Alpha-interferon does not increase the efficacy of 5-fluorouracil in advanced colorectal cancer

Two meta-analyses were conducted to quantify the benefit of combining α-IFN to 5FU in advanced colorectal cancer in terms of tumour response and survival. Analyses were based on a total of 3254 individual patient data provided by principal investigators of each trial. The meta-analysis of 5FU ± LV vs. 5FU ± LV + α-IFN combined 12 trials and 1766 patients. The meta-analysis failed to show any statistically significant difference between the two treatment groups in terms of tumour response or survival. Overall tumour response rates were 25% for patients receiving no α-IFN vs. 24% for patients receiving α-IFN (relative risk, RR = 1.02), and median survivals were 11.4 months for patients receiving no α-IFN vs. 11.5 months for patients receiving α-IFN (hazard ratio, HR = 0.95). The meta-analysis of 5FU + LV vs. 5FU + α-IFN combined 7 trials, and 1488 patients. This meta-analysis showed an advantage for 5FU + LV over 5FU + α-IFN which was statistically significant in terms of tumour response (23% vs. 18%; RR = 1.26;P = 0.042), and of a borderline significance for overall survival (HR = 1.11;P = 0.066). Metastases confined to the liver and primary rectal tumours were independent favourable prognostic factors for tumour response, whereas good performance status, metastases confined to the liver or confined to the lung, and primary tumour in the rectum were independent favourable prognostic factors for survival. We conclude that α-IFN does not increase the efficacy of 5FU or of 5FU + LV, and that 5FU + α-IFN is significantly inferior to 5FU + LV, for patients with advanced colorectal cancer. © 2001 Cancer Research Campaign http://www.bjcancer.com


Trial selection
Two meta-analyses were conducted concomitantly. In the first one we considered all properly randomised trials comparing 5FU with or without folinic acid (5FU ± LV) to the same 5FU ± LV regimen plus α-IFN (5FU ± LV + α-IFN). In the second meta-analysis we considered all properly randomised trials comparing 5FU + LV to 5FU + α-IFN. In both meta-analyses, α-IFN must have consisted of α-2a-interferon or α-2b-interferon, and patients must have been included in the trial before July 1996. The search for relevant trials was initiated in October 1996 by consulting MEDLINE, Physician Data Query (PDQ), the proceedings of major conferences since 1989, and through contacts with principal investigators. A total of 20 relevant trials were identified, but 3 of them (335 patients) could not be included in the meta-analysis, due to lack of data or information on the trial (Kreuser et al, 1995); (Kosmidis et al, 1996); (Recchia et al, 1996).
Meta-analysis of 5FU ± LV vs. 5FU ± LV + α-IFN (Table 1) The comparison of 5FU versus 5FU + α-IFN was addressed in 7 trials, the Roche International Clinical Research Center (RICRC) trial (Greco et al, 1996), the Palermo trial (Palmeri et al, 1998), the Ancona trial (Piga et al, 1996), two Royal Marsden Hospital (RMH) trials (Hill et al, 1995a+b), the trial from France (Dufour et al, 1996), and the Eastern Cooperative Oncology Group, Cancer and Leukemia Group B (ECOG/CALGB) trial (O'Dwyer et al, 1996). The ECOG/CALGB trial (O'Dwyer et al, 1996) was not considered in the first meta-analysis, because unlike the other trials, the planned dose of 5FU and its mode of administration were not the same in the 2 treatment groups. In most trials, the 5FU regimen was close to the Wadler regimen (Wadler et al, 1989), consisting of an initial 5-day 5FU infusion followed by a weekly 5FU infusion. The dose of 5FU varied from 500 to 750 mg/m 2 /day. The dose of α-IFN varied from 3 to 10 MU, 3 times a week. Based on the impact of the mode of 5FU administration on tumour response and survival (Meta-Analysis Group In Cancer,  Seymour et al, 1996 Same + α-IFN 6 MU every other day d1 to d12 1998), trials were further stratified according to the duration of 5FU infusion. Bolus 5FU were administered in 5 comparisons (Hill et al, 1995;Dufour et al, 1996;Greco et al, 1996;Piga et al, 1996;Palmeri et al, 1998) and continuous infusion 5FU in one comparison (Hill et al, 1995b). The comparison of 5FU + LV versus 5FU + LV + α-IFN was addressed in 6 trials, the Gruppo Oncologico dell'Italia Meridionale (GOIM) trial (Colucci et al, 1999), the Roma trial (Cassano et al, 1996), the trial from Hungary (Pajkos et al, 1997), the trial from Argentina (Pensel et al, 1993), the Medical Research Council (MRC) trial (Seymour et al, 1996), and the AIO trial (Köhne et al, 1998). The AIO trial (Köhne et al, 1998) and the trial from Hungary (Pajkos et al, 1997) were multiple-arm trials. Two trials (MRC (Seymour et al, 1996), AIO (Köhne et al, 1998)) used a continuous infusion 5FU. Trials were stratified according to 5FU schedule of administration (5FU bolus and 5FU continuous infusion), and in terms of modulation of 5FU by leucovorin.

Protocol for the meta-analysis
In March 1997, all principal investigators received a protocol for the meta-analyses, and were asked to provide individual patient data. Information requested for every randomised patient was date of randomisation, tumour measurability (i.e. measurable or nonmeasurable tumours), treatment assigned by randomisation, age, gender, performance status according to the ECOG scale, primary tumour site (colon or rectum), prior adjuvant chemotherapy, prior chemotherapy for metastatic disease, site of metastases, overall response status with the first assigned treatment, date of response or progression with the first allocated treatment, cross-over to another treatment arm, date of death or last visit, survival status, and cause of death if applicable. Data on toxicity were not collected.

Data collection
All individual patient data were received by April 1999. Data were extensively checked and discussed with all collaborators present at a plenary meeting of the Meta-Analysis Group In Cancer held in Atlanta, GA, in May 1999.

Tumour response and survival
Complete response (CR) and partial response (PR) criteria adopted in individual trials followed the World Health Organization recommendations (Miller et al, 1981) and were similar in all trials. Patients experiencing minimal response, stable disease or progressive disease were considered to have no response for the purpose of the meta-analyses. In the MRC trial (Seymour et al, 1996) and in the trial from Hungary (Pajkos et al, 1997) chemotherapy was stopped after 6 months in the absence of tumour progression. In all 5-FU and interferon in advanced colorectal cancer 613 British Journal of Cancer (2001) 84(5), 611-620 © 2001 Cancer Research Campaign Table 2 Randomised clinical trials comparing 5FU + LV to 5FU + α-IFN in advanced colorectal cancer

Comparison
Patients Treatment arms 5FU + LV vs. 5FU + α-IFN, with the same dose of 5FU in both arms GOIRC 238 5FU 600 mg/m 2 bolus, + l-folinic acid 250 mg/m 2 bolus, + HU 3 g once a week for 6 weeks followed by a 2 week-break Di Costanzo et al, Same without l-folinic acid + l-folinic acid + α-IFN 3 MU three times a week 1995 AIO 187 5FU 2 600 mg/m 2 continuous infusion + LV 500 mg/m 2 bolus, once a week for 6 weeks followed by a 2 week-break Köhne et al, 1998 Same without LV + α-IFN 3 MU three times a week Turkey 46 5FU 500 mg/m 2 /d bolus d1 to d5 + l-folinic acid 100 mg/m 2 , then weekly, every 4 weeks Aykan et al, 1996 Same without l-folinic acid + IFN 5 MU three times a week other trials treatment was maintained until disease progression or severe toxicity. Duration of survival was calculated from the date of randomisation to the date of death, whatever its cause.

Statistical methods
The statistical methods for meta-analyses based on individual patient data have been described in detail in previous publications (ACCMP, 1992;ACCMP, 1994;MAGIC, 1996;MAGIC, 1998a;MAGIC, 1998b). All analyses were based on an intention to treat basis, without any patient exclusion. Tumour responses were compared through relative risks (RR) in individual trials and overall (MAGIC, 1998b). Prognostic factors for response were identified through a logistic regression model (Cox, 1970). Survival times were compared through hazard ratios (HR) in individual trials and overall (Peto et al, 1977). Prognostic factors for survival were identified through a proportional hazards regression model (Cox, 1972). All P values were two-sided.

Patient characteristics
A total of 3254 were included in the analyses. The main patient characteristics are listed in Table 3 and 4. As could be expected in large series of patients, there was no imbalance between the experimental and the control groups for either of the comparisons of interest. 84% of patients had died at the time of analysis.

Meta-analysis of 5FU + LV vs. 5FU + α-IFN
1488 patients were included in this meta-analysis. The ECOG/CALGB trial (O'Dwyer et al 1996) allowed the inclusion of patients with non-measurable disease. After exclusion of these patients, 1305 patients were eligible for tumour response assessment.
Analyses stratified by type of 5FU administration showed that the advantage of 5FU + LV over 5FU + α-IFN was limited to the group of trials using the same 5FU schedules in both treatment arms (RR = 1.80; 95% CI = 1.29-2.51; P = 0.0005).
Survival analysis showed a small trend in favour of 5FU + LV over 5FU + α-IFN, but this advantage was not statistically significant (overall HR = 1.11; 95% CI = 0.99-1.24; P = 0.066) (Figure 4). Median survivals were 11.7 months for patients allocated to 5FU + LV and 11.3 months for patients allocated to 5FU + α-IFN. The survival difference reached statistical significance in the group of trials using the same 5FU schedules in both treatment arms (HR = 1.29; 95% CI 1.07-1.57; P = 0.008). There was some heterogeneity in this group of trials, but which did not reach a statistically significant level (P = 0.67).

Prognostic factor analyses
Individual patient data used for the two meta-analyses were combined to identify prognostic factors for response and survival (3254 patients). Sex, age, performance status (PS), primary tumour site, previous adjuvant chemotherapy, metastatic site, and allocated treatment (no α-IFN vs. α-IFN) were considered in these analyses. In a logistic regression model, metastases confined to the liver (P < 10 −4 ), and primary rectal tumours (P = 0.042) were the independent favourable prognostic factors for tumour response. Tumour response rates were 26% for patients with metastases confined to the liver versus 20% for the others. Patients with rectal cancer had a 26% tumour response rate, vs. 22% with colon tumour.

5-FU and interferon in advanced colorectal cancer 615
British Journal of Cancer (2001) 84(5) Figure 1 Tumour response relative risks in individual trials and overall for the meta-analysis 5FU ± LV vs. 5FU ± LV + α-IFN In a Cox regression model, good PS (P < 10 −4 ), metastases confined to the liver or confined to the lung (P = 0.0002 and P = 0.004 respectively), and primary tumour in the rectum (P = 0.003) were the independent favourable prognostic factors for survival. One-year survival was 59% for patients with PS 0, 47% for patients with PS 1, and 30% for patients with PS 2 or worse; 54% for patients with metastases confined to the liver, 46% for the others; 61% for patients with metastases confined to the lung, 49% for the others; 56% for patients with primary rectal cancer, 47% for the others.

DISCUSSION
Pre-clinical studies indicate that α-IFN may increase the cytotoxicity of 5FU in a variety of tumour cell lines (Elias and Crissman, 1988;Wadler et al, 1990). Several mechanisms of interaction between 5FU and interferon have been demonstrated. In vitro data published by Elias and Crissman (Elias and Crissman, 1988) suggest that the enzyme thymidylate synthase might be a target for this interaction. Moreover, the presence of thymidine in the culture medium tends to block the synergic effect (Neefe and John, 1991). Interferon may also modify the plasma pharmacokinetics of 5FU (Lindley et al, 1990;Danhouser et al, 1991). Finally, 5FU may influence the immunomodulatory actions of interferon (Neefe and John, 1991). However, despite more than 3000 patients included in randomized trials, the clinical impact of combining α-IFN to 5FU remained debatable.
The 2 meta-analyses presented here address the efficacy of α-IFN combined with 5FU in advanced colorectal cancer. Tumour response rate and survival were the two main end points. Toxicity was not studied, since at the time of beginning these meta-analyses individual trials had already demonstrated that the addition of α-IFN to a 5FU regimen led to an increased risk of neutropenia, mucositis, and neurotoxicity, and was associated with flu-like syndromes. α-IFN also produced a significant impairment of quality of life in the MRC trial (Seymour et al, 1996).
The meta-analysis of trials comparing 5FU ± LV to a similar 5FU regimen plus α-IFN failed to show any difference between control and experimental arms in terms of tumour response or survival. The tumour response rate with 5FU bolus alone reported in the group of trials comparing 5FU to 5FU + α-IFN was rather high (19%), compared to tumour responses reported for patients receiving 5FU bolus in the 4 meta-analyses previously performed by our group, which varied between 11% and 14% (ACCMP, 1992(ACCMP, , 1994MAGIC, 1996MAGIC, , 1998a. This may reflect a selection of patients with favourable prognostic characteristics in trials included in the present meta-analysis, but does not invalidate our finding of no difference between 5FU alone and 5FU + α-IFN. It should also be noted that the doses of 5FU delivered in the 5FU alone arms were generally high compared with the 5FU doses reported in our previous meta-analyses. In contrast, the meta-analysis of trials comparing 5FU + LV to 5FU + α-IFN showed higher response rates and a trend towards longer survival in favour of 5FU + LV. In this set of trials, the overall tumour response rate and the median survival of patients receiving 5FU + LV (23% and 13 months, respectively) were remarkably similar to those reported previously in the metaanalysis of trials comparing 5FU to 5FU + LV (ACCMP, 1992), (23% and 11.5 months, respectively). Thus, the advantage of 5FU + LV over 5FU + α-IFN observed in the present meta-analysis does not seem to be due to some selection bias that might have favoured patients allocated to the 5FU + LV arm.
In this meta-analysis, the stratification of trials by type of 5FU administration (Figures 3 and 4) Figure 4 Survival hazard ratios in individual trials and overall for the meta-analysis 5FU + LV vs. 5FU + α-IFN advantage of 5FU/LV over 5FU+IFN in the group of trials using the same 5FU schedules in both arms. By contrast, there was no difference between the two treatment arms when 5FU was administered by bolus in the 5FU/LV arm and by continuous infusion in the 5FU + IFN. This could be linked to the tumour response and survival advantage of 5FU continuous infusion over 5FU bolus demonstrated in one of our previous meta-analyses (MAGIC, 1998a). 5FU dose intensity is not a valid parameter when comparing bolus versus infusion or mixed regimens. Consequently, no attempt was made to stratify trials according to 5FU dose intensity.
The prognostic factor analysis confirms well-established results, such as the key role of performance status for survival. Other findings are less classical, such as the role of primary and metastatic tumour sites, and are currently under investigation by our group, on the basis of 7000 individual patient data with advanced colorectal cancer. In the adjuvant setting, a trial conducted by the National Surgical Adjuvant Breast and Bowel Project (NSABP-C05) also failed to show any advantage for 5FU + LV + α-IFN over 5FU + LV in patients with stage II-III colon cancer (Wolmark et al, 1998). On-going studies are currently addressing the interest of other types of interferon, such as α-2c IFN and β-IFN (Villar Grimalt et al, 1999). However, new agents, such as CPT-11 (irinotecan) (Douillard et al, 2000;Saltz et al, 2000) or oxaliplatin (de Gramont et al, 2000) have demonstrated clinical benefits in advanced colorectal cancer, and are therefore more plausible candidates for the adjuvant setting.
We conclude that α-IFN does not increase the efficacy of 5FU in advanced colorectal cancer, and should not be offered in routine clinical practice.