Participants

The median age of participants in the population included in the meta‐analyses was 60 years (range 53 to 72 years). The proportion of participants reported as having metastatic disease ranged from 69% in Al‐Batran 2013 to 100% in Nicolaou 1982. We could not extract the percentage of people with metastasis in 5 out of 19 studies included in the meta‐regression analyses. For information on individual studies, see Characteristics of included studies.

With regard to baseline differences, we saw a five‐year difference in median age between the study arms in one trial (Rao 2010). In another, the proportion of the metastatic sites involved per participant was not similar at baseline: 29% of participants in the intervention group had one site involved, compared to 58% in the BSC group (Thuss‐Patience 2011). Performance status was well balanced in all studies, with no differences greater than 15% between study arms. The percentage of participants with ECOG‐2 or ‐3 was in the range of 0% to 35%.

Seven studies included in the meta‐analyses took place completely or partially in Asia (Bang 2010; Fuchs 2014; Huang 2009; Lordick 2013; Ohtsu 2011; Shen 2014; Wilke 2014; Xu 2013a). One study was translated from Chinese (Huang 2009).

Two studies selected participants based on biomarker expression. The effect of a targeted therapeutic can be expected to depend on the presence of the target in the tumor. The first included only EGFR‐positive participants (Lorenzen 2009). The second included only participants that were eligible if their tumor samples, stained with HER2, were scored as 3 or more on immunohistochemistry or if they tested positive using fluorescent in situ hybridization (FISH) analysis (HER2:CEP17 ratio ≥ 2) (Bang 2010). The results of these studies cannot be extrapolated to people that do not have biomarker expression as defined by these studies.

Comparisons

Studies excluded from the comparisons

Five studies included only eligible participants, however, these studies were not eligible for inclusion in any of our comparisons. Waddell 2013 studied panitumumab and adjusted the control regimen of epirubicin, oxaliplatin, and capecitabine in the experimental arm. One study was not eligible because it compared leucovorin and 5‐FU versus S‐1 on a background of cisplatin (Pang 2014). Three studies used an equal number of agents in each arm. These studies compared: atofluding versus ftorafur on a background of either mitomycin C plus etoposide, cisplatin plus hydroxycamptothecin, cisplatin plus vindesine, or mitomycin C plus adriamycin (Li 2002), epirubicin versus mitomycin on a background of cisplatin plus 5‐FU (Ross 2002), and adriamycin versus methotrexate or 5‐FU (Ezdinli 1980).

There were 15 studies that also included participants with gastric cancer that we excluded from the comparisons because they did not compare the addition of an agent to an unaltered control regimen. These studies compared: S1 versus 5‐FU on a background of cisplatin (Ajani 2010); oxaliplatin versus cisplatin on a background of 5‐FU plus leucovorin (Al‐Batran 2008); cisplatin versus oxaliplatin on a background of epirubicin plus 5‐FU or capecitabine, and 5‐FU versus capecitabine on a background of epirubicin plus cisplatin or oxaliplatin (Cunningham 2008); irinotecan plus folinic acid versus cisplatin on a background of 5‐FU (Dank 2008); continuous 5‐FU plus cisplatin versus bolus 5‐FU plus leucovorin (Duffour 2006); irinotecan versus cisplatin on a background of capecitabine (Moehler 2010); 5‐FU versus cisplatin on a background of irinotecan (Pozzo 2004); irinotecan versus 5‐FU on a background of docetaxel (Roy 2012); and epirubicin plus cisplatin versus doxorubicin plus methotrexate on a background of 5‐FU (Waters 1999). Five studies did add an agent to a control regimen but adjusted their control regimens in the experimental arm. They investigated docetaxel plus cisplatin (Ajani 2005); docetaxel plus oxaliplatin, either with or without 5‐FU (Van Cutsem 2015); lapatinib (Lorenzen 2015); docetaxel plus oxaliplatin with or without capecitabine (Van Cutsem 2015); or cisplatin and 5‐FU with or without docetaxel (Van Cutsem 2006). One study was not eligible for inclusion because it tested two agents (cisplatin, 5‐FU) versus one agent (capecitabine) (Tebbutt 2010).

In order to analyze the data from studies with a mixed participant population, including both eligible and ineligible participants, we requested individual participant data from authors of studies that we included after the search round of 3 October 2013. Only one author responded and provided individual participant data. Subsequently, we decided to investigate the information from the studies that included both esophageal and/or GE‐junction, mixed with gastric cancer participants, in a sensitivity analysis to assess the influence of gastric cancer participants on the outcome of the individual studies.

Outcomes

OS and toxicity were the most commonly described outcomes, followed by PFS, time to progression (TTP), and objective response rate. Studies did not always classify toxicity according to NCI‐CTC or WHO. Studies published before 2010 did not report quality of life with validated methods, and where reported, authors did not always report this outcome separately for esophageal and GE‐junction cancer subgroups.

Overall survival

Eleven studies in 1347 participants contributed data to this meta‐analysis (Bang 2010; Bleiberg 1997; Dutton 2014; Ford 2014; Fuchs 2014; Huang 2009; Levard 1998; Lordick 2013; Lorenzen 2009; Nicolaou 1982; Wilke 2014). These studies included only people with esophageal and/or GE‐junction cancer, or they reported reported the results separately for this group. The overall HR in favor of the arm with the additional agent was 0.75 (95% CI 0.68 to 0.84, high‐quality evidence), showing an OS benefit (Analysis 1.1; Figure 3). On average, participants in the arm with the additional chemotherapy or targeted therapy agent received 2.1 chemotherapy or targeted therapy agents, versus an average of 1.0 agents in the control arms. Median OS, weighted for study size, in the arm with the additional agent was 6.7 months versus 5.7 months in the control arm. We could not include two studies because they did not report median overall survival data (Bang 2010; Huang 2009). Cochrane's Q test for heterogeneity showed a non‐significant amount of heterogeneity (I² = 5%, P = 0.40), which indicated that results of the different studies were consistent in their findings. We did not note any small study effects such as publication bias (Figure 4).

Figure 3

---

---

Forest plot of the main analysis: chemotherapy or targeted therapy agent(s) plus control arm versus control arm, outcome: 1.1 Overall survival.

Figure 4

---

---

Funnel plot of the main comparison: chemotherapy or targeted therapy agent(s) plus control arm versus control arm, outcome: 1.1 Overall survival.

Two studies included in the meta‐analysis selected participants based on biomarker expression (Bang 2010; Lorenzen 2009). We could not include three studies under this comparison in the meta‐analysis because of insufficient information. Eatock 2013 investigated trebananib. Median OS times at the time of analysis were 9.1 months, 9.4 months, and 12.8 months for the 10.0 mg/kg, 3.0 mg/kg, and control arms, respectively. Wilkes 2011 studied thalidomide and reported that survival was not affected by group allocation. Xu 2013a found higher, statistically significant (P < 0.05) OS in the nimotuzumab group relative to the control group.

Progression‐free survival

Of all studies included in this comparison, only studies that investigated a targeted therapy agent reported PFS. Therefore, we describe the results of this analysis in more detail under subcomparison 4, 'Progression‐free survival'. The analysis contained five studies in 883 participants. The addition of a targeted therapeutic agent probably leads to an HR of 0.64 (95% CI 0.45 to 0.92, moderate‐quality evidence; Analysis 1.2; Figure 5). Eatock 2013 investigated trebananib in esophageal adenocarcinoma participants but did not provide enough detail for inclusion in the meta‐analysis. Median PFS times at the time of analysis were 4.2 months, 4.9 months, and 5.2 months for the 10.0 mg/kg, 3.0 mg/kg, and control arms, respectively.

Figure 5

---

---

Forest plot of the main analysis: chemotherapy or targeted therapy agent(s) plus control arm versus control arm, outcome: 1.2 Progression free survival.

Toxicity

All 11 studies included in the main analysis reported on toxicity. All studies reported the frequency of toxic effects, although not all used WHO or NCI‐CTC guidelines. Five studies did not report toxicity for the esophageal and GE‐junction cancer subgroup separately (Bang 2010; Ford 2014; Fuchs 2014; Lordick 2013; Wilke 2014). Therefore, in this review, we only compared toxicity between study arms, not between studies. Nicolaou 1982 and Levard 1998 did not use WHO or NCI‐CTC guidelines. All other studies used NCI‐CTC, except for two that used WHO guidelines (Bleiberg 1997; Huang 2009).

In Bang 2010, the most common grade 3 and 4 adverse events reported with trastuzumab plus chemotherapy versus chemotherapy alone were neutropenia (27% versus 30%), anemia (12% versus 10%), and diarrhea (9% versus 4%). Treatment‐related mortality was 3% in the trastuzumab plus chemotherapy arm, versus 1% in the chemotherapy alone arm. Bleiberg 1997 reported that grade 3 and 4 adverse events occurred most frequently in the 5‐FU cisplatin arm versus the cisplatin alone arm. These were nausea or vomiting (27% versus 11%) and leukocytopenia and thrombocytopenia (both 12% versus 0%). Seven (16%) treatment‐related deaths occurred in the treatment arm. Dutton 2014 reported that any grade 3, 4, or 5 toxicities occurred in 45% of participants in the gefitinib plus control arm versus 39% of participants in the control arm. Two treatment‐related deaths occurred in the placebo group and one in the gefitinib group. Ford 2014 found that grade 4 toxicities occurred more frequently in participants treated with docetaxel compared to participants in the control arm (21% versus 4%). Neutropenia, infections, and febrile neutropenia were the toxicities that differed most between the study arms. None of the deaths were attributed to the treatment. In both arms of Fuchs 2014, 2% of the participants died due to drug‐related toxicity. Ramucirumab was not associated with increased rates of fatigue, decreased appetite, vomiting, anemia, or other notable toxic effects. Huang 2009 used WHO guideline classifications but did not specify the grade of the side effects that occurred. In Lordick 2013, any grade 3 or 4 adverse events occurred in 83% of participants in the cetuximab plus control arm versus 77% in the control arm. Nine per cent of participants in the chemotherapy plus cetuximab arm and eight per cent of participants in the control arm had an adverse event leading to death. Lorenzen 2009 reported that grade 3 and 4 adverse events, which occurred more frequently in the cetuximab group, were diarrhea (16% versus 0%), neutropenia (22% versus 13%), and rash (6% versus 0%). Additionally, they reported one (3%) treatment‐related death in the control arm and none in the experimental arm. Wilke 2014 found that the most frequently occurring grade 3, 4, and 5 adverse events in the ramucirumab arm versus the control arm were neutropenia (41% versus 19%), leukopenia (18% versus 7%), and hypertension (15% versus 3%). In both arms, 2% of participants had adverse events leading to death with a causal relation to the study drugs.

Overall, palliative chemotherapy and/or targeted therapy appears to increase the frequency of treatment‐related toxicity of at least grade 3. Treatment‐related deaths were rare in most studies, and there is no clear evidence that treatment‐related deaths occur more frequently in the study arms with an additional chemotherapy or targeted therapy agent.

Quality of life

Five studies included in the main analysis did not report on quality of life (Bleiberg 1997; Lordick 2013; Lorenzen 2009; Nicolaou 1982; Wilkes 2011). Four studies did not report quality of life separately for the esophageal and GE‐junction cancer subgroup (Bang 2010; Ford 2014; Fuchs 2014; Wilke 2014). Two studies reported on quality of life but did not use validated methods (Huang 2009; Levard 1998). Huang 2009 used improvement after treatment in the treatment arm versus the control arm for the Karnofsky score (33% versus 10%) and body weight (27% versus 6.7%). Levard 1998 studied dysphagia in order to judge the quality of life during the survival of their participants. Bang 2010 (see Satoh 2014) reported that trastuzumab plus chemotherapy versus chemotherapy alone prolonged time to 10% definitive deterioration in all QLQ‐C30 and QLQ‐STO22 scores (Aaronson 1993; Blazeby 2004), including QLQ‐C30 global quality of life score, from 6.4 months to 10.2 months. Dutton 2014 reported no differences between the gefitinib and placebo groups in global quality of life measured with QLQ‐C30. However, odynophagia worsened for participants on placebo and improved significantly for participants on gefitinib. Ford 2014 reported that the mean quality‐adjusted life weeks (QLQ‐C30) were 12.1 weeks (standard deviation (SD) 0.84) for the docetaxel group and 9.3 weeks (SD 0.73) for the control group. Fuchs 2014 reported a trend toward better quality of life (QLQ‐C30) at six weeks for participants in the ramucirumab group compared to those in the placebo group (P = 0.23). Median time of deterioration to a score of 2 or worse in ECOG performance status was 2.4 months (95% CI 1.3 to not reached) in the placebo group and 5.1 months (95% CI 1.9 to 16.8) in the ramucirumab group. Wilke 2014 reported that baseline and end‐of‐treatment results for global quality of life from the QLQ‐C30 and index scores from the EQ‐5D‐3L were similar in the treatment groups (EuroQol 1990). Overall, the studies reporting quality of life did so in different ways. Although recent studies often use the QLQ‐C30, the outcomes were reported in the form of either mean values with SD, change from the baseline, proportions improved, or mean area under the curve. This prohibited a meta‐analysis of quality of life outcomes. The five studies were not representative for all the studies in this analysis, as four of them tested a targeted agent, and four did not report data separately for the esophageal and GE‐junction cancer subgroup. However, the quality of life improved in the arms with the additional agent.

Overall survival

Five studies in 750 participants contributed data to this meta‐analysis (Dutton 2014; Ford 2014; Fuchs 2014; Levard 1998; Nicolaou 1982). For overall survival, we found an HR of 0.81 (95% CI 0.71 to 0.92, high‐quality evidence; Analysis 4.1;Figure 6) in favor of the chemotherapy or targeted therapy arm. Median OS, weighted for study size, in the chemotherapy arm was 4.7 months versus 4.2 months in the BSC arm. Only two studies used first‐line therapies (Levard 1998; Nicolaou 1982), while the others used second‐line. Cochrane's Q test for heterogeneity was non‐significant (I² = 0%, P = 0.56), which indicated that results of the five studies were consistent in their findings.

Figure 6

---

---

Forest plot of subcomparison 1: chemotherapy or targeted therapy agent(s) plus BSC versus BSC, outcome 4.1: overall survival.

Progression‐free survival

Two studies in 540 participants contributed data to this meta‐analysis (Dutton 2014; Fuchs 2014). Both studies assessed a targeted therapeutic agent. The other three studies did not report on PFS. The overall HR of 0.58 (95% CI 0.28 to 1.18, very low‐quality evidence; Analysis 4.2; Figure 7) in favor of targeted therapy demonstrated that there was a larger, non‐significant effect estimate toward a PFS benefit when participants received targeted therapy; however, we are uncertain of these findings due to the very low quality of evidence. Median progression‐free survival was only available from Dutton 2014 and was 1.6 months in the targeted therapy arm versus 1.2 months in the BSC arm. Cochrane's Q test for heterogeneity showed significant heterogeneity (I² = 85%, P = 0.01), which indicated that results of the two studies were not consistent.

Figure 7

---

---

Forest plot of subcomparison 1: chemotherapy or targeted therapy agent(s) plus BSC versus BSC, outcome 4.1: progression free survival.

Overall survival

This meta‐analysis included four studies in 769 participants (Dutton 2014; Ford 2014; Fuchs 2014; Wilke 2014). Three studies included participants with GE‐junction cancer only. In Dutton 2014, 24% of the esophageal cancers were SCC. Two studies investigated ramucirumab (Fuchs 2014; Wilke 2014), which we analyzed separately in subgroup analysis 4b. The overall HR of 0.71 (95% CI 0.54 to 0.94, moderate‐quality evidence; Analysis 5.1) in favor of the arm with the additional agent demonstrated that an additional chemotherapeutic or targeted agent leads to a survival benefit. Median OS, weighted for study size, was 5.1 months in the chemotherapy arm versus 4.4 months in the BSC arm. Cochrane's Q test for heterogeneity showed significant heterogeneity (I² = 57%, P = 0.07), indicating that results of the three included studies were quite inconsistent in their findings.

Progression‐free survival

Three studies in 677 participants contributed data to this meta‐analysis (Dutton 2014; Fuchs 2014; Wilke 2014). All three studies investigated targeted agents. Two studies (both of ramucirumab) included participants with GE‐junction cancer only. In the third study, 24% of the esophageal cancers were SCC. The overall HR of 0.51 (95% CI 0.29 to 0.90, low‐quality evidence; Analysis 5.2) in favor of the targeted therapy arms demonstrated that there may be a PFS benefit for people receiving targeted therapy agents. Cochrane's Q test for heterogeneity showed substantial heterogeneity (I² = 83%, P < 0.001), which indicated that results of the three studies were inconsistent in their findings.

Overall survival

Five studies in 358 participants contributed data to this meta‐analysis (Bleiberg 1997; Ford 2014; Huang 2009; Levard 1998; Nicolaou 1982). The overall HR of 0.73 (95% CI 0.63 to 0.85, moderate‐quality evidence; Analysis 1.1; Figure 3, subcomparison 3) in favor of the arm with the additional chemotherapy agent demonstrates that there is probably a significant survival benefit in people receiving an additional chemotherapy agent. Median survival time, weighted for study size, was 6.9 months in the chemotherapy arm versus 5.8 months in the control arm. We could not include Huang 2009 in the median OS analysis because it did not report this outcome. Cochrane's Q test for heterogeneity showed non‐significant heterogeneity (I² = 0%, P = 0.50), which indicated that results of the five studies were consistent in their findings. Considering the limited number of studies and the low heterogeneity in the model, we consider that the quality of evidence is moderate.

One study included in the comparison did not contribute to the meta‐analysis for subcomparison 3 for OS because the publication did not provide enough information. Wilkes 2011 only reported that survival was not affected by group allocation or whether the participant was able to complete the protocol.

Progression‐free survival

None of the four studies reported on PFS.

Overall survival

Six studies with 989 participants contributed to this meta‐analysis (Bang 2010; Dutton 2014; Fuchs 2014; Lordick 2013; Lorenzen 2009; Wilke 2014). The overall HR in favor of the arm containing a targeted agent was 0.75 (95% CI 0.63 to 0.90, high‐quality evidence). This analysis demonstrated a survival benefit for participants randomized to the arm with the additional targeted agents (Analysis 1.1; Figure 3, subcomparison 4). Median OS in the arm with the additional targeted agent, weighted for study size, was 6.7 months versus 5.7 months in the control arm. We could not include Bang 2010 because it did not report median OS data. Cochrane's Q test for heterogeneity showed low heterogeneity (I² = 24%, P = 0.25), indicating that results of the different studies were quite consistent in their findings.

Two studies included in the meta‐analysis selected participants based on biomarker expression (Bang 2010; Lorenzen 2009). Xu 2013a found higher PFS and OS in the nimotuzumab group. However, the study did not provide enough information to extract HRs, so we could not include it in the analyses.

Progression‐free survival

Five studies in 883 participants contributed data to this meta‐analysis (Dutton 2014; Fuchs 2014; Lordick 2013; Lorenzen 2009; Wilke 2014). The five studies investigated targeted therapies. In this analysis, the overall HR, in favor of the treatment arm that contained a targeted therapy agent, was 0.64 (95% CI 0.45 to 0.92, moderate‐quality evidence). This showed that there is probably a PFS benefit for people treated with the targeted therapy agent (Analysis 1.2;Figure 5). Median progression‐free survival, weighted for study size, was 2.9 months in the arm with the additional targeted therapy agent versus 2.4 months in the control arm. We could not include two studies in the analysis because they did not report median PFS data (Fuchs 2014; Wilke 2014). Cochrane's Q test for heterogeneity showed substantial heterogeneity (I² = 79%, P < 0.001), which indicated that results of the individual studies were quite inconsistent in their findings. One study included in the meta‐analysis selected participants based on biomarker expression (Lorenzen 2009).

Overall survival

Five studies in 538 participants contributed data to this meta‐analysis (Bang 2010; Ford 2014; Fuchs 2014; Lordick 2013; Wilke 2014). For overall survival, we found an HR of 0.66 (95% CI 0.54 to 0.81, high‐quality evidence; Analysis 9.1) in favor of the experimental arm. Median OS, weighted for study size, was 7.1 months in the added agent arm versus 6.0 months in the control arm. We could not include Bang 2010 because it did not report median overall survival data. Two studies were first‐line therapy regimens (Bang 2010; Lordick 2013), and three were second‐line therapies (Ford 2014; Fuchs 2014; Wilke 2014). Cochrane's Q test for heterogeneity was non‐significant (I² = 0%, P = 0.55), which indicated that results from the five studies were consistent in their findings. Eatock 2013 investigated trebananib in participants with esophageal adenocarcinoma but did not provide enough detail for inclusion in the meta‐analysis. Median OS at the time of analysis were 9.1 months, 9.4 months, and 12.8 months for the 10.0 mg/kg, 3.0 mg/kg, and control arms, respectively.

Progression‐free survival

Four studies in 713 participants contributed data to this meta‐analysis (Dutton 2014; Fuchs 2014; Lordick 2013; Wilke 2014). For progression‐free survival, we found an HR of 0.62 (95% CI 0.38 to 1.00, very low‐quality evidence; Analysis 9.2) in favor of the experimental arm; however, we are uncertain of these results due to the low‐quality evidence. Median OS, weighted for study size, was 1.8 months in the added agent arm versus 1.7 months in the control arm. We could not include two studies in the analysis because they did not report median PFS data (Fuchs 2014; Wilke 2014). One study was on first‐line therapy regimens (Lordick 2013), and three were second‐line therapies (Dutton 2014; Fuchs 2014; Wilke 2014). Cochrane's Q test for heterogeneity was non‐significant (I² = 84%, P < 0.001), which indicated that results of the four studies were very inconsistent in their findings. Eatock 2013 investigated trebananib in participants with esophageal adenocarcinoma but did not provide enough detail for inclusion in the meta‐analysis. Median PFS times at the time of analysis were 4.2 months, 4.9 months, and 5.2 months for the 10.0 mg/kg, 3.0 mg/kg, and control arms, respectively.

Overall survival

Four studies in 268 participants contributed data to this meta‐analysis (Bleiberg 1997; Levard 1998; Lorenzen 2009; Nicolaou 1982). For overall survival, we found an HR of 0.76 (95% CI 0.65 to 0.90, high‐quality evidence; Analysis 10.1) in favor of the experimental arm. Median OS, weighted for study size, was 8.0 months in the added agent arm versus 6.5 months in the control arm. All studies were first‐line therapy regimens. Cochrane's Q test for heterogeneity was non‐significant (I² = 0%, P = 0.95), which indicated that results of the four studies were consistent in their findings. Xu 2013a found that the OS of the nimotuzumab group was increased relative to the control group.

Progression‐free survival

Two studies in 168 participants contributed data to this meta‐analysis (Dutton 2014; Lorenzen 2009). For overall survival, an HR of 0.72 (95% CI 0.55 to 0.96, low‐quality evidence; Analysis 10.2) showed that there may be a benefit in favor of the experimental arm. Median OS, weighted for study size, was 1.7 months in the added agent arm versus 1.2 months in the control arm. All studies were first‐line therapy regimens. Cochrane's Q test for heterogeneity was non‐significant (I² = 0%, P = 0.97), which indicated that results of the four studies were consistent in their findings. Xu 2013a found that the PFS of the nimotuzumab group was higher than in the control group.