Clin Oncol 1996; 15: 201a (Abstr 972).

37. Fountzilas G, Gossios K, Zisiadis A et al. Prognostic variable in patients with advanced colorectal cancer treated with fluorouracil and leucovorin-based chemotherapy. Med Pediatr Oncol 1996; 26: 305–317.[ISI][Medline]

38. Lavin P, Mittelman A, Douglass H Jr et al. Survival and response to chemotherapy for advanced colorectal adenocarcinoma: an Eastern Cooperative Oncology Group report. Cancer 1980; 46: 1536–1543.[ISI][Medline]

39. Saltz LB, Cox JV, Blanke C et al. Irinotecan plus fluorouracil and leucovorin for metastatic colorectal cancer. Irinotecan Study Group. N Engl J Med 2000; 343: 905–914.[Abstract/Free Full Text]

40. Giacchetti S, Perpoint B, Zidani R et al. Phase III multicenter randomized trial of oxaliplatin added to chronomodulated fluorouracil–leucovorin as first-line treatment of metastatic colorectal cancer. J Clin Oncol 2000; 18: 136–147.[Abstract/Free Full Text]

41. Cunningham D, Pyrhonen S, James RD et al. Randomised trial of irinotecan plus supportive care versus supportive care alone after fluorouracil failure for patients with metastatic colorectal cancer. Lancet 1998; 352: 1413–1418.[ISI][Medline]

42. Rougier P, Van CE, Bajetta E et al. Randomised trial of irinotecan versus fluorouracil by continuous infusion after fluorouracil failure in patients with metastatic colorectal cancer [published erratum appears in Lancet 1998; 352: 1634]. Lancet 1998; 352: 1407–1412.[ISI][Medline]

43. de Gramont A, Figer A, Seymour M et al. Leucovorin and fluorouracil with or without oxaliplatin as first line treatment in advanced colorectal cancer. J Clin Oncol 2000; 18: 2938–2947.[Abstract/Free Full Text]

44. Knight RD, Miller LL, Pirotta N et al. First line irinotecan (C), Fluorouracil (F), Leukovorin (L) especially improves survival (OS) in metastatic colorectal cancer (MCRC) patients (PT) with favorable prognostic indicators. Proc Am Soc Clin Oncol 2000; 19: 246a (Abstr 991).

45. Littlewood TJ, Rapoport BL, Bajetta E et al. Possible relationship of hemoglobin levels with survival in anemic cancer patients receiving chemotherapy. Proc Am Soc Clin Oncol 2000; 19: 605a (Abstr 2381).

46. Glimelius B, Linne T, Hoffman K et al. Epoetin beta in the treatment of anemia in patients with advanced gastrointestinal cancer. J Clin Oncol 1998; 16: 434–440.[Abstract]

47. Salonga D, Danenberg KD, Johnson M et al. Colorectal tumors responding to 5-fluorouracil have low gene expression levels of dihydropyrimidine dehydrogenase, thymidylate synthase, and thymidine phosphorylase. Clin Cancer Res 2000; 6: 1322–1327.[Abstract/Free Full Text]

48. Peters GJ, van der Wilt CL, van Groeningen CJ et al. Thymidylate synthase inhibition after administration of fluorouracil with or without leucovorin in colon cancer patients: implications for treatment with fluorouracil. J Clin Oncol 1994; 12: 2035–2042.[Abstract]

49. Sturm I, Köhne CH, Wolff G et al. Analysis of the p53/BAX pathway in colorectal cacner: Low BAX is a negative prognostic factor in patients with resected liver metastases. J Clin Oncol 1999; 17: 1–11.[Free Full Text]