Gemcitabine-based combinations for inoperable pancreatic cancer: Have we made real progress?

A meta-analysis of 20 phase 3 trials




Several attempts have been made at improving the efficacy of gemcitabine in advanced pancreatic cancer by combining it with other chemotherapeutic or molecularly targeted agents. However, randomized trials have produced conflicting results.


All prospective, randomized, phase 3 trials that compared single-agent gemcitabine with gemcitabine-based combinations were considered eligible for the current analysis. A literature-based meta-analysis was performed, event-based relative risk ratios with 95% confidence intervals were derived through both a fixed-effect model approach and a random-effect model approach, and overall survival (OS) was explored as the primary endpoint. To estimate the magnitude of the eventual benefit, absolute differences and the number of patients needed to treat (NNT) for 1 patient to benefit were calculated. A sensitivity analysis for OS was performed according to the type of agent used in combination with gemcitabine.


Twenty trials that involved 6296 patients were identified. No significant differences in the primary endpoint were observed in the overall population or in the sensitivity analysis. Conversely, a significant advantage was evident with regard to both progression-free survival (PFS) and the overall response rate (ORR) in the overall population, with an absolute benefit of 2.6% (NTT = 39 patients) and 3.0% (NNT = 33 patients). Platinum combinations led to the greatest absolute benefits for PFS and ORR compared with single-agent gemcitabine (10% and 6.5%, respectively), but this did not result in an OS benefit. Improvement in PFS, but not in the ORR, was correlated with an improvement in OS.


Single-agent gemcitabine remains the standard of care for patients with advanced pancreatic cancer. However, platinum/gemcitabine combinations appeared to improve PFS and the ORR and, thus, may be considered in selected patients. Cancer 2007. © 2007 American Cancer Society.

Pancreatic adenocarcinoma (PDAC) is a deadly disease, and its mortality closely approaches its incidence. Indeed, pancreatic cancer accounted for an estimated 33,730 newly diagnosed cancers and >32,000 estimated deaths in the United States alone in 2006, representing the fourth leading cause of cancer-related death in both sexes.1 Because of the combination of late-stage disease at presentation (localized, 8%; regional, 26%; metastatic, 52%) and intrinsic resistance to conventional treatments, it may be argued that PDAC has the worst overall prognosis among solid tumors, with 5-year survival rates that do not exceed 5%. Even in the small percentage of patients who present with localized disease and, thus, are amenable to surgery with curative intent, long-term survival barely exceeds 15%.1 In patients with advanced, inoperable disease, the objectives of systemic treatment are palliative: Weekly treatment with gemcitabine (G) (Gemzar; Eli Lilly, Indianapolis, Ind) at a dose of 1000 mg/m2 infused over 30 minutes provides a small but statistically significant overall advantage over bolus 5-fluorouracil (5-FU) in terms of both clinical benefit response (23.8% vs 4.8%; P = .0022) and survival (median survival, 5.65 months vs 4.41 months; P = .0025) and currently is considered the treatment of choice for patients with inoperable disease.2 Building upon the favorable toxicity profile of G, in the past 10 years, many attempts have been made to improve outcomes by combining it with a number of other chemotherapeutic agents. Results from phase 3 trials, however, have failed to demonstrate convincingly the superiority of doublet regimens over single-agent G, although agents with different mechanism of action have been combined (platinum salts, fluoropyrimidines, and camptothecines). The overall negative results with G-based combinations generally have been attributed to a lack of statistical power to detect small differences in survival. Although this hypothesis may be supported by the observation that hazard ratios (HRs) for survival were similar across trials, although significant differences were not always demonstrated,3, 4 such conflicting results called for a cumulative analysis that could detect real survival differences, albeit small, with adequate statistical power.

More recently, possible improvements in treatment efficacy also have been explored using novel, molecularly targeted agents, which were introduced in the clinical setting in the past 10 years. However, randomized clinical trials (RCTs) evaluating the addition of matrix metalloprotease inhibitors (MMP-I), farnesyl-transferase inhibitors (FTI), epidermal growth factor receptor (EGFR) inhibitors, and antigastrin vaccines to single-agent G largely have been disappointing.5–9 To assess the potential benefit in efficacy and activity of G-based combinations, including both classic cytotoxics and molecularly targeted agents, over standard single-agent G for the first-line treatment of advanced, inoperable PDAC, a literature-based meta-analysis of all randomized trials that explored this topic was performed.


Endpoint Definition

We considered the G-based combination systemic therapy arm as the experimental arm and the single-agent G arm as the standard comparator. Analyses were conducted to detect possible significant differences in primary and secondary endpoints. The primary endpoint for the analysis of the magnitude of eventual benefit was overall survival (OS), which was defined as the time (in months) between randomization and death from any cause. Secondary endpoints were 1) progression-free survival (PFS), which was defined as the time (in months) between randomization and either disease progression or death from any cause; and 2) the overall response rate (ORR) according to the World Health Organization definition.10

Search Strategy

The deadline for trial publication to be eligible for the current analysis was November 30, 2006. Updated results of eligible RCTs were gathered through Medline (PubMed:, American Society of Clinical Oncology (ASCO) (, European Society for Medical Oncology, (ESMO) (, and Federation of European Cancer Societies ( website searches. The following keywords used in the search: pancreas, advanced, chemotherapy, gemcitabine, metastatic, palliative, review, metanalysis, meta-analysis, pooled analysis, randomized, phase 3, comprehensive review, systematic review. In addition to the online search, references from reviews and original articles also were scanned to capture missing clinical trial data that met our eligibility criteria. Furthermore, presentations at major international meetings that had advanced pancreatic cancer as the main topic were checked. No language restrictions were applied. We gathered all phase 3 prospective and randomized trials that were published as formal articles in peer-reviewed journals or that were presented as abstracts at ASCO, European Cancer Conference, or ESMO meetings up to the end of November 2006 in which previously untreated patients were randomized to receive single-agent G (control arm) or G in combination with a different chemotherapeutic agent or a new targeted agent (experimental arm).

Data Extraction and Synthesis

From each arm within each trial, the number of deaths, progressions, and responses were obtained. All data were reviewed and computed separately by 2 independent investigators. The log of the relative risk ratio (RR) was estimated for each considered endpoint, and the 95% confidence interval (95% CI) was derived.11, 12 By using this method, it was possible to apply both a fixed-effect model and a random-effect model according to the inverse variance and the Mantel-Haenzel method. To test for heterogeneity between trials, the Q statistic was used. Patients were accounted and filled into 2 × 2 tables using the intention-to-treat assignment where applicable. The random-effect estimation was preferred in the presence of significant heterogeneity. To test for effect robustness and to identify the contribution of different drug classes to the effects of the combinations, a sensitivity analysis was performed in 4 different subpopulations according to the combination drug: G plus a platinum salt (P-G), G plus a fluoropyrimidine (F-G), G plus other classic cytotoxic agents (other-G), and G plus molecularly targeted agents (ND-G). Absolute benefits for each outcome were calculated as follows: (absolute benefit = exp [RR × log(control survival)] − control survival).13 The number of patients needed to treat for a single patient to benefit was determined (NNT: 1/[(absolute benefit)/100]).14

The pooled analysis calculations were determined by using Comprehensive Meta-Analysis Software, version 1.0.23 (CMA; Biostat, Englewood, NJ),12 Potential correlations between OS and surrogate endpoints (ORR and PFS) were explored according to a linear regression model considering both the actual outcome for each single arm and the calculated relative risk (RR) for each outcome in paired comparisons. Correlations were estimated according to the Pearson (r) and R2 coefficients (parametric) and the Spearman (ρ) coefficient (nonparametric). Because differences in median OS across the trials that were included in the current analysis ranged from 0.4 months to 1.5 months, a model also was derived to calculate the target sample size to correctly identify 0.4-, 0.7-, and 1.5-month benefits in OS using PFS as a surrogate endpoint.


Selected Trials

Twenty RCTs were identified that accrued 6296 patients who were randomized to receive single-agent G versus any G-based combination (Table 1). All selected RCTs reported OS data and, thus, were eligible for the primary endpoint analysis. P-G combinations were evaluated in 7 RCTs that involved 1354 patients (cisplatin in 4 RCTs that involved 382 patients; cisplatin, epirubicin, and 5-FU in 1 RCT that involved 104 patients; and oxaliplatin in 2 RCTs that involved 868 patients).3, 15–19 For Eastern Cooperative Oncology Group trial E6201, which randomized patients between standard-infusion G, fixed-dose-rate G, and oxaliplatin plus G, only the standard-infusion G and oxaliplatin plus G arms were considered for the current analysis.20 F-G combinations were evaluated in 5 RCTs that involved 1659 patients (5-FU with or without folinic acid in 2 RCTs that involved 788 patients, capecitabine in 2 RCTs that involved 852 patients, and uracyl-tegafur in 1 RCT that involved 19 patients).4, 20–23 Other-G combinations were evaluated in 4 RCTs involved 1404 patients; the combination drugs were topoisomerase I inhibitors in 3 RCTs (irinotecan in 2 RCTs that involved 490 patients and exatecan in 1 RCT that involved 349 patients) and pemetrexed (in 1 RCT that involved 565 patients).24–27 Four additional trials randomized 1879 patients to receive G versus ND-G, including the matrix metalloprotease inhibitor (MMP-I) marimastat (239 patients),7 the FTI tipifarnib (688 patients),5 the EGFR inhibitor erlotinib (569 patients),28 and the antigastrin vaccine G17DT (383 patients).9 Three of the 20 RCTs that were selected reported a statistically significant OS benefit for the experimental arm.4, 15, 28 The range of patient accrual to the individual trials included in our analysis was from 19 patients to 688 patients. At the time of our survey, 11 RCTs were published in extenso, and 9 were available in an abstract/presentation form.

Table 1. Randomized Clinical Trials Selected for Inclusion in the Meta-analysis
StudyYearComparatorSourceNo. of patientsSignificant survival benefit
  1. CDDP indicates cisplatin; G, gemcitabine; Epi, epirubicin; 5-FU, 5-fluorouracil; FA, folinic acid; UFT: uracyl-tegafur; G17DE, antigastrin vaccine.

Colucci et al.152002CDDP+GArticle107Yes
Wang et al.192002CDDP+GAbstract34No
Li and Chao182004CDDP+GAbstract46No
Reni et al.172005CDDP-Epi-5-FU+GArticle104No
Louvet et al.32005Oxaliplatin+GArticle313No
Heinemann et al.162006CDDP+GArticle198No
Poplin et al.382006Oxaliplatin+GAbstract555No
Berlin et al.2020025-FU+GArticle322No
Riess et al.2320055-FU-FA+GAbstract466No
Hermann et al.212005Capecitabine+GAbstract319No
Cunningham et al.42005Capecitabine+GAbstract533Yes
Rocha Lima et al.272004Irinotecan+GArticle360No
Oettle et al.242005Pemetrexed+GArticle565No
Abou-Alfa et al.262006Exatecan+GArticle349No
Stathopoulos et al.252006Irinotecan+GArticle130No
Bramhall et al.72002Marimastat+GArticle239No
Van Cutsem et al.52004Tipifarnib+GArticle688No
Shapiro et al.92005G17DT+GAbstract383No
Moore et al.282007Erlotinib+GAbstract569Yes

Combined Analysis

The combined results from all trials that compared single-agent G versus any G-based combination are summarized in Table 2, and the Forest plot for OS is shown in Figure 1. For the primary endpoint of OS, no significant differences were observed between single-agent G and G-based combinations (RR, 0.93; 95% CI, 0.84–1.03; P = .17) with no significant heterogeneity (P = .99). Conversely, the analysis of PFS slightly but significantly favored the combination arm, with an RR of 0.91 (95% CI, 0.84–0.98; P = .015) without significant heterogeneity (P = .29); this difference translated into a 2.6% absolute benefit (NNT = 39 patients). A similar advantage for G-based combinations was observed in terms of the ORR (RR, 1.57; 95% CI, 1.31–1.86; P < .0001); and, although heterogeneity was observed between different trials (P = .026), the difference in favor of the combination arm remained significant utilizing a random-effect model (Table 2). This difference translated into a 3% absolute ORR benefit (NNT = 33 patients).

Figure 1.

Overall survival (OS) analysis of all randomized controlled trials (RCTs). All RCTs that compared gemcitabine (G)-based combinations (Drug + G) with single-agent G (n = 20 trials) were taken into account for the OS analysis, and the effects of the combination were analyzed using both a fixed-effect model (Fixed) and a random-effect model (Random). Results are shown as Forest plots, and P values are indicated. Individual trials have been grouped according to the combination drug for the sensitivity analysis. F-G indicates fluoropyrimidines; ND, new drugs; other, nonplatinum/nonfluoropyrimidine chemotherapeutic agents; P, platinum salts. Note that risk ratios (RR) are to be interpreted as follows: with regard to the endpoints of OS and progression-free survival, an RR <1.0 indicates fewer events (deaths or progressions) in the experimental arm and, thus, favors the combination; with regard to the endpoint of objective response rate, a RR <1.0 indicates fewer objective responses in the experimental arm and, thus, favors single-agent G.

Table 2. Comparison Between Gemcitabine-based Combinations and Single-agent Gemcitabine: All Randomized Clinical Trials
VariableNo. of patients (No. of RCTs)RR [95% CI]PHeterogeneityAB, %NNT
  • RCTs indicates randomized clinical trials; RR, relative risk; 95% CI, 95% confidence interval; AB, absolute benefit; NNT, number needed to treat; OS, overall survival; PFS, progression-free survival; ORR: overall response rate.

  • *

    Fixed-effect model.

  • Random-effect model.

OS6296 (20)0.93 [0.84–1.03].170.99
PFS5174 (17)0.91 [0.84–0.98].0150.292.639
ORR6296 (20)1.57 [1.31–1.86]*<.00010.026333
  1.51 [1.17–1.94].001   

Sensitivity Analysis

The impact of different classes of cytotoxic/molecularly targeted agents was explored further by clustering the selected RCTs into 4 separate groups (P-G, F-G, other-G, and ND-G). Figure 1 shows that none of the combined treatment groups demonstrated a significant OS advantage compared with single-agent G, with no significant heterogeneity among different trials. Platinum-containing combinations homogeneously demonstrated a significant advantage in terms of PFS (RR, 0.67; 95% CI, 0.53–0.83; P = .0004) and ORR (RR, 1.77; 95% CI, 1.32–2.37; P = .0001) (Table 3). These differences translated into a 10% absolute PFS benefit (NNT = 10 patients) and a 6.5% absolute ORR benefit (NNT = 15 patients), respectively, for P-G combinations (Table 3). A similar ORR effect was observed for other-G combinations (RR, 1.96; 95% CI, 1.41–2.93; P = .0005) that did not translate into a PFS or OS benefit (data not shown). A trend toward improved PFS was observed for the F-G group (RR, 0.86; 95% CI, 0.73–1.02).

Table 3. Comparison Between Combinations of Gemcitabine Plus Platinum Salts and Single-agent Gemcitabine
VariableNo. of patients (No. of RCTs)RR [95% CI]PHeterogeneityAB, %NNT
  1. RCTs indicates randomized clinical trials; RR, relative risk; 95% CI, 95% confidence interval; AB, absolute benefit; NNT, number needed to treat; OS, overall survival; PFS, progression-free survival; ORR: overall response rate.

OS1354 (7)0.83 [0.67–1.03].100.99
PFS765 (5)0.67 [0.53–0.83].00040.781010
ORR1354 (7)1.77 [1.32–2.37].00010.266.515

Correlation Between Primary and Secondary Endpoints

We explored correlations between primary and secondary endpoints among the trials that were included in our analysis. Nineteen RCTs that involved 6288 patients and 17 RCTs that involved 4882 patients were identified for the ORR/OS and PFS/OS correlation, respectively. When considering RRs, the ORR did not correlate significantly with OS (r = −0.17 and R2 = 0.03, P = .46; ρ = −0.18, P = .44) (Fig. 2), whereas PFS showed a strong linear correlation (r = 0.91 and R2 = 0.82, P < .0001; ρ = 0.59, P = .01) (Fig. 3). Similar results were obtained when considering the ORR and the median OS (r = 0.23 and R2 = 0.06, P = .14; ρ = 0.16, P = .33) and the median PFS and OS (r = 0.75 and R2 = 0.56, P < .0001; ρ = 0.90, P < .0001; data not shown). Based on these data, we also derived a model to calculate sample size estimates using PFS as a surrogate endpoint to observe OS improvements of 0.4 months, 0.7 months, and 1.5 months favoring the combination versus single-agent therapy. According to our model, 0.5-, 1-, and 2-month improvements in PFS would need to be observed to translate into the target OS advantages. This would translate into sample size estimates for the relevant clinical trials of 2370 patients, 678 patients, and 222 patients, respectively.

Figure 2.

Analysis of the correlation between the overall response rate (ORR) and overall survival (OS). The relative risks (RR) for ORR (x-axis) were plotted against the RR for OS (y-axis) for each trial that was included in the meta-analysis, and a regression analysis was performed. Pearson correlation test (r) and R2 values are shown. Each circle represents an individual trial, and its diameter is proportional to the number of patients included.

Figure 3.

Analysis of the correlation between progression-free survival (PFS) and overall survival (OS). The relative risks (RR) for PFS (x-axis) were plotted against the RR for OS (y-axis) for each trial that was included in the meta-analysis, and a regression analysis was performed. Pearson correlation test (r) and R2 values are shown. Each circle represents an individual trial, and its diameter is proportional to the number of patients included.


Since the introduction of G as first-line treatment for advanced pancreatic cancer in 1997,2 many attempts have been made to improve on the results obtained with single-agent chemotherapy. A number of randomized studies combining G with other chemotherapeutic agents, mostly in doublets, have been conducted with conflicting results. More recently, novel molecularly targeted agents also have been tested in combination with G. Although only 3 of the studies that we considered for the current meta-analysis reported a survival advantage in favor of the combination arm,4, 15, 28 a better ORR was observed in the majority of trials, and the overall negative results in terms of OS have been attributed largely to a lack of statistical power to detect small but significant differences. For example, both in the French Multidisciplinary Cooperative Oncology Group/Italian Group for the Study of Digestive Tract Cancer intergroup trial, which evaluated the combination of G and oxaliplatin,3 and in the trial conducted by Cunningham et al., who evaluated the combination of G and capecitabine,4 a comparably significant ORR benefit was observed, but only in the latter trial was a statistically significant survival advantage observed, possibly by virtue of a more adequately powered sample size (533 patients vs 313 patients). This issue is of paramount importance in the evaluation of therapeutic progress in pancreatic cancer, in which multiple factors other than sample size, such as inherent chemoresistance and poor patient performance status, may obscure the potential advantages of multiagent chemotherapy.

One possible interpretation of the results from RCTs is that single-agent G remains the standard of care in advanced PDAC, given the absence of a reported benefit for combination therapy in the majority of trials, the large NNT required to observe 1 benefit, and the poor performance status of most patients. However, a significant benefit in terms of activity generally was believed sufficient to introduce G-based doublets (especially doublets that contained platinum salts or fluoropyrimidines) as one of the front-line options for clinical practice, as reflected by their inclusion into several guidelines. Therefore, we believed that a cumulative analysis could help put such discrepancies in perspective and allow verification or rejection of the hypothesis that there was a beneficial effect of combination therapy with adequate statistical power. Based on the results of the current meta-analysis, which included all phase 3 trials that compared G with G-based combinations, single-agent G remains the standard of care for advanced pancreatic cancer. Indeed, no statistical difference in OS, which was our primary endpoint, was observed in favor of the combination arm. Similar results have been reported recently by the Cochrane Collaboration Group.29 In that meta-analysis, despite a significant 6-month survival benefit in favor of platinum-based combinations, no significant advantage at 6 months or 12 months was determined when all trials were included. Even in the most favorable group (ie, those who received the platinum-based regimen), the survival trend was not maintained at 12 months. Strikingly different results have been reported recently by Heinemann et al.30: in their meta-analysis, they observed a slightly significant OS benefit in favor of combinations, particularly those that contained a platinum salt or a fluoropyrimidine, over standard G. Methodological issues may explain the apparently discrepant results achieved by our current meta-analysis and that of Heinemann et al. First, we systematically excluded Phase II randomized trials.31 Second, the statistical methodology applied by Heinemann et al. differed from ours in the selection of effect measurement (HR vs RR).30 On closer inspection, however, the results obtained by these 3 meta-analyses using different statistical approaches and trial-selection criteria ultimately convey a common message: The overall reduction in the risk of death obtained by using combination regimens is minimal in all 3 studies (RR: 0.88, 0.91, and 0.93 for the Cochrane Collaboration Group study, Heinemann et al's study, and the current meta-analysis, respectively). The interpretation of these results from a clinical standpoint would be that the improvement attainable with G-based combinations is small and has to be weighed against the increased toxicity of combination regimens,32 an issue that has to be taken into account when considering different therapeutic options in the palliative setting of advanced PDAC treatment.

It is disappointing that the introduction of new, molecularly targeted agents has had little impact on the survival of patients with advanced PDAC. Rather than leading to the dismissal of the combination of targeted agents with G as ineffective, these results call for an in-depth analysis of potential reasons for failure. The first reason may be the choice of agents, such as MMP-I, that largely are ineffective as single agents; indeed, combined analysis of the 2 randomized studies in head-to-head comparisons of MMP-I with G uniformly demonstrated a significantly worse outcome for patients who were allocated to experimental treatment (RR for OS 1.45; 95% CI; 1.01–2.10; P = .04 and RR for ORR .14; 95% CI; .05–.40; P < .0001, respectively; data not shown. P = .14, 95% CI; .05–.40, P < .0001, respectively; data not shown).33 Second, the chosen agent may fail to hit its putative target(s): This appears to be the case for FTIs, which may fail to block K-Ras activity effectively because of alternative activation pathways, such as geranyl-geranylation.34 General methodological issues, such as insufficient target validation, inadequate Phase II testing, and a rush to bring new compounds into phase 3 trials, also may have contributed to the general failure of new approaches to have an impact on the survival of patients with PDAC. At the moment, the most promising among the therapeutic strategies that employ molecularly targeted agents appears to be the addition of EGFR-targeted drugs to G. Indeed, in the second largest trial among all 20 RCTs that were considered in the current analysis,28 a small but statistically significant OS advantage was observed for the combination of erlotinib and G compared with standard, single-agent G.

The usefulness of the response rate as a surrogate endpoint for survival is hotly debated. Indeed, response according to the Response Criteria in Solid Tumors may be difficult to determine in PDAC, in which the evaluation of disease response in the locally advanced setting may be obscured by extensive desmoplasia and inflammatory changes; moreover, the accuracy and usefulness of objective tumor shrinkage in the evaluation of novel therapeutics with predominantly cytostatic, rather than cytotoxic, activity has been challenged. In that respect, the finding of a lack of correlation between ORR and OS suggests that PFS is a more reliable surrogate marker for survival in advanced PDAC trials; other surrogate markers, such as a drop in CA19.9 serum levels during treatment,35, 36 may prove useful but will require prospective validation.

Overall, the results of the current meta-analysis do not indicate a significant impact of multiagent, systemic treatments on the survival of patients with advanced PDAC and do not support their routine use in clinical practice despite the observed advantage in terms of ORR and PFS. In view of the potential limitations of literature-based meta-analyses and of the conflicting results reported by other authors using similar approaches,29, 30, 32 we believe that a meta-analysis based on individual patient data would be of great value. Although progress in the systemic treatment of patients with advanced PDAC in the past 10 years appears to be small, rather than fostering therapeutic nihilism, the current data should prompt a profound revision of our approach to clinical research in pancreatic cancer37 that takes into account the need for identification of novel targets, more accurate preclinical/early clinical target validation, more extensive and innovative Phase II testing, and identification and validation of alternative surrogate markers for clinical efficacy.


Supported in part by grants from The Italian Ministry of Health and the Italian Association for Cancer Research.