Salvage systemic therapy for advanced gastric and oesophago‐gastric junction adenocarcinoma

Yoko Tomita; Max Moldovan; Rachael Chang Lee; Amy HC Hsieh; Amanda Townsend; Timothy Price

doi:10.1002/14651858.CD012078.pub2

. 2020 Nov 19;2020(11):CD012078. doi: 10.1002/14651858.CD012078.pub2

Salvage systemic therapy for advanced gastric and oesophago‐gastric junction adenocarcinoma

Yoko Tomita ^1,^✉, Max Moldovan ², Rachael Chang Lee ³, Amy HC Hsieh ¹, Amanda Townsend ¹, Timothy Price ¹

Editor: Cochrane Gut Group

PMCID: PMC8094513 PMID: 33210731

Abstract

Background

Salvage systemic therapy has become the new standard of care in patients with advanced gastric and oesophago‐gastric junction (OGJ) adenocarcinoma, following disease progression on first‐line fluoropyrimidine and platinum‐containing chemotherapy. Pharmacological agents proven to be effective in this setting include both chemotherapy and biological therapy, however, the consensus on the best salvage systemic therapy has not been reached.

Objectives

To assess the effects of systemic chemotherapy and biological therapy, either alone or in combination, on overall survival (OS) and progression‐free survival (PFS) in patients with advanced gastric and OGJ adenocarcinoma, whose disease has progressed on, or relapsed after first‐line fluoropyrimidine and platinum‐containing chemotherapy. Adverse events (AEs), tumour response rate (TRR) and quality of life (QoL) associated with systemic chemotherapy and/or biological therapy were additionally assessed.

Search methods

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, trial registries and proceedings of the major oncology conferences up to October 2020. We additionally handsearched the reference lists of studies. No language restriction was applied.

Selection criteria

We included randomised controlled trials (RCTs) comparing salvage systemic therapy (chemotherapy and/or biological therapy) and either another type of salvage systemic therapy, placebo, best supportive care (BSC) or no treatment in patients with gastric and OGJ adenocarcinoma refractory to first‐line fluoropyrimidine and platinum‐containing chemotherapy.

Data collection and analysis

Two review authors independently performed selection of eligible studies and the primary author extracted study characteristics and outcome data from included studies. We assessed the quality and risk of bias of eligible studies according to the Cochrane Handbook for Systematic Reviews of Interventions. We expressed pooled estimates of effect using hazard ratio (HR) calculated using an inverse variance random‐effects model for time‐to‐event data, and risk ratio (RR) calculated using Mantel‐Haenszel random‐effects model for binary data. The certainty of evidence was graded using GRADEpro.

Main results

We identified 17 RCTs with 5110 participants for inclusion in this review. Tweenty‐nine studies are ongoing and twenty studies are awaiting classification. No studies examined the following comparisons: chemotherapy combined with biological therapy versus placebo, BSC or no treatment, chemotherapy combined with biological therapy versus biological therapy, biological therapy versus biological therapy and chemotherapy combined with biological therapy versus chemotherapy combined with biological therapy.

Chemotherapy versus placebo, best supportive care or no treatment

Chemotherapy probably improves OS (HR = 0.66, 95% CI 0.52 to 0.83, moderate‐certainty evidence) based on two studies involving 547 participants and improves PFS (HR = 0.57, 95% CI 0.47 to 0.69, high‐certainty evidence) based on one study involving 507 participants over placebo and BSC. Chemotherapy probably increases serious AEs (SAEs) (RR = 1.38, 95% CI 1.20 to 1.59, moderate‐certainty evidence) based on one study involving 503 participants.

Biological therapy versus placebo, best supportive care or no treatment

Biological therapy improves OS (HR = 0.55, 95% CI 0.41 to 0.73, high‐certainty evidence) and probably improves PFS (HR = 0.33, 95% CI 0.19 to 0.57, moderate‐certainty evidence) over placebo based on three studies involving 781 participants. There is currently insufficient evidence for increased SAEs from biological therapy (RR = 1.14, 95% CI 0.95 to 1.37, low‐certainty evidence) based on two studies involving 638 participants.

Chemotherapy versus biological therapy

This comparison only considered immunotherapy. There is probably no evidence of a difference for OS (HR = 0.82, 95% CI 0.66 to 1.02, moderate‐certainty evidence) between chemotherapy and immunotherapy, and immunotherapy probably reduces PFS (HR = 1.27, 95% CI 1.03 to 1.57, moderate‐certainty evidence) based on one study involving 395 participants. SAEs may be less frequent with immunotherapy compared to chemotherapy (RR = 0.41, 95% CI 0.30 to 0.57, low‐certainty evidence).

Chemotherapy combined with biological therapy versus chemotherapy

Addition of biological therapy to chemotherapy probably does not improve OS (HR = 0.93, 95% CI 0.83 to 1.04, moderate‐certainty evidence) and we are uncertain whether it improves PFS (HR = 0.87, 95% CI 0.74 to 1.02, very low‐certainty evidence) based on seven studies involving 2743 participants. We are similarly uncertain whether combined chemotherapy and biological therapy increases SAEs (RR = 1.17, 95% CI 0.95 to 1.44, very low‐certainty evidence) based on four studies involving 1618 participants.

Chemotherapy versus chemotherapy

There is no evidence of a difference for OS and PFS between irinotecan and paclitaxel (HR = 1.13, 95% CI 0.86 to 1.48, low‐certainty evidence for OS; HR = 1.14, 95% CI 0.88 to 1.48, low‐certainty evidence for PFS) based on one study involving 219 participants. Similarly, there is no evidence to indicate improved OS and PFS from addition of another chemotherapy to docetaxel (HR = 1.05, 95% CI 0.72 to 1.54, low‐certainty evidence for OS; HR = 0.75, 95% CI 0.52 to 1.09, low‐certainty evidence for PFS) based on two studies involving 121 participants. Grade ≥ 3 neutropenia occurred commonly with both mono‐ and poly‐chemotherapy except for docetaxel‐S1 and EOX chemotherapy.

Authors' conclusions

Survival outcome of patients with advanced gastric and OGJ adenocarcinoma whose disease progressed on first‐line fluoropyrimidine and platinum‐containing chemotherapy can be improved by chemotherapy and biological therapy. Biological therapy, in particular, achieves this without clear increase in SAEs or QoL impairment. Whether biological therapy is preferred over chemotherapy is still unclear and there is no evidence of a difference for OS outcome, although immunotherapy may be associated with less SAEs. Addition of biological therapy to chemotherapy and poly‐chemotherapy are associated with frequent treatment‐related toxicity without clear survival benefit.

Plain language summary

Which treatments work best for advanced stomach cancer that has not responded to standard chemotherapy?

What is advanced stomach cancer?

Gastric (stomach) cancer usually begins in the mucous‐producing cells lining the stomach. Oesophago‐gastric junction (OGJ) cancer starts where the food pipe (oesophagus) joins the stomach. Advanced cancer is cancer that has spread to nearby tissues, or to another part of the body, despite treatment.

Treatments for gastric and OGJ cancer include:

• operation to remove the cancer;

• chemotherapy (medicines that kill cancer cells);

• radiotherapy (radiation to kill cancer cells); and

• biological therapy (medicines made from proteins and other substances that occur naturally in the body).

Biological therapies include immunotherapy (medicines that help the immune system to recognise and kill cancer cells) and therapies that target something in, or surrounding, the cancer, such as the cancer's blood supply. Standard chemotherapy usually combines two medicines containing fluoropyrimidine and platinum.

When standard chemotherapy for advanced cancer has not worked, further treatment aims to slow the growth of the cancer to help people live longer. Further treatments include: other chemotherapy medicines, biological therapies, and best supportive care (care that helps a person cope with life‐limiting illness and its treatment).

Why we did this Cochrane Review

Stomach and OGJ cancer are difficult to treat. We wanted to find out which treatments work best to control these cancers and help people live longer, when standard chemotherapy has not worked.

What did we do?

We searched for studies that looked at chemotherapy and/or biological therapies for advanced stomach or OGJ cancer that had not responded to standard chemotherapy. We looked for studies in which the treatment each person received was decided at random. These studies usually give the most reliable evidence about the effects of treatments.

Search date

We included evidence published up to October 2020.

What we found

We found 17 studies in 5110 people with advanced stomach or OGJ cancer. Studies compared further chemotherapy and/or biological therapies, given by mouth or through the bloodstream (systemic), with:

• another systemic chemotherapy and/or biological therapy;

• a placebo ('dummy' treatment);

• best supportive care; and

• no treatment.

The studies looked at:

• how long people lived;

• any adverse (unwanted) effects; and

• their quality of life (well‐being).

What are the results of our review?

People probably live longer after further chemotherapy (irinotecan or trifluridine plus tipiracil) than with placebo treatment or best supportive care. But chemotherapy probably increases serious unwanted effects, including diarrhoea, fever, and lower numbers of red and white blood cells.

People may live as long after irinotecan chemotherapy as after paclitaxel chemotherapy. Adding another chemotherapy (oxaliplatin or cisplatin) to docetaxel may not affect how long people live.

People live longer after biological therapy (nivolumab, apatinib or regorafenib) than with placebo treatment. We did not find enough evidence about whether biological therapy increases unwanted effects.

People given immunotherapy (pembrolizumab) probably live as long as people given chemotherapy (paclitaxel), but may not have as many unwanted effects as with chemotherapy.

Combining chemotherapy with biological therapy probably does not help people live longer than chemotherapy alone, and we are uncertain whether it increases unwanted effects.

How reliable are these results?

We are moderately confident that chemotherapy probably helps people to live longer than placebo treatment or best supportive care. We are confident that people live longer on biological therapy than placebo treatment. We think more evidence is unlikely to change this result.

We are less confident about the results for unwanted effects. Some studies had missing data or did not report these; and in some studies people and their doctors knew which treatment was given, which could have affected the study results. These results are likely to change when more evidence becomes available.

Conclusions

If advanced stomach or OGJ cancer has not responded to standard chemotherapy, further chemotherapy or biological therapy help people to live longer than placebo treatment, best supportive care, or no treatment. However, chemotherapy is more clearly associated with unwanted effects than biological therapy.

We are unsure if biological therapies work better than chemotherapy, but they may cause fewer unwanted effects. Combining chemotherapy and biological therapies may cause more unwanted effects without giving any extra benefit.

Summary of findings

Background

A glossary of terms is provided in Appendix 1.

Description of the condition

Despite gradual decline in its occurrence during recent decades, gastric cancer remains a major health burden internationally. It is the fifth most common cancer and there were over one million incident cases in 2018 worldwide (Ferlay 2018). The condition is particularly prevalent in Eastern Asia where 60% of worldwide cases occur. Gastric cancer is the third most common cause of cancer‐related death; it contributed to 783,000 deaths in 2018 worldwide. Causes of gastric cancer are multifactorial, although infection with Helicobacter pylori (H.pylori) is considered to be the primary carcinogenic step. Other risk factors include smoking, high salt consumption, and processed meat intake as well as genetic polymorphisms in hosts (Rawla 2019). Histologically, 95% of gastric cancer demonstrates adenocarcinoma.

Oesophago‐gastric junction (OGJ) cancer affects the border between the oesophagus and the stomach, and the vast majority of this type of cancer is adenocarcinoma. Unlike gastric adenocarcinoma, the incidence of OGJ adenocarcinoma has increased significantly since 1970 in Western countries. In England its age‐standardised incidence rose by 2.6 fold between the 1970s and 1990s, before plateauing (Offman 2018). Because of its location, the definition and staging of OGJ cancer has been the source of controversy. In the eighth edition of the American Joint Committee in Cancer (AJCC) staging system, adenocarcinomas with epicentres no more than 2 cm into the gastric cardia are staged as oesophageal cancer, while those extending further are staged as stomach cancers (Rice 2017). Risk factors for OGJ cancer include smoking, gastro‐oesophageal reflux disease, Barrett's oesophagus and obesity, while a diet high in fibre and H.pylori infection are inversely associated with the condition (Buas 2013). The shifts in dietary practices towards increased fat intake and meat consumption, together with the decline in prevalence of H.pylori infection in Western countries, are thought to have contributed to the rise in OGJ cancer's incidence in these countries. Improved site classification of OGJ and gastric cardia adenocarcinoma may also explain this change (Corley 2004).

Both gastric and OGJ cancers have a dismal prognosis. The five‐year survival rates of these cancers have been reported to be 32% and 20% in the USA, respectively (SEER 2018). Their high mortalities are explained by the majority of patients with these cancers presenting with either advanced disease or relapse following curative intervention. For both gastric and OGJ cancer, where radical resection remains the only treatment offering potential cure, the conditions are considered advanced if complete resection is not possible either due to local extension or the presence of distant metastases. Surgically curable early gastric cancers are usually asymptomatic and only infrequently detected outside screening programs. In many parts of the world where screening programs do not exist, more than 50% of gastric cancer cases are diagnosed at advanced stage (Wesolowski 2009). Similarly, more than 80% of patients with OGJ cancer have advanced disease at diagnosis (Siewert 2005). Prognoses of advanced gastric and OJG cancers are poor with the five‐year survival rate being less than 10%, however, selected patients still benefit from local and systemic therapies of palliative intention (SEER 2018).

Description of the intervention

Systemic therapy refers to the treatment of cancer using pharmacological agents, whereby drugs commonly administered orally or intravenously travel through the bloodstream to reach and affect cancer cells around the body. There are currently two types of systemic pharmacological therapy utilised in the management of gastric and OGJ cancers: chemotherapy and biological therapy. While chemotherapy is a group of cytotoxic chemicals used against cancer, biological therapy involves the use of substances derived from living organisms. Biological therapy is categorised based on its mechanism of action and the main types of biological therapy currently in clinical use or having been evaluated in large clinical trials for stomach and OGJ cancers include anti‐HER2 antibody, vascular endothelial growth factor receptor (VEGFR)‐targeted therapy and immunotherapy.

Fluoropyrimidine and platinum‐containing chemotherapy has been established as the backbone of first‐line systemic therapy for both advanced gastric and OGJ cancers, with some debate as to the benefit of additional taxane or anthracycline; however, the disease eventually progresses on the treatment (Cunningham 2008; Kang 2009; Van Cutsem 2006). Gastric and OGJ cancers, which relapse shortly after the perioperative or adjuvant chemotherapy containing these two agents, are similarly considered to be resistant to this regimen. For patients whose disease has progressed on first‐line chemotherapy, best supportive care (BSC) was previously the main management approach.

Salvage systemic therapy refers to second‐line and beyond chemotherapy and biological therapy, and they are administered following failure of first‐line systemic therapy. With an increasing number of studies being published supporting its benefit, second‐line pharmacological management of advanced gastric and OGJ cancer has now become the standard of care for patients with reasonable fitness (Kanagavel 2015). Chemotherapy such as irinotecan and taxanes were the first to be demonstrated to improve survival of these patients following the failure of fluoropyrimidine and platinum‐containing therapy (Ford 2014; Kang 2012; Thuss‐Patience 2011). More recently, there has been a rapid emergence of several biological therapy agents with proven anticancer activity for advanced gastric cancer in the salvage setting, expanding the pharmacological options for these cancers (Pavlakis 2016; Wilke 2014).

How the intervention might work

Salvage systemic therapy is administered with an intention to prolong and maximise quality of life (QoL). Both salvage chemotherapy and biological therapy have been demonstrated to delay progression of cancer growth and extend survival in a proportion of patients with gastric and OGJ cancers (Ford 2014; Kang 2012; Pavlakis 2016; Thuss‐Patience 2011; Wilke 2014). These therapies can also improve or delay the deterioration of cancer‐related symptoms such as fatigue, nausea/vomiting, pain and the level of functioning experienced by patients (Al‐Batran 2014; Ford 2014; Thuss‐Patience 2011; Wilke 2014).

Salvage systemic therapy can cause treatment‐related adverse events (AEs) in some patients. The type and the extent of AEs experienced depend on the exact therapy administered and the risk factors held by individual patients. The list of such treatment‐related AEs includes anaemia, diarrhoea, fatigue, infection, loss of appetite, nausea/vomiting, neuropathy, skin rash and stomatitis. When selecting patients with gastric and OGJ cancer for salvage systemic therapy, the expected benefit of the therapy needs to be weighed against the potential toxicities.

Why it is important to do this review

There are now multiple studies supporting the benefit of salvage systemic therapy to improve the survival of patients with advanced gastric and OGJ cancer. Pharmacological agents proven to be effective in this setting include both chemotherapy and newly available biological therapy. Studies comparing the efficacy of these agents vary in size, comparator interventions, and consequently, and consensus on the best salvage systemic therapy has not been reached. There also remain uncertainties about risk‐benefit trade‐off of salvage systemic therapy due to the expected short survival of these patients and potential toxicities associated with treatment.

Objectives

Primary objective

Secondary objectives

To assess the effect of the aforementioned intervention on tumour response, AEs and QoL.
To assess the impact of patients' geographical regions (East and South‐East Asia versus the rest of the world) on the survival benefit of salvage systemic therapy for advanced gastric and OGJ adenocarcinoma.
To compare the survival benefit of salvage systemic therapy for advanced gastric and OGJ adenocarcinoma in the second versus third‐line and beyond setting.

Methods

Criteria for considering studies for this review

Types of studies

We considered both blinded and open‐label randomised studies, in which a type of salvage systemic therapy is compared with either another type of salvage systemic therapy, placebo, BSC or no treatment. Quasi‐randomised studies were accepted as long as they were parallel‐group randomised studies. We included studies reported as full text or as abstract only as well as unpublished studies, if they met the other inclusion criteria.

Types of participants

We included adults patients over 18 years of age with a histological or cytological diagnosis of locally advanced and unresectable or metastatic gastric and OGJ adenocarcinoma, whose disease had progressed on first‐line fluoropyrimidine and platinum‐containing palliative chemotherapy. Patients with disease progression on, or disease relapse within 6 months of fluoropyrimidine and platinum‐containing neoadjuvant/perioperative/adjuvant chemotherapy, who were no longer suitable for potentially curative surgery, were also included in the review.

The original intention of this systemic review was to assess the effects of systemic therapy in patients with advanced gastric carcinoma, following progression on the standard first‐line chemotherapy. The decision to include OGJ adenocarcinoma in the review, however, was made based on the fact that the majority of clinical studies on advanced gastric cancer enrol patients with OGJ cancer and extracting the proposed primary endpoints for patients with gastric cancer alone was not always possible. OGJ cancer is commonly treated using the same approach as gastric cancer and distinguishing gastric cardia cancer involving OGJ from distal oesophagus and OGJ cancer extending inferiorly to involve gastric cardia remains controversial. In the most recent eighth edition of the American Joint Committee in Cancer (AJCC) staging system, the definition of cancer location for OGJ cancer has changed from the position of the upper edge of the cancer to its epicentre and some even suggest the genetic signature of OGJ cancers may be more accurate in determining the cell of origin for cancer staging rather than its gross location (Hayakawa 2016).

We excluded studies with the following characteristics:

studies in which patients have gastric or OGJ cancer other than adenocarcinoma in histology, unless these patients can be separated out for the purpose of outcome assessment;
studies in which patients have oesophageal cancer other than OGJ cancer, unless these patients can be separated out for the purpose of outcome assessment;
studies in which 10% or more of patients previously received chemotherapy which did not contain both fluoropyrimidine and platinum as first‐line therapy, unless these patients can be separated out for the purpose of outcome assessment; and
studies which do not mention the Eastern Cooperative Oncology Group (ECOG) performance score (PS), and those in which 10% or more of patients have an ECOG PS more than 2, unless these patients can be separated out for the purpose of outcome assessment.

When studies met the above listed exclusion criteria, we contacted investigators or study sponsors in order to obtain individual patient data for the purpose of primary and secondary outcome analysis and subgroup analysis where possible.

Types of interventions

We included studies comparing systemically administered (parenteral or oral) chemotherapy and biological therapy, either alone or in combination, with or without BSC, to another systemically administered therapy, placebo, BSC or no treatment.

We allowed the following co‐interventions in both experimental and control arms of studies, provided they were not part of the randomised treatment and administered for the purposes of symptom control alone:

surgery; and
radiotherapy.

Types of outcome measures

We assessed following outcomes.

Primary outcomes

Overall survival (OS), defined as survival time from the start of the intervention until death from any cause.
Progression‐free survival (PFS), defined as survival time without disease progression from the start of the intervention.
Serious adverse events (SAEs), assessed using Common Terminology Criteria for Adverse Events (CTCAE) and defined as Grade ≥ 3.

OS and PFS were chosen as the primary outcomes as they are considered to be the most clinically relevant efficacy assessment tools for patients and clinicians to decide on administration of the intervention. Disease progression for the purpose of determining OS and PFS was defined according to Response Evaluation Criteria in Solid Tumours (RECIST) or immune‐related Response Evaluation Criteria in Solid Tumours (irRECIST). irRECIST was developed to take account of pseudo‐progression, a feature unique to immunotherapy. It involves one‐dimensional measurement as in the case of RECIST, however, confirmation of disease progression is needed minimum four weeks after the first assessment indicating disease progression.

SAEs are also important in predicting the tolerability of the intervention and for its benefit‐risk assessment. CTCAE Grade 3 AEs generally refer to any type of AEs which require medical intervention and/or hospitalisation, but are not life‐threatening, while Grade 4 AEs are those which are life‐threatening and Grade 5 AEs are those resulting in death.

Secondary outcomes

Tumour response rate (TRR), assessed using RECIST or irRECIST.
Any adverse events (AAEs), assessed using CTCAE.
QoL, assessed using validated tools.

TRR was chosen for the secondary outcomes to assess the anticancer activity of the intervention. As patients included in this review generally had very limited prognoses, AAEs and QoL associated with the intervention were thought to be important aspects of their management.

We aimed to compare the following treatment groups for the primary and secondary outcomes.

Chemotherapy versus placebo, BSC or no treatment.
Biological therapy versus placebo, BSC or no treatment.
Chemotherapy combined with biological therapy versus placebo, BSC or no treatment.
Chemotherapy versus biological therapy.
Chemotherapy combined with biological therapy versus chemotherapy.
Chemotherapy combined with biological therapy versus biological therapy.
Chemotherapy versus chemotherapy.
Biological therapy versus biological therapy.
Chemotherapy combined with biological therapy versus chemotherapy combined with biological therapy.

Reporting of the outcomes listed here was not an inclusion criterion for the review.

Search methods for identification of studies

No restrictions were placed on the language of publication when searching the electronic databases or reviewing reference lists in identified studies. We translated abstracts of non‐English language papers and assessed them for potential inclusion in the review as necessary.

Electronic searches

We searched the following electronic databases:

Cochrane Central Register of Controlled Trials (CENTRAL) (inception to Issue 9, 2020) (Appendix 2) (which includes RCTs from Cochrane Gut Group Specialised Register);
EMBASE (1974 to October 2020) (via Ovid) (Appendix 3); and
MEDLINE (1946 to October 2020) (via Ovid) (Appendix 4).

We additionally searched:

WHO ICTRP (http://www.who.int/ictrp/en/) (Appendix 5); and
ClinicalTrials.gov (https://clinicaltrials.gov/) (Appendix 6).

Proceedings of the following oncology conferences were also handsearched (2000 to October 2020):

American Society of Clinical Oncology Annual Meeting;
American Society of Clinical Oncology Gastrointestinal Cancer Symposium; and
European Society of Medical Oncology Congress.

Searching other resources

We checked the reference lists of all relevant primary studies and review articles for additional references.

Data collection and analysis

Selection of studies

Two review authors independently screened for inclusion the titles and abstracts of all the studies we identified as a result of the search and coded them as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. We retrieved the full‐text study reports/publication and two review authors independently screened the full text and identified studies for inclusion. These authors also identified and recorded reasons for exclusion of the ineligible studies. We resolved any disagreement through discussion or, if required, we consulted a third person. We identified and excluded duplicates and collated multiple reports of the same study so that each study rather than each report was the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram (Figure 1) and Characteristics of excluded studies (Liberati 2009).

Data extraction and management

We used a standard data collection form for study characteristics and outcome data. One review author extracted study characteristics and outcome data from included studies. We extracted the following study characteristics.

Methods: study design, total duration of study and run in, number of study centres, study setting, withdrawals, and date of study.
Participants: number, mean age, age range, geographical location, gender, staging of disease, diagnostic criteria, inclusion criteria, exclusion criteria, and ECOG PS.
Interventions: intervention, comparison, concomitant medications, and excluded medications.
Outcomes: primary and secondary outcomes specified and collected, and time points reported.
Notes: funding for study and notable conflicts of interest of trial authors.

We noted in Characteristics of included studies if outcome data were not reported in a usable way. One review author copied the data from the data collection form into the Review Manager file. We double‐checked that the data were entered correctly by comparing the study reports against the data presented in the systematic review. A second review author spot‐checked study characteristics for accuracy against the trial report.

Assessment of risk of bias in included studies

Two review authors independently assessed risk of bias of each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Any disagreement was resolved by discussion or by involving a third assessor. We assessed the risk of bias according to the following domains.

Random sequence generation.
Allocation concealment.
Blinding of participants and personnel.
Blinding of outcome assessment.
Imbalance in baseline characteristics across treatment arms.
Incomplete outcome data.
Selective outcome reporting.
Other bias.

We graded each potential source of bias as high, low or unclear risk and provided a justification for our judgment in the 'Risk of bias' table. We summarised the 'Risk of bias' judgements across different studies for each of the domains listed (Figure 2). We considered attrition risk separately for QoL and the rest of the outcomes. Where information on risk of bias relates to unpublished data or correspondence with an author, we noted this in the 'Risk of bias' table.

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

When considering treatment effects, we took into account the risk of bias for the studies that contributed to that outcome.

Measures of treatment effect

We analysed time‐to‐event data (OS and PFS) as hazard ratios (HRs) with a 95% confidence intervals (CIs) and dichotomous data as risk ratios (RRs) with a 95% CIs. We planned to analyse continuous data as mean difference (MD) or standardised mean difference (SMD) with 95% CIs, ensuring that higher scores for continuous outcomes have the same meaning for the particular outcome, explaining the direction and reporting where the directions were reversed if this was necessary.

We undertook meta‐analysis only where this was meaningful, i.e. if the treatments, participants and underlying clinical questions were similar enough for pooling to make sense.

Where multiple study arms were reported in a single study and two comparisons (e.g. drug A versus placebo and drug B versus placebo) must be entered into the same meta‐analysis, we halved the control group to avoid double counting.

Unit of analysis issues

We included studies with a parallel group design where participants are individually assigned to one of the treatment groups with a single observation for each outcome from each participant. The unit of analysis therefore was individuals assigned to each group.

Dealing with missing data

We contacted investigators or study sponsors in order to verify key study characteristics and obtain missing numerical outcome data where possible (e.g. when a study was identified as an abstract only).

Assessment of heterogeneity

Heterogeneity secondary to clinical diversity was expected to be observed between the studies. This included variation in participants (age, ethnicity, baseline ECOG PS, volume of cancer, study eligibility), interventions (chemotherapy and/or biological therapy administered, intensity/dose, intervention in the control group) and outcomes (follow‐up duration), as well as variation in study methodology. We assessed heterogeneity among the studies in each analysis with visual inspection and statistically using the Chi‐square (Chi²) test and the I‐square (I²) statistic. We used a P value threshold of 0.10 to determine statistical significance for the Chi² test, and considered an I² of 30% or less to be low degree, 30% to 60% to be moderate degree, and 60% or more to be a high degree of heterogeneity. We recognised that there is uncertainty in the I² measurement when there are few studies in a meta‐analysis.

When we identified substantial heterogeneity, we explored possible sources of heterogeneity using the pre‐specified sensitivity and subgroup analysis described below.

Assessment of reporting biases

We attempted to contact study investigators asking them to provide missing outcome data. When this was not possible, and the missing data were thought to introduce serious bias, the impact of including such studies in the overall assessment of results was explored by a sensitivity analysis.

As we were unable to pool more than ten studies for any of the meta‐analyses performed in this review, planned assessment using funnel plots to explore possible publication biases was not carried out.

Data synthesis

We used Review Manager 5.4.1 (Review Manager 2020) for pooling of data at a study level, and statistical analysis. Given the potential heterogeneity in the included studies as described above, we used a random‐effects model for both primary and secondary outcome analysis and subgroup analysis.

'Summary of findings' tables

We created 'Summary of findings' tables using the following outcomes: OS, PFS, SAEs, TRR, AAEs and QoL. We used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the quality of a body of evidence as it relates to the studies which contribute data to the meta‐analysis for the pre‐specified outcomes. We used methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook (Higgins 2011) and used GRADEpro software. We justified all decisions to down‐ or up‐grade the quality of studies using footnotes, and made comments to aid readers' understanding of the review where necessary. We considered whether there is any additional outcome information that was not able to be incorporated into meta‐analysis, and noted this in the comments and stated if it supports or contradicts the information from the meta‐analysis.

Subgroup analysis and investigation of heterogeneity

We carried out the following subgroup analyses.

Line of treatment (second line versus third line and beyond).
Geographical region (East and South‐East Asia versus the rest of the world).
Extent of the disease (locally advanced versus metastatic).
Type of biological therapy (non‐immunotherapy versus immunotherapy).

Geographical region rather than ethnicity was chosen as this additionally accounts for the regional difference in the preferred first‐line systemic therapy for gastric and OGJ cancer, and also to maximise the data attainment.

The following outcomes were used in subgroup analysis:

OS; and
PFS.

We used the formal statistical test, Chi² test, with significance set at 10%, to test for subgroup interactions.

We aimed to compare the following treatment groups for the subgroup analysis.

Chemotherapy versus placebo, BSC or no treatment.
Biological therapy versus placebo, BSC or no treatment.
Chemotherapy combined with biological therapy versus placebo, BSC or no treatment.
Chemotherapy versus biological therapy.
Chemotherapy combined with biological therapy versus chemotherapy.
Chemotherapy combined with biological therapy versus biological therapy.

Sensitivity analysis

We performed sensitivity analysis to assess the effect of risk of bias and heterogeneity in the included studies. We repeated the OS and PFS analysis with the following conditions.

Use of a fixed‐effect model.
Exclusion of studies with high risk or unclear risk of bias for randomisation, blinding or attrition.
Exclusion of studies with incomplete reporting of primary outcome (HR for OS and/or PFS estimated from the test or acquired directly from investigators).
Exclusion of studies evaluating HER2‐targeted therapy in biomarker selected participants.

If any additional clinical or methodological variations suitable for sensitivity analysis arose during the review process, we considered this.

Reaching conclusions

We based our conclusions only on findings from the quantitative or narrative synthesis of included studies for this review. We avoided making specific recommendations for practice and our implications for research gave the reader a clear sense of where the focus of any future research in the area should be and what the remaining uncertainties are.

We conducted the review according to the published protocol and reported any deviations from it in the 'Differences between protocol and review' section of the systematic review.

Results

Description of studies

We categorised the relevant studies into four categories and their characteristics were summarised see Characteristics of included studies, Characteristics of excluded studies, Characteristics of studies awaiting classification and Characteristics of ongoing studies.

Results of the search

Through the electronic search in October 2020, we identified 9099 unique records; 3024 from EMBASE, 375 from CENTRAL, 1934 from MEDLINE, 3294 from WHO ICTRP and 472 from ClinicalTrials.gov. Handsearching found no relevant abstracts from conference proceedings. Of 9093 records left after removing duplicates, 8999 records were considered not suitable for further assessment; 8509 records were excluded after screening title and abstract and 442 records were excluded as they were related records for single trials. Tweenty‐nine studies are ongoing and nineteen studies are awaiting classification as there were no publication on results available (n = 7) or their abstracts provided insufficient information for screening in the absence of full‐text articles (n = 12). One study (Zhao 2019) identified in the top‐up search performed immediately prior to publication is not yet fully incorporated into the review and is added to studies awaiting classifications; it will be incorporated into the review at the next update. Ninety‐four published abstracts or full‐text articles were assessed in detail and seventy‐six of them were considered to have failed screening based on non‐RCT (n = 9), included participants with oesophageal cancer other than oesophago‐gastric junction (OGJ) adenocarcinoma (n = 4), included participants who received intervention as first‐line systemic therapy (n = 32), equal or more than 10%, or unknown proportion of participants received non‐fluoropyrimidine and platinum‐containing chemotherapy as the first‐line (n = 29), participants had stable disease on first‐line fluoropyrimidine and platinum‐containing chemotherapy (n = 1) and eligibility for inclusion could not be assessed on the published full‐text article (n = 1). Details of search results are shown in the PRISMA flow diagram (Figure 1).

Included studies

Fourteen studies (Bang 2017; Hironaka 2013; Kim 2015; Lee 2017; Ling 2018; Makiyama 2018; Pauligk 2017; Shah 2018; Shitara 2018; Shitara 2018b; Thuss‐Patience 2011; Thuss‐Patience 2017; Yi 2012; Wilke 2014) were considered to have satisfied our selection criteria and included into this review. Three studies (Kang 2017; Li 2013; Pavlakis 2016) did not provide sufficient information for screening whether more than 90% of participants from these studies received and progressed on fluoropyrimidine and platinum‐containing chemotherapy, however, whether these two agents were administered concurrently was not clear in the texts. Although no response was obtained from study authors, these studies were considered highly relevant and included in the review. A total of 17 studies with 5110 participants were considered in this review.

Study design

Eight studies had double‐blind placebo‐controlled design (Bang 2017; Kang 2017; Li 2013; Pauligk 2017; Pavlakis 2016; Shah 2018; Shitara 2018b; Wilke 2014), while another six studies had open‐label design (Hironaka 2013; Makiyama 2018; Shitara 2018; Thuss‐Patience 2011; Thuss‐Patience 2017; Yi 2012). Three studies (Kim 2015; Lee 2017; Ling 2018) did not mention blinding status, however, were most likely open‐label trials. Seven studies (Bang 2017; Kang 2017; Pavlakis 2016; Shah 2018; Shitara 2018; Shitara 2018b; Thuss‐Patience 2017) were multinational, eight studies (Hironaka 2013; Kim 2015; Lee 2017; Li 2013; Makiyama 2018; Pauligk 2017; Thuss‐Patience 2011; Yi 2012) were multi‐centre trials in a single country, and two studies (Ling 2018; Yi 2012) were single‐centre trials. Nine studies (Bang 2017; Hironaka 2013; Kang 2017; Pauligk 2017; Shah 2018; Shitara 2018; Shitara 2018b; Thuss‐Patience 2011; Wilke 2014) were phase 3, six studies (Kim 2015; Lee 2017; Li 2013; Makiyama 2018; Pavlakis 2016; Yi 2012) were phase 2, and one study (Thuss‐Patience 2017) was phase 2/3. One study (Ling 2018) did not describe the phase.

Participants

Reported median age of participants varied from 52 to 65 years with the youngest and oldest participant being 19 and 85 years old, respectively. All the studies when data were available, had more male than female participants with the proportion of male participants ranging from 65.4% to 80.8%. In total, 3379 out of 4721 participants in this review were estimated to be males with two studies (Makiyama 2018; Pauligk 2017) not providing any information on the gender of participants. The number of participants with ECOG PS 2 was low in the majority of included studies; seven studies (Bang 2017; Kang 2017; Li 2013; Pavlakis 2016; Shah 2018; Shitara 2018b; Wilke 2014) only enrolled participants with ECOG PS 0‐1 and in five studies (Hironaka 2013; Kim 2015; Lee 2017; Shitara 2018; Thuss‐Patience 2017), the proportion of participants with ECOG PS 2 was less than 6.0%. In three studies (Ling 2018; Thuss‐Patience 2011: Yi 2012) more than 10% of the study population had ECOG PS 2 with the highest being 24.3%.

Four studies (Hironaka 2013; Kim 2015; Lee 2017; Ling 2018) included participants with only gastric cancer. For nine studies (Bang 2017; Kang 2017; Pavlakis 2016; Shah 2018; Shitara 2018; Shitara 2018b; Thuss‐Patience 2011; Thuss‐Patience 2017; Wilke 2014), which reported the number of participants with gastric and OGJ adenocarcinoma, 57.5% to 97.0% of the study participants had gastric cancer. The remaining four studies (Li 2013; Makiyama 2018; Pauligk 2017; Yi 2012) enrolled participants with both gastric and OGJ adenocarcinoma, however, the number of participants with each condition was not provided.

Five studies (Hironaka 2013; Kim 2015; Lee 2017; Ling 2018; Thuss‐Patience 2011) included participants with metastatic or recurrent disease, while the other 12 studies (Bang 2017; Kang 2017; Li 2013; Makiyama 2018; Pauligk 2017; Pavlakis 2016; Shah 2018; Shitara 2018; Shitara 2018b; Thuss‐Patience 2017; Wilke 2014; Yi 2012) included participants with locally advanced unresectable disease and those with metastatic or recurrent disease. When known, the proportion of participants with metastatic disease ranged between 57.0% to 100.0%. Participants with recurrent disease were evaluated by seven studies (Hironaka 2013; Kang 2017; Kim 2015; Lee 2017; Makiyama 2018; Pauligk 2017; Pavlakis 2016) .

Nine studies (Bang 2017; Hironaka 2013; Kang 2017; Li 2013; Shitara 2018; Shitara 2018b; Thuss‐Patience 2011; Thuss‐Patience 2017; Wilke 2014) reported the number of participants who had previous gastrectomy or resection of primary tumours and the proportion such participants varied between 30.0% to 76.6%. Kim 2015 reported the proportion of participants who had received palliative operation to be 46.2%, however, it did not describe the type of surgery performed.

Study treatment was administered as second‐line systemic therapy in 12 studies (Bang 2017; Hironaka 2013; Kim 2015; Lee 2017; Ling 2018; Makiyama 2018; Shah 2018; Shitara 2018; Thuss‐Patience 2011; Thuss‐Patience 2017; Wilke 2014; Yi 2012) with 3522 participants, second‐ or third‐line in one study (Pavlakis 2016) with 147 participants, second‐ to fourth‐line in one study (Pauligk 2017) with 300 participants, and third‐line and beyond in three studies (Kang 2017; Li 2013; Shitara 2018b) with 1141 participants. Six studies (Pavlakis 2016; Shah 2018; Shitara 2018; Shitara 2018b; Thuss‐Patience 2017; Wilke 2014) were multinational trials with participating centres located worldwide, nine studies (Bang 2017; Hironaka 2013; Kang 2017; Kim 2015; Lee 2017; Li 2013; Ling 2018; Makiyama 2018; Yi 2012) recruited participants from Asian countries and two studies from Germany (Pauligk 2017; Thuss‐Patience 2011). In total, 2327 participants from 13 studies (Bang 2017; Hironaka 2013; Kang 2017; Kim 2015; Lee 2017; Li 2013; Ling 2018; Makiyama 2018; Pavlakis 2016; Shitara 2018; Shitara 2018b; Wilke 2014; Yi 2012) were enrolled from Asian countries with additional 157 participants in one study (Thuss‐Patience 2017) being enrolled from countries in the Asia‐Pacific region. The number participants enrolled from Asian countries was unknown for one study (Shah 2018), which was a multinational trial.

Three studies selected participants based on biomarker expression. Only participants with HER2 positivity were enrolled in two studies (Makiyama 2018; Thuss‐Patience 2017). In the third study (Shitara 2018), the study's predefined primary outcomes were overall survival (OS) and progression‐free survival (PFS) in PD‐L1 combined positive score ≥ 1 participants and the enrolment was restricted to participants with PD‐L1 combined positive score of ≥ 1 after 489 out of 592 participants were enrolled.

Interventions

Taxanes were the most commonly investigated single‐agent chemotherapy: paclitaxel was examined in six studies (Bang 2017; Hironaka 2013; Makiyama 2018; Pauligk 2017; Shah 2018; Wilke 2014), docetaxel in three studies (Kim 2015; Lee 2017; Yi 2012) and either paclitaxel or docetaxel in one study (Thuss‐Patience 2017). Irinotecan was assessed in two studies (Hironaka 2013; Thuss‐Patience 2011) and Trifluridine/tipiracil in one study (Shitara 2018b). Combination chemotherapy was compared with either single‐agent chemotherapy or another combination chemotherapy in five studies: docetaxel‐oxaliplatin (Kim 2015), docetaxel‐cisplatin (Lee 2017), docetaxel‐S1 (Lee 2017), EOX (Ling 2018) and FOLFIRI (Ling 2018). Five studies examined single‐agent biological therapy: nivolumab (Kang 2017), apatinib (Li 2013), regorafenib (Pavlakis 2016), pembrolizumab (Shitara 2018) and trastuzumab emtansine (Thuss‐Patience 2017). Alternatively, biological therapy was examined as an addition to taxane chemotherapy:olaparib‐paclitaxel (Bang 2017), trastuzumab‐paclitaxel (Makiyama 2018), everolimus‐paclitaxel (Pauligk 2017), napabucasin‐paclitaxel (Shah 2018), ramucirumab‐paclitaxel (Wilke 2014) and sunitinib‐docetaxel (Yi 2012).

Outcome measures

Outcomes on efficacy: overall survival (OS), progression‐free survival (PFS) and tumour response rate (TRR).

World Health Organization (WHO) criteria were used for efficacy assessment in one study (Thuss‐Patience 2011) and the method was not described in two studies published as conference abstracts (Makiyama 2018; Pauligk 2017). WHO criteria have been previously shown to be comparable to Response Evaluation Criteria in Solid Tumours (RECIST) criteria in evaluating the response of colorectal carcinoma (Choi 2005). In the remaining studies RECIST or immune‐related Response Evaluation Criteria in Solid Tumours (irRECIST) was used to assess for radiological disease progression. OS and PFS were measured in all the studies included in this review, except in Yi 2012 where time to progression (TTP) was reported in place of PFS. Three studies (Makiyama 2018; Pauligk 2017; Shah 2018) did not specify if efficacy analysis was performed on the intention‐to‐treat population. As these studies specified radiologically measurable or evaluable disease as one of their inclusion criteria, we assumed efficacy analysis was performed on the intention‐to‐treat population. One study (Thuss‐Patience 2011) reported PFS and TRR on participants in the irinotecan arm and not the best supportive care (BSC) arm. Two studies (Bang 2017; Pauligk 2017) reported hazard ratios (HRs) for survival outcomes with 97.5% confidence intervals (CIs and P values; 95% CIs were estimated under the assumption for Gaussian distribution. Three studies (Kim 2015; Lee 2017; Ling 2018) reported survival outcomes without referring to HRs and they were estimated from the published Kaplan‐Meire curves for two of the studies (Kim 2015; Lee 2017). Kaplan‐Meier curves from Ling 2018 indicated median overall survival (mOS) and median progression‐free survival (mPFS), which differed from those described in its text, raising a question on the accuracy of these Kaplan‐Meier curves and the estimated HRs from them. TRR was assessed on subpopulation of randomised participants, those with measurable disease at baseline in three studies (Hironaka 2013; Kang 2017; Thuss‐Patience 2017). One study (Pauligk 2017) did not report TRR and was excluded from meta‐analysis for this outcome. 4806, 4766 and 4328 participants were considered to have been included in OS, PFS and TRR analyses, respectively.

Outcomes on adverse events: serious adverse events (SAEs) and adverse events (AEs).

Apart from three studies (Makiyama 2018; Pauligk 2017; Shah 2018), which did not specify the assessment method for safety, National Cancer Institute : Common Terminology Criteria for Adverse Events (NCI CTCAE) version 2.0‐4.03 was used for measuring AEs and SAEs. Minimal descriptions were provided on AEs for three studies (Makiyama 2018; Pauligk 2017; Shah 2018).Two studies (Pavlakis 2016; Thuss‐Patience 2011) only reported data on SAEs alone with the second study only measuring AEs in the irinotecan arm and not the BSC arm. For the remaining studies, comprehensive results on AEs were available.

Outcomes on quality of life (QoL)

Seven studies listed patient‐reported QoL as their outcomes and assessed QoL using one or combinations of three questionnaires: European Organisation for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire C30 (QLQ‐C30) in all the studies (Bang 2017; Lee 2017; Li 2013; Pavlakis 2016; Thuss‐Patience 2011; Thuss‐Patience 2017; Wilke 2014), EORTC gastric module (QLQ‐STO22) in four studies (Bang 2017; Lee 2017; Pavlakis 2016; Thuss‐Patience 2017), and the EuroQol five dimensions health status questionnaire (EQ‐5D) in two studies (Thuss‐Patience 2017; Wilke 2014). Two studies (Bang 2017; Pauligk 2017) resulted in separate publications focusing on the QoL outcomes. QoL outcomes of another study (Shitara 2018b) were similarly going to be published separately, however, this was not available at the time of preparing this review.

Excluded studies

The two commonest reasons for studies being excluded after the assessment of abstracts or full‐text articles were enrolment of participants who received intervention as first‐line systemic therapy (n = 32) and equal or more than 10%, or unknown proportion of participants having received non‐fluoropyrimidine and platinum‐containing chemotherapy as first‐line therapy (n = 29). Study authors were contacted via email when insufficient data were available for screening, however, only a few responses were received and in those occasions study data were owned by sponsors and could not be released.

Risk of bias in included studies

Potential source of bias in included studies are described in Characteristics of included studies and summarised in 'Risk of bias' summary table (Figure 2).

Allocation

Randomisation method was not described for four studies: one full‐text article (Kim 2015) and three conference abstracts (Makiyama 2018; Pauligk 2017; Shah 2018). In an additional four studies (Lee 2017; Li 2013; Ling 2018; Yi 2012), authors only reported random sequence method or allocation concealment method, but not both.

Blinding

Eight studies (Bang 2017; Kang 2017; Li 2013; Pauligk 2017; Pavlakis 2016; Shah 2018; Shitara 2018b; Wilke 2014) had double‐blind placebo‐controlled design and were considered to be at low risk of performance and detection bias, except for Shah 2018 where the data safety monitoring board recommended unblinding after an interim analysis suggested meeting the primary endpoint at final analyses was unlikely. In three of the six open‐label studies included in the review (Hironaka 2013; Shitara 2018; Yi 2012) response assessment was performed centrally by masked reviewers, therefore these studies were thought to be at high risk of performance bias, but not detection bias. Two open‐label studies (Thuss‐Patience 2011; Thuss‐Patience 2017) were considered to be at high risk of both performance and detection bias. No blinding status for outcome assessment was provided for the remaining one study (Makiyama 2018).

Imbalance in baseline characteristics (confounding)

Twelve studies (Bang 2017; Hironaka 2013; Kang 2017; Li 2013; Ling 2018; Pavlakis 2016; Shitara 2018; Shitara 2018b; Thuss‐Patience 2011; Thuss‐Patience 2017; Wilke 2014; Yi 2012) were assessed to be at low risk of confounding based on the description of stratification at randomisation and well‐balanced patient and disease characteristics at baseline. In two studies (Kim 2015; Lee 2017) the risk of confounding was unclear due to some imbalances in patient and disease characteristics at baseline. Detailed study description was not available for Makiyama 2018, Pauligk 2017, and Shah 2018, which were published as conference abstracts, to adequately assess their risk of confounding.

Efficacy and safety outcomes

All the studies performed the survival analysis on the intention‐to‐treat population and assessed to be at low risk of attrition bias, except for the three conference abstracts (Makiyama 2018; Pauligk 2017; Shah 2018) where details of analysis performed was unknown. Five studies (Hironaka 2013; Lee 2017; Li 2013; Pavlakis 2016; Yi 2012) modified their intention‐to‐treat population by removing participants who were found to be ineligible for the study after randomisation, withdrew the consent and/or did not receive any study treatment, however, in these studies the intention‐to‐treat population still constituted more than 95% of the total randomised population and considered to be at low risk of attrition bias.

Quality of life (QoL) outcomes

QoL endpoints were measured in seven studies. 84.2% of participants completed QoL questionnaires at least one post‐baseline time point in Wilke 2014, and this study was assessed to be low risk of attrition bias. For Bang 2017 and Pavlakis 2016, they were rated 65.9% and 62.6%, respectively and these studies were considered to be at high risk of attrition bias. For Lee 2017, more than 60% of participants completed baseline QoL questionnaires and for Thuss‐Patience 2011 the completion rate of the QoL questionnaire was said to be "poor". Therefore, these studies were similarly considered to be at high risk of attrition bias. For the remaining two studies (Li 2013; Thuss‐Patience 2017), detailed results including the number of participants completing the relevant questionnaires were not provided. For Shitara 2018b, in which QoL outcome is yet to be published, and for the studies which did not measure QoL outcome, their attrition bias risk was categorised as low in 'Risk of bias' summary table (Figure 2).

Selective reporting

One study (Thuss‐Patience 2011) reported OS for both treatment arms, but not PFS, TRR and AE outcomes, which were provided for the active treatment arm alone and not the BSC arm. In the same study, QoL was not reported due to poor completion rate of EORTC QLQ‐C30 questionnaire by the participants, which authors explained limited any meaningful analysis. Detailed results on AEs were not given for the two conference abstracts (Pauligk 2017; Makiyama 2018), however, all the predefined efficacy outcomes were reported in these studies. The other 14 studies in the review provided results for both AAEs and SAEs, except for Pavlakis 2016 and Thuss‐Patience 2011, which only provided the result for SAEs. Four studies (Bang 2017; Lee 2017; Li 2013; Thuss‐Patience 2017) provided variable amount of results on QoL in a descriptive manner. In two studies (Pavlakis 2016; Wilke 2014), separate articles focusing on QoL outcomes alone were published.

At least a brief summary of their protocols was available through relevant clinical trial registries for all the studies except for three studies (Kim 2015; Ling 2018; Thuss‐Patience 2011). All the efficacy outcomes relevant to this review (predefined in the case of studies with protocols available) were reported in all the studies apart from Thuss‐Patience 2011. Considering the order of importance of different outcomes based on the purpose of this review to be efficacy outcomes > safety outcomes > QoL outcomes, one study (Thuss‐Patience 2011) was assessed to be at high risk of reporting bias, three studies (Makiyama 2018; Pauligk 2017; Shah 2018) to be unclear and the rest of studies to be at low risk of reporting bias.

Other potential sources of bias

Three studies (Kim 2015; Lee 2017; Thuss‐Patience 2011) were terminated prematurely due to poor accrual, making these studies at high risk of recruitment bias. Enrolment was restricted to a biomarker‐selected population after 82.6% of the study participants had enrolled in Shitara 2018, however, this did not affect the study's primary outcome results and hence this study was considered to be still low risk of recruitment bias. Other sources of bias could not be adequately assessed in three studies (Makiyama 2018; Pauligk 2017; Shah 2018) published as conference abstracts.

Effects of interventions

See: Table 1; Table 2; Table 3; Table 4; Table 5

Summary of findings 1. Chemotherapy versus placebo, best supportive care (BSC) or no treatment.

Chemotherapy compared to placebo, BSC or no treatment for advanced gastric and oesophago‐gastric junction adenocarcinoma
Patient or population: advanced gastric and oesophago‐gastric junction adenocarcinoma Setting: second‐line and beyond Intervention: chemotherapy Comparison: placebo, BSC or no treatment
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with chemotherapy	Risk with placebo, BSC or no treatment
Overall survival	mOS weighed for study size		HR 0.66 (0.52 to 0.83)	547 (2 RCTs)	⊕⊕⊕⊝ MODERATE ¹	2 studies compared chemotherapy (irinotecan and trifluridine/tipiracil) to placebo or BSC. Heterogeneity was low (I² = 7%, P = 0.30).
	5.6 months	3.5 months
Progression‐free survival	mPFS		HR 0.57 (0.47 to 0.69)	507 (1 RCT)	⊕⊕⊕⊕ HIGH	1 study compared trifluridine/tipiracil to placebo.
	2.0 months	1.8 months
Serious adverse events	Study population		RR 1.38 (1.20 to 1.59)	503 (1 RCT)	⊕⊕⊕⊝ MODERATE ²	1 study compared trifluridine/tipiracil to placebo.
	797 per 1,000 (693 to 918)	577 per 1,000
Tumour response rate			RR 2.17 (0.63 to 7.48)	435 (1 RCT)	⊕⊕⊝⊝ LOW ³	1 study compared trifluridine/tipiracil to placebo.
	97 per 1,000 (28 to 335)	45 per 1,000
Any adverse events			RR 1.04 (1.00 to 1.09)	503 (1 RCT)	⊕⊕⊕⊕ HIGH	1 study compared trifluridine/tipiracil to placebo.
	972 per 1,000 (935 to 1,000)	935 per 1,000
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval ;HR: hazard ratio; mOS: median overall survival; mPFS: median progression‐free survival; RCT: randomised controlled trial; RR: Risk ratio.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

Biological therapy compared to placebo, BSC or no treatment for advanced gastric and oesophago‐gastric junction adenocarcinoma
Patient or population: advanced gastric and oesophago‐gastric junction adenocarcinoma Setting: second‐ and third‐line Intervention: biological therapy Comparison: placebo, BSC or no treatment
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with biological therapy	Risk with placebo, BSC or no treatment
Overall survival	mOS weighed for study size		HR 0.55 (0.41 to 0.73)	781 (3 RCTs)	⊕⊕⊕⊕ HIGH	1 study compared nivolumab to placebo and 2 studies compared VEGFR‐targeted agents (apatinib and regorafenib) to placebo. Heterogeneity was moderate (I² = 55%, P = 0.08).
	5.2 months	3.9 months
Progression‐free survival	mPFS weighed for study size		HR 0.33 (0.19 to 0.57)	781 (3 RCTs)	⊕⊕⊕⊝ MODERATE ¹	1 study compared nivolumab to placebo and 2 studies compared VEGFR‐targeted agents (apatinib and regorafenib) to placebo. Heterogeneity was high (I² = 87%, P < 0.0001).
	2.1 months	1.3 months
Serious adverse events	Study population		RR 1.14 (0.95 to 1.37)	638 (2 RCTs)	⊕⊕⊝⊝ LOW ²	1 study compared nivolumab to placebo and the other study compared regorafenib to placebo. Heterogeneity was low (I² = 3%, P = 0.31).
	481 per 1,000 (401 to 578)	422 per 1,000
Tumour response rate	Study population		RR 5.12 (1.23 to 21.27)	687 (3 RCTs)	⊕⊕⊝⊝ LOW ³	1 study compared nivolumab to placebo and 2 studies compared VEGFR‐targeted agents (apatinib and regorafenib) to placebo. Heterogeneity was low (I² = 13%, P = 0.33).
	470 per 1,000 (113 to 1,000)	92 per 1,000
Any adverse events	Study population		RR 1.08 (1.00 to 1.17)	491 (1 RCT)	⊕⊕⊕⊕ HIGH	1 study compared nivolumab to placebo.
	906 per 1,000 (839 to 981)	839 per 1,000
Quality of life (QoL	)Similar QoL was experienced by participants treated with biological therapy and placebo, except for improved insomnia and pain associated with apatinib and regorafenib treatment, respectively. More participants receiving regorafenib experienced diarrhoea, and sore throat and mouth.			283 (2 RCTs)	⊕⊕⊝⊝ LOW ^{4 5}	QoL was assessed in 2 studies comparing VEGFR‐targeted agents (apatinib and regorafenib) to placebo, using EORTC QLQ‐C30 +/‐ QLQ‐STO22 and EQ‐5D.
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; HR: hazard ratio; mOS: median overall survival; mPFS: median progression‐free survival; RCT: randomised controlled trial; RR: Risk ratio.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

Chemotherapy compared to biological therapy for advanced gastric and oesophago‐gastric junction adenocarcinoma
Patient or population: advanced gastric and oesophago‐gastric junction adenocarcinoma Setting: second‐line Intervention: biological therapy Comparison: chemotherapy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with biological therapy	Risk with Chemotherapy
Overall survival	mOS		HR 0.82 (0.66 to 1.02)	395 (1 RCT)	⊕⊕⊕⊝ MODERATE ¹	1 study compared pembrolizumab to paclitaxel in PD‐L1 CPS ≥ 1 population.
	9.1 months	8.3 months
Progression‐free survival	mPFS		HR 1.27 (1.03 to 1.57)	395 (1 RCT)	⊕⊕⊕⊝ MODERATE ²	1 study compared pembrolizumab to paclitaxel in PD‐L1 CPS ≥ 1 population.
	1.5 months	4.1 months
Serious adverse events	Study population		RR 0.41 (0.30 to 0.57)	570 (1 RCT)	⊕⊕⊝⊝ LOW ^{3 4}	1 study compared pembrolizumab to paclitaxel.
	143 per 1,000 (104 to 198)	348 per 1,000
Tumour response rate	Study population		RR 1.17 (0.72 to 1.88)	395 (1 RCT)	⊕⊕⊝⊝ LOW ⁵	1 study compared pembrolizumab to paclitaxel in PD‐L1 CPS ≥ 1 population.
	159 per 1,000 (98 to 255)	136 per 1,000
Any adverse events	Study population		RR 0.63 (0.56 to 0.71)	570 (1 RCT)	⊕⊕⊕⊝ MODERATE ⁴	1 study compared pembrolizumab to paclitaxel.
	530 per 1,000 (471 to 597)	841 per 1,000
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; HR: hazard ratio; mOS: median overall survival; mPFS: median progression‐free survival; RCT: randomised controlled trial;RR: Risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Chemmotherapy compared to chemotherapy combined with biological therapy for advanced gastric and oesophago‐gastric junction adenocarcinoma
Patient or population: advanced gastric and oesophago‐gastric junction adenocarcinoma Setting: second‐ to fourth‐line Intervention: chemotherapy combined with biological therapy Comparison: chemotherapy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with chemotherapy combined with biological therapy	Risk with chemotherapy
Overall survival	mOS weighed for study size		HR 0.93 (0.83 to 1.04)	2743 (7 RCTs)	⊕⊕⊕⊝ MODERATE ¹	1 study each examined PARP inhibitor (olaparib), anti‐HER2 antibody (trastuzumab), cancer stem cell inhibitor (napabucasin), mTOR inhibitor (everolimus), anti‐VEGFR2 antibody (ramucirumab) and multi‐tyrosine kinase inhibitor (sunitinib) together with chemotherapy. The other study examined anti‐HER2 antibody conjugated with chemotherapy (trastuzumab‐emtansine). The control arm was taxane in all studies. Heterogeneity was moderate (I² = 36%, P = 0.15).
	8.1 months	7.2 months
Progression‐free survival	mPFS weighed for study size		HR 0.87 (0.74 to 1.02)	2743 (7 RCTs)	⊕⊝⊝⊝ VERY LOW ^{2 3 4}	1 study each examined PARP inhibitor (olaparib), anti‐HER2 antibody (trastuzumab), cancer stem cell inhibitor (napabucasin), mTOR inhibitor (everolimus), anti‐VEGFR2 antibody (ramucirumab) and multi‐tyrosine kinase inhibitor (sunitinib) together with chemotherapy. The other study examined anti‐HER2 antibody conjugated with chemotherapy (trastuzumab‐emtansine). The control arm was taxane in all studies. Heterogeneity was high (I² = 71%, P = 0.002).
	3.9 months	3.1 months
Serious adverse events	Study population		RR 1.17 (0.95 to 1.44)	1618 (4 RCTs)	⊕⊝⊝⊝ VERY LOW ^{3 5 6}	1 study each examined PARP inhibitor (olaparib), anti‐VEGFR2 antibody (ramucirumab) and multi‐tyrosine kinase inhibitor (sunitinib) together with chemotherapy. The other study examined anti‐HER2 antibody conjugated with chemotherapy (trastuzumab‐emtansine). The control arm was taxane in all studies. Heterogeneity was high (I² = 86%, P < 0.0001).
	735 per 1,000 (686 to 791)	613 per 1,000
Tumour response rate	Study population		RR 1.35 (0.99 to 1.85)	2404 (6 RCTs)	⊕⊝⊝⊝ VERY LOW ^{2 3 4}	1 study each examined PARP inhibitor (olaparib), anti‐HER2 antibody (trastuzumab), cancer stem cell inhibitor (napabucasin), anti‐VEGFR2 antibody (ramucirumab) and multi‐tyrosine kinase inhibitor (sunitinib) together with chemotherapy. The other study examined anti‐HER2 antibody conjugated with chemotherapy (trastuzumab‐emtansine). The control arm was taxane in all studies. Heterogeneity was high (I² = 66%, P = 0.01).
	294 per 1,000 (216 to 403)	218 per 1,000
Any adverse events	Study population		RR 1.01 (1.00 to 1.03)	1513 (3 RCTs)	⊕⊕⊕⊝ MODERATE ⁷	1 study each examined PARP inhibitor (olaparib) and anti‐VEGFR2 antibody (ramucirumab). The other study examined anti‐HER2 antibody conjugated with chemotherapy (trastuzumab‐emtansine). The control arm was taxane in all studies. Heterogeneity was low (I² = 0%, P = 0.78).
	987 per 1,000 (977 to 1000)	977 per 1,000
Quality of life (QoL)	Median time to deterioration of EORTC QLQ‐C30 global health status for olaparib‐paclitaxel vs placebo‐paclitaxel		HR 0.88 (0.74 to 1.04)	1154 (2 RCTs)	⊕⊕⊝⊝ LOW ^{4 8}	QoL was assessed by 2 out of 7 studies using EORTC QLQ‐C30 plus QLQ‐SOT22 or EQ‐5D‐3L. 1 study each investigated PARP inhibitor (olaparib) and anti‐VEGFR2 antibody (ramucirumab). The control arm was taxane in both studies. Meta‐analysis was performed for time to deterioration of EORTC QLQ‐C30 global health status. Heterogeneity was very low (I² = 0%, P = 0.47).
	3.4 months	2.4 months
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; HR: hazard ratio; mOS: median overall survival; mPFS: median progression‐free survival; RCT: randomised controlled trial; RR: Risk ratio.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

Chemotherapy compared to chemotherapy for advanced gastric and oesophago‐gastric junction adenocarcinoma
Patient or population: advanced gastric adenocarcinoma Setting: second‐line Intervention: chemotherapy Comparison: chemotherapy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with chemotherapy	Risk with chemotherapy
Overall survival (non‐taxane monotherapy vs taxane monotherapy)	mOS		HR 1.13 (0.86 to 1.48)	219 (1 RCT)	⊕⊕⊝⊝ LOW ¹	1 study compared irinotecan to paclitaxel.
	8.4 months (irinotecan)	9.5 months (paclitaxel)
Overall survival (taxane‐containing doublet therapy vs taxane monotherapy)	mOS weighed for study size		HR 1.05 (0.72 to 1.54)	121 (2 RCTs)	⊕⊕⊝⊝ LOW ²	1 study compared docetaxel‐oxaliplatin to docetaxel and the other study compared docetaxel‐cisplatin and docetaxel‐S1 to docetaxel. Heterogeneity was moderate (I² = 44%, P = 0.17).
	7.0 months (taxane‐containing doublet therapy)	8.8 months (docetaxel)
Overall survival (non‐taxane containing therapy vs another non‐taxane containing therapy)	mOS (P = 0.17)		Not applicable	107 (1 RCT)	⊕⊕⊕⊕ HIGH	1 study compared FOLFIRI to EOX.
	18.5 months (FOLFIRI)	19.3 months (EOX)
Progression‐free survival (non‐taxane monotherapy vs taxane monotherapy)	mPFS		HR 1.14 (0.88 to 1.48)	219 (1 RCT)	⊕⊕⊝⊝ LOW ¹	1 study compared irinotecan to paclitaxel.
	2.3 months (irinotecan)	3.6 months (paclitaxel)
Progression‐free survival (taxane‐containing doublet therapy vs taxane monotherapy)	mPFS weighed for study size		HR 0.75 (0.52 to 1.09)	121 (2 RCTs)	⊕⊕⊝⊝ LOW ³	1 study compared docetaxel‐oxaliplatin to docetaxel and the other study compared docetaxel‐cisplatin and docetaxel‐S1 to docetaxel. Heterogeneity was moderate (I² = 59%, P = 0.09).
	3.4 months (taxane‐containing doublet therapy)	1.6 months (docetaxel)
Progression‐free survival (non‐taxane containing therapy vs another non‐taxane containing therapy)	mPFS (P = 0.46)		Not applicable	107 (1 RCT)	⊕⊕⊕⊕ HIGH	1 study compared FOLFIRI to EOX.
	8.1 months (FOLFIRI)	7.4 months (EOX)
Serious adverse effects	Grade ≥ 3 neutropenia was common occurring in > 20% of participants receiving both mono chemotherapy and polychemotherapy except for docetaxel‐S1 and EOX arms.			448 (4 RCTs)	⊕⊕⊕⊕ HIGH	1 study each compared docetaxel‐cisplatin and docetaxel‐S1 to docetaxel, irinotecan to paclitaxel, docetaxel‐oxaliplatin to docetaxel and FOLFIRI to EOX. All 4 studies measured serious adverse events, however, none of them reported the total number of participants with any serious adverse events.
Tumour response rate (non‐taxane monotherapy vs taxane monotherapy)	Study population		RR 0.65 (0.34 to 1.26)	179 (1 RCT)	⊕⊕⊝⊝ LOW ⁴	1 study compared irinotecan to paclitaxel.
	136 per 1,000 (71 to 263) (irinotecan)	209 per 1,000 (paclitaxel)
Tumour response rate (taxane‐containing doublet therapy vs taxane monotherapy)	Study population		RR 1.47 (0.56 to 3.89)	121 (2 RCTs)	⊕⊝⊝⊝ VERY LOW ^{5 6}	1 study examined both docetaxel‐cisplatin and docetaxel‐S1 and the other study examined docetaxel‐oxaliplatin. Control arm was docetaxel in both studies. Heterogeneity was very low (I² = 0%, P = 0.92).
	147 per 1,000 (56 to 389) (taxane‐containing doublet therapy)	100 per 1,000 (taxane monotherapy)
Tumour response rate (non‐taxane containing therapy vs another non‐taxane containing therapy)	Study population		RR 0.89 (0.48 to 1.67)	107 (1 RCT)	⊕⊕⊝⊝ LOW ⁷	1 study compared FOLFIRI to EOX chemotherapy.
	254 per 1,000 (137 to 477) (FOLFIRI)	286 per 1,000 (EOX)
Adverse events	Neutropenia, anaemia, diarrhoea and anorexia were some of the commonly experienced adverse events across all the studies. Neuropathy was associated with paclitaxel, docetaxel‐S1 and EOX treatment.			448 (4 RCTs)	⊕⊕⊕⊝ MODERATE ⁸	4 studies measured any adverse events, however, none of them reported the total number of participants with any adverse events.
Quality of life	No difference was observed for global QoL between docetaxel‐cisplatin, docetaxel‐S1 and docetaxel arms, however, worse QoL was reported for participants receiving the doublet chemotherapy for several symptom categories including physical functioning, fatigue and appetite loss.			70 (1 RCT)	⊕⊝⊝⊝ VERY LOW ^{9 10}	1 study comparing docetaxel‐cisplatin to docetaxel‐S1 and docetaxel reported QoL. QoL was assessed using EORTC QLQ‐C30 and gastric modules QLQ‐STO22.
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; HR: hazard ratio; mOS: median overall survival; mPFS: median progression‐free survival; RR: Risk ratio; RCT: randomised controlled trial.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Overall survival	2	547	Hazard Ratio (IV, Random, 95% CI)	0.66 [0.52, 0.83]
1.2 Progression‐free survival	1	507	Hazard Ratio (IV, Random, 95% CI)	0.57 [0.47, 0.69]
1.3 Serious adverse events	1	503	Risk Ratio (M‐H, Random, 95% CI)	1.38 [1.20, 1.59]
1.4 Tumour response rate	1	435	Risk Ratio (M‐H, Random, 95% CI)	2.17 [0.63, 7.48]
1.5 Any adverse events	1	503	Risk Ratio (M‐H, Random, 95% CI)	1.04 [1.00, 1.09]
1.6 Subgroup analysis by line of treatment: overall survival	2	547	Hazard Ratio (IV, Random, 95% CI)	0.66 [0.52, 0.83]
1.6.1 Second line	1	40	Hazard Ratio (IV, Random, 95% CI)	0.48 [0.25, 0.92]
1.6.2 Third line and beyond	1	507	Hazard Ratio (IV, Random, 95% CI)	0.69 [0.56, 0.85]
1.7 Subgroup analysis by geographical region: overall survival	2	547	Hazard Ratio (IV, Random, 95% CI)	0.66 [0.54, 0.81]
1.7.1 East and South‐East Asia	1	73	Hazard Ratio (IV, Random, 95% CI)	0.77 [0.46, 1.29]
1.7.2 Rest of the world	2	474	Hazard Ratio (IV, Random, 95% CI)	0.65 [0.52, 0.80]
1.8 Subgroup analysis by geographical region: progression‐free survival	1	507	Hazard Ratio (IV, Random, 95% CI)	0.52 [0.33, 0.82]
1.8.1 East and South‐East Asia	1	73	Hazard Ratio (IV, Random, 95% CI)	0.33 [0.19, 0.57]
1.8.2 Rest of the world	1	434	Hazard Ratio (IV, Random, 95% CI)	0.61 [0.49, 0.76]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Overall survival	3	781	Hazard Ratio (IV, Random, 95% CI)	0.55 [0.41, 0.73]
2.2 Progression‐free survival	3	781	Hazard Ratio (IV, Random, 95% CI)	0.33 [0.19, 0.57]
2.3 Serious adverse events	2	638	Risk Ratio (M‐H, Random, 95% CI)	1.14 [0.95, 1.37]
2.4 Tumour response rate	3	687	Risk Ratio (M‐H, Random, 95% CI)	5.12 [1.23, 21.27]
2.5 Any adverse events	1	491	Risk Ratio (M‐H, Random, 95% CI)	1.08 [1.00, 1.17]
2.6 Subgroup analysis by line of treatment: progression‐free survival	2	288	Hazard Ratio (IV, Random, 95% CI)	0.28 [0.18, 0.44]
2.6.1 Second‐line	1	62	Hazard Ratio (IV, Random, 95% CI)	0.49 [0.28, 0.86]
2.6.2 Third‐line and beyond	2	226	Hazard Ratio (IV, Random, 95% CI)	0.24 [0.17, 0.34]
2.7 Subgroup analysis by geographical region: progression‐free survival (VEGFR‐targeted therapy)	2	288	Hazard Ratio (IV, Random, 95% CI)	0.24 [0.11, 0.50]
2.7.1 East and South‐East Asia	2	195	Hazard Ratio (IV, Random, 95% CI)	0.17 [0.12, 0.24]
2.7.2 Rest of the world	1	93	Hazard Ratio (IV, Random, 95% CI)	0.61 [0.39, 0.95]
2.8 Subgroup analysis by types of biological therapy: overall survival	3	781	Hazard Ratio (IV, Random, 95% CI)	0.55 [0.41, 0.73]
2.8.1 VEGFR‐targeted therapy	2	288	Hazard Ratio (IV, Random, 95% CI)	0.50 [0.31, 0.79]
2.8.2 Immunotherapy	1	493	Hazard Ratio (IV, Random, 95% CI)	0.63 [0.51, 0.78]
2.9 Subgroup analysis by types of biological therapy: progression‐free survival	3	781	Hazard Ratio (IV, Random, 95% CI)	0.33 [0.19, 0.57]
2.9.1 VEGFR‐targeted therapy	2	288	Hazard Ratio (IV, Random, 95% CI)	0.26 [0.15, 0.45]
2.9.2 Immunotherapy	1	493	Hazard Ratio (IV, Random, 95% CI)	0.60 [0.49, 0.73]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Overall survival	1	395	Hazard Ratio (IV, Random, 95% CI)	0.82 [0.66, 1.02]
3.2 Progression‐free survival	1	395	Hazard Ratio (IV, Random, 95% CI)	1.27 [1.03, 1.57]
3.3 Serious adverse events	1	570	Risk Ratio (M‐H, Random, 95% CI)	0.41 [0.30, 0.57]
3.4 Tumour response rate	1	395	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.72, 1.88]
3.5 Any adverse events	1	570	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.56, 0.71]
3.6 Subgroup analysis by geographical region: overall survival	1	395	Hazard Ratio (IV, Random, 95% CI)	0.84 [0.66, 1.06]
3.6.1 East and South‐East Asia	1	104	Hazard Ratio (IV, Random, 95% CI)	0.90 [0.59, 1.37]
3.6.2 Rest of the world	1	291	Hazard Ratio (IV, Random, 95% CI)	0.81 [0.61, 1.08]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Overall survival	7	2743	Hazard Ratio (IV, Random, 95% CI)	0.93 [0.83, 1.04]
4.2 Progression‐free survival	7	2743	Hazard Ratio (IV, Random, 95% CI)	0.87 [0.74, 1.02]
4.3 Serious adverse events	4	1618	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.95, 1.44]
4.4 Tumour response rate	6	2404	Risk Ratio (M‐H, Random, 95% CI)	1.35 [0.99, 1.85]
4.5 Any adverse events	3	1513	Risk Ratio (M‐H, Random, 95% CI)	1.01 [1.00, 1.03]
4.6 QoL	2	1154	Hazard Ratio (IV, Random, 95% CI)	0.88 [0.74, 1.04]
4.7 Subgroup analysis by geographical region: overall survival	6	2030	Hazard Ratio (IV, Random, 95% CI)	0.88 [0.78, 0.99]
4.7.1 East and South‐EastAsia	5	1099	Hazard Ratio (IV, Random, 95% CI)	0.95 [0.78, 1.16]
4.7.2 Rest of the world	3	931	Hazard Ratio (IV, Random, 95% CI)	0.84 [0.73, 0.97]
4.8 Subgroup analysis by geographical region: progression‐free survival	5	1684	Hazard Ratio (IV, Random, 95% CI)	0.77 [0.68, 0.87]
4.8.1 East and South‐East Asia	4	942	Hazard Ratio (IV, Random, 95% CI)	0.79 [0.68, 0.92]
4.8.2 Rest of the world	2	742	Hazard Ratio (IV, Random, 95% CI)	0.74 [0.57, 0.95]
4.9 Subgroup analysis by extent of disease: overall survival	2	859	Hazard Ratio (IV, Random, 95% CI)	0.95 [0.68, 1.33]
4.9.1 Locally advanced	1	14	Hazard Ratio (IV, Random, 95% CI)	1.56 [0.31, 7.85]
4.9.2 Metastatic	2	845	Hazard Ratio (IV, Random, 95% CI)	0.93 [0.64, 1.35]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
5.1 Overall survival (non‐taxane monotherapy vs taxane monotherapy)	1	219	Hazard Ratio (IV, Random, 95% CI)	1.13 [0.86, 1.48]
5.2 Overall survival (taxane‐containing doublet therapy vs taxane monotherapy)	2	121	Hazard Ratio (IV, Random, 95% CI)	1.05 [0.72, 1.54]
5.3 Progression‐free survival (non‐taxane monotherapy vs taxane monotherapy)	1	219	Hazard Ratio (IV, Random, 95% CI)	1.14 [0.88, 1.48]
5.4 Progression‐free survival (taxane‐containing doublet therapy vs taxane monotherapy)	2	121	Hazard Ratio (IV, Random, 95% CI)	0.75 [0.52, 1.09]
5.5 Tumour response rate (non‐taxane monotherapy vs taxane monotherapy)	1	179	Risk Ratio (M‐H, Random, 95% CI)	0.65 [0.34, 1.26]
5.6 Tumour response rate (taxane‐containing doublet therapy vs taxane monotherapy)	2	121	Risk Ratio (M‐H, Random, 95% CI)	1.47 [0.56, 3.89]
5.7 Tumour response rate (non‐taxane containing therapy vs another non‐taxane containing therapy)	1	107	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.48, 1.67]

*Study characteristics*
Methods	International double‐blind placebo‐controlled phase 3 RCT In the overall population, median duration of follow‐up for OS was 11.1 months in olaparib‐paclitaxel group and 9.9 months in placebo‐paclitaxel group In ATM‐negative participants, median duration of follow‐up for OS was 12.6 months for olaparib‐paclitaxel group and 15.4 months for placebo‐paclitaxel group
Participants	Diagnosis: advanced gastric cancer including OGJ cancer Previous treatment status: disease progression following or during first‐line chemotherapy consisting of only 5‐fluoropyrimidine and platinum doublet combination Number of participants: 525 enrolled during 3rd September 2013 and 28th March 2016 Median age: 58 (49 to 67) years in Arm A and 59 (50 to 65) years in Arm B Number of male participants: 359 ECOG PS status: 224 PS0, 299 PS1 and 2 unknown Extent of disease: 9 locally advanced, 514 metastatic and 2 unknown Location of primary cancer: 14 OGJ, 509 gastric and 2 unknown Study locations: 58 centres in 4 countries in Asia (China, Japan, South Korea, and Taiwan) Other relevant key eligibility criteria were: life expectancy of at least 16 weeks; at least 1 measurable or non‐measurable lesion; acceptable bone marrow, liver, and renal function. Participants who had received more than 1 previous chemotherapy regimen for the treatment of advanced gastric cancer and those with history of previous treatment with PARP inhibitor or taxane were excluded. Of 525 participants randomised, 1 participant in each arm withdrew consents before receiving study treatment.
Interventions	Arm A (263 participants): oral olaparib 100 mg twice daily continuously in combination with intravenous paclitaxel 80 mg/m² on Day 1, 8, 15 of 28‐day cycle Arm B (262 participants): oral placebo twice daily continuously in combination with intravenous paclitaxel 80 mg/m² on Day 1, 8, and 15 of 28‐day cycle
Outcomes	Primary outcome: OS Secondary outcome: PFS, ORR, time to deterioration of health‐related QoL, safety, tolerability
Notes	QLQ‐STO22 (HRQoL22‐question scale developed specifically for participants with gastric cancer) was one of secondary outcomes, which was going to be reported separately, but publication not yet available. AstraZeneca provided organisational support and had a role in study design, data collection, data analyses, data interpretation, and the writing of the report. NCT01924533
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation scheme was produced by the Global Randomisation system at AstraZeneca that generates random numbers and randomisation codes were assigned sequentially as participants became eligible for randomisation
Allocation concealment (selection bias)	Low risk	Randomisation scheme was loaded into an interactive voice and web response system database
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Olaparib and matching placebo treatment were masked to participants, investigators and study monitors
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Olaparib and matching placebo treatment were masked to participants, investigators and study monitors
Imbalance in baseline characteristics across arms (confounding)	Low risk	No stratification factors were included in the study. Demographic and baseline disease characteristics were generally well‐balanced between the treatment arms in the overall population, and no major differences were noted. In the ATM‐negative population minor imbalances existed; 48% and 37% of participants had diffuse histology type, 46% and 33% of participants had > 2 metastases at baseline and 67% and 54% of had undergone a gastrectomy in the olaparib‐paclitaxel and placebo‐paclitaxel arm, respectively.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Response assessment was performed on intention‐to‐treat population (all participants assigned to treatment) while adverse events were assessed on safety population (all participants who received ≥ 1 dose) with 1 participant in Arm B (placebo) counted in Arm A (olaparib).
Incomplete outcome data (attrition bias) for QoL outcome	High risk	Only time to deterioration of health‐related‐QoL measured by EORTC QLQ‐C30 has been so far published. 93% and 66% of the participants provided QLQ‐C30 data at baseline and ≥ 1 post‐baseline time point, respectively.
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported
Other bias	Low risk	No obvious potential source of bias

*Study characteristics*
Methods	Multicentre open‐label phase 3 RCT Media follow‐up period 17.6 months
Participants	Diagnosis: histologically‐confirmed metastatic or recurrent gastric adenocarcinoma Previous treatment status: disease progression confirmed by CT, endoscopy or other imaging techniques during or within 1 month after last dose of first line chemotherapy with fluoropyrimidine plus platinum (in case of treatment with adjuvant or neoadjuvant chemotherapy consisting of fluoropyrimidine plus platinum, disease progression during treatment or within 6 months after treatment completion was allowed) Number of participants: 219 enrolled during August 2007 and August 2010 Median age: 64.5 (37 to 75) years in Arm A and 65 (38 to 75) years in Arm B Number of male participants: 171 ECOG PS status: 211 PS0‐1 and 8 PS2 Study locations: 37 centres in Japan Other relevant key eligibility criteria were: adequate bone marrow, hepatic and renal functions. Participants with prior history of chemotherapy with taxanes or irinotecan and those with severe peritoneal metastases were excluded. Of 223 participants randomly allocated, 5 in paclitaxel arm and 2 in irinotecan arm discontinued study treatment due to withdrawal of consent.
Interventions	Arm A (111 participants): intravenous paclitaxel 80 mg/m² on Day 1, 8, 15 of 28‐day cycle Arm B (112 participants): intravenous irinotecan 150 mg/m² on Day 1 and 15 of 28‐day cycle
Outcomes	Primary outcome: OS Secondary outcome: PFS, RR, safety
Notes	The study was funded by Yakult Pharmaceutical Industry and Daiichi Sankyo UMIN000001252
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using minimisation method
Allocation concealment (selection bias)	Low risk	Random assignment was carried out centrally at the data centre
Blinding of participants and personnel (performance bias) All outcomes	High risk	Open‐label study
Blinding of outcome assessment (detection bias) All outcomes	Low risk	An independent statistician and data analysis centre performed primary analysis for OS and all investigators remained blinded to data until analysis was completed.
Imbalance in baseline characteristics across arms (confounding)	Low risk	Random assignment was carried out using minimisation method with the following adjustment factors; institution, ECOG PS (0‐1 vs 2) & measurable lesions (presence vs absence). Baseline characteristics of patients in the intention‐to‐treat population were well‐balanced between the two treatment arms.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Survival analyses were performed on intention‐to‐treat population (all randomly assigned participants who met eligibility criteria with those found to be ineligible after random assignment being excluded), whereas safety analyses were performed on safety population (all randomly assigned participants who received ≥ 1 dose of study treatment). Response rate was assessed on participants with ≥ 1 measurable lesion at baseline.
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported
Other bias	Low risk	No obvious potential source of bias

*Study characteristics*
Methods	Multicentre phase 2 RCT
Participants	Diagnosis: histologically‐confirmed metastatic or recurrent gastric adenocarcinoma Previous treatment status: objective radiographic disease progression either during first‐line chemotherapy or within 6 months after last dose of cisplatin‐based adjuvant chemotherapy. 96.6% of participants in Arm A (docetaxel) and 96% of participants in Arm B (docetaxel‐oxaliplatin) had fluoropyrimidine‐platinum containing chemotherapy regimen. Number of participants: 52 enrolled during January 2009 to January 2012 Median age: 54 (Arm A) and 59 (Arm B) years Number of male participants: 42 ECOG PS status: 8 PS0, 42 PS1 and 2 PS2 Study locations: 6 centres in South Korea Other relevant key eligibility criteria were: at least 1 measurable lesion, life expectancy > 12 weeks; adequate organ function. Participants with following conditions were excluded: previous exposure to docetaxel or oxaliplatin; symptomatic peripheral neuropathy of Grade ≥ 2; brain metastasis; history of other tumour except BCC or successfully treated cervical carcinoma; another serious medical condition; hypersensitivity to atropine; pregnant/breast‐feeding.
Interventions	Arm A (27): intravenous docetaxel 36 mg/m² on Day 1 and 8 of 21‐day cycle up to max 9 cycles Arm B (25): intravenous docetaxel 36 mg/m² on Day 1 and 8 and intravenous oxaliplatin on Day 1 of 21‐day cycle up to max 9 cycles
Outcomes	Primary outcome: ORR Secondary outcome: PFS, OS
Notes	Trial was terminated early due to poor px accrual rate.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not described
Allocation concealment (selection bias)	Unclear risk	Not described
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Not described
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not described
Imbalance in baseline characteristics across arms (confounding)	Unclear risk	Randomisation was stratified by ECOG PS (0‐1 vs 2). Higher proportion of participants in the docetaxel arm had metastases of lymph nodes and bladder/ureter at base line; 44.4% and 14.8% in the docetaxel arm, respectively and 32% and 4% in the docetaxel‐oxaliplatin arm, respectively. They were considered unlikely to affect the outcome of this review. 32% of participants in the docetaxel‐oxaliplatin arm had received paclitaxel‐S1‐cisplatin previously, compared to 22.2% of participants in the docetaxel arm.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Both efficacy and safety analyses were performed on intention‐to‐treat population (all enrolled participants)
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Low risk	No protocol was available, however, all important outcomes were reported
Other bias	High risk	Trial was terminated early due to poor px accrual rate

*Study characteristics*
Methods	Multicentre phase 2 RCT Median follow‐up time 7.3 months (range 1.6 to 31.5)
Participants	Diagnosis: histologically‐confirmed metastatic gastric adenocarcinoma Previous treatment status: disease progression during or within 6 months of completion of first‐line chemotherapy with S1‐ or capecitabine‐cisplatin; Number of participants: 69 Median age: 56 (34 to 68) (Arm A), 55 (38 to 74) (Arm B) and 55 (39 to 68) (Arm C) years Number of male participants: 52 ECOG PS status: 4 PS0, 61 PS1 and 4 PS2 Study locations: 3 centres in South Korea Other relevant key eligibility criteria were: measurable lesion; adequate major organ function. Participants with following conditions were excluded: concurrent uncontrolled medical conditions; other malignancies within past 3 years; prior treatment with taxane; pre‐existing Grade ≥ 2 neuropathy. Of 72 participants randomly allocated, 1 in docetaxel‐S‐1 arm withdrew consent before receiving treatment.
Interventions	Arm A (23 participants): intravenous docetaxel 75 mg/m² on Day 1 of 21‐day cycle Arm B (23 participants): intravenous docetaxel 60 mg/m² on Day 1 and intravenous cisplatin 60 mg/m² on Day 1 of 21‐day cycle Arm C (23 participants):intravenous docetaxel 60 mg/m² on Day 1 and oral S1 30 mg/m² on Day 1 to 14 of 21‐day cycle Study treatment was continued until disease progression, unacceptable toxicity or withdrawal of consent by participants.
Outcomes	Primary outcome: ORR Secondary outcome: PFS, OS, safety, QoL
Notes	Trial was terminated early due to slow recruitment. Study was supported by the Research Institute and Hospital, National Cancer Center, Republic of Korea. NCT00980603
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using random permutation method
Allocation concealment (selection bias)	Unclear risk	Not described
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Not described
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not described
Imbalance in baseline characteristics across arms (confounding)	Unclear risk	Randomisation was stratified according to study site, ECOG PS (0‐1 vs 2), and best response to first‐line chemotherapy. Smaller proportion of participants in the docetaxel‐S1 arm were males; 60.9% in docetaxel‐sunitinib arm compared to 78.3% and 87% in the docetaxel and docetaxel‐cisplatin arms, respectively. 95.7% of participants in the docetaxel arm had upfront metastatic disease compared to 78.3% in the docetaxel and docetaxel‐cisplatin arms. More participants in the docetaxel‐S1 arm had ≥ 2 metastases at baseline.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Efficacy analyses were performed on modified intention‐to‐treat population (all randomised participants excluding those who were deemed ineligible or never started study treatment), whereas safety analyses were performed on safety population (all participants who received at least 1 dose of study treatment).1 participant in Arm B (docetaxel‐cisplatin) was considered ineligible and excluded from efficacy analyses, but included in safety analyses. 2 participants in Arm C did not receive study treatment and excluded from both efficacy and safety analyses. Modified intention‐to‐treat population was 95.8% of randomised participants.
Incomplete outcome data (attrition bias) for QoL outcome	High risk	More than 60% of participants in each arm completed baseline QoL questionnaire and at least one post‐treatment questionnaire. Although it was described compliance was similar in 3 arms, the exact proportion of participants completing QoL questionnaire at different time points were not provided.
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported
Other bias	High risk	Trial was terminated early due to slow recruitment.

*Study characteristics*
Methods	Multicentre double‐blind placebo‐controlled phase 2 RCT
Participants	Diagnosis: histologically‐confirmed advanced or metastatic gastric cancer including OGJ adenocarcinoma Previous treatment status: lack of response or intolerance to at least 2 chemotherapy including both platinum and fluoropyrimidine. Whether participants had received fluoropyrimidine and platinum concurrently was not described. Number of participants: 141 with enrolment starting June 2009 and participants being observed until October 2010 Median age: 55 (Arm A), 53 (Arm B), 54 (Arm C) years Number of male participants: 109 Extent of disease: 5 locally advanced, 136 metastatic Study locations: 15 centres in China Other relevant key eligibility criteria were: at least 1measurable lesion; acceptable haematologic, hepatic and renal function. Participants with uncontrolled BP on medication (> 140/90 mmHg), bleeding tendency or those receiving thrombolytics or anticoagulants were excluded.
Interventions	Arm A (47 participants): oral apatinib 850 mg once daily continuously on 28‐day cycle Arm B (46 participants): oral apatinib 425 mg twice daily continuously on 28‐day cycle Arm C (48 participants): oral placebo 2 tablets in the morning and 1 tablet in the evening continuous on 28‐day cycle Study treatment was continued until disease progression, intolerable toxicity or withdrawal of consent by participants. of 141 participants randomly assigned, 6 in placebo arm, 7 in apatinib 850 mg arm and 6 in apatinib 425 mg arm withdrew consent.
Outcomes	Primary outcome: PFS Secondary outcome: DCR, ORR, OS, QoL, safety
Notes	Fudan University Shanghai Cancer Center funded Manette Marais, PhD to assist with writing and editing the article. NCT00970138
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not described
Allocation concealment (selection bias)	Low risk	Random assignment was centrally managed
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Double‐blind placebo‐controlled trial
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Double‐blind placebo‐controlled trial
Imbalance in baseline characteristics across arms (confounding)	Low risk	Randomisation was stratified according to the number of metastatic organs. The baseline characteristics of patients in the different groups were similar with regard to age, sex ratio, surgical history, disease stage, and number of metastatic organs.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Both response and safety analyses were performed on intention‐to‐treat population (all randomised participants excluding those who withdrew consent before receiving study treatment) . The intention‐to‐treat population was 97.2% of the randomly assigned participants.
Incomplete outcome data (attrition bias) for QoL outcome	Unclear risk	No information on the number of participants who completed QoL questionnaire or the results of QoL questionnaire was provided.
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported
Other bias	Low risk	No obvious potential source of bias

*Study characteristics*
Methods	Single‐centre RCT Study enrolment commenced 1st January 2010 and follow‐up ended 26th September 2015
Participants	Disease: advanced GC confirmed by oesophago‐gastro‐duodenoscopy or post‐surgical pathology, including those inoperable or relapsed/metastasis after surgery Previous treatment status: disease progression after DCF chemotherapy Number of participants: 107 enrolled between 1st January 2010 and 1st June 2013 Median age: 53 (31 to 71) years Number of male participants: 70 ECOG PS status: 81 PS0‐1 and 26 PS2 Extent of disease: 46 locally advanced and 61 metastatic Study locations: single centre in China Other relevant key eligibility criteria were: adequate routine bloods including liver and kidney function; normal ECG
Interventions	Arm A (56 participants): intravenous epirubicin 50 mg/m² on Day 1, intravenous oxaliplatin 85 mg/m² on Day 1 and oral capecitabine 625 mg/m² twice daily D1‐14 of 21‐day cycle Arm B (51 participants): intravenous irinotecan 180 mg/m² on Day 1, intravenous leucovorin 400 mg/m² on Day 1, intravenous 5‐FU, 400mg/m² on Day 1 over 24 hours of 14‐day cycle
Outcomes	Endpoints: RR, DCR, safety, PFS, OS
Notes	Text in Chinese. No HR was reported.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using a digit randomisation table
Allocation concealment (selection bias)	Unclear risk	Unable to assess on the published full text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Unable to assess on the published full text
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unable to assess on the published full text
Imbalance in baseline characteristics across arms (confounding)	Low risk	The baseline characteristics of patients in the different groups were similar with regard to age, sex ratio, ECOG PS, surgical history, disease stage, histological subtype, and number of metastatic organs.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Efficacy analyses were performed on intention‐to‐treat population (all the randomised participants)
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Low risk	No protocol was available, however, all important outcomes were reported
Other bias	Low risk	No obvious potential source of bias

*Study characteristics*
Methods	Multicentre open‐label phase 2 RCT
Participants	Diagnosis: HER2‐positive advanced unresectable or recurrent G/GEJ cancer Previous treatment status: progression first‐line chemotherapy with trastuzumab + fluoropyrimidine + platinum; ECOG0‐2 Number of participants: 89 enrolled between December 2012 to October 2016 Study locations: 8 centres in Japan Other relevant key eligibility criteria were: ECOG PS 0‐2; expected survival ≥ 90 days; measurable or evaluable lesion, adequate bone marrow and kidney/liver functions; left ventricular Ejection fraction (LVEF) ≥ 50% measured by echocardiography or multi‐gated acquisition scan (MUGA)
Interventions	Arm A (44 participants): intravenous paclitaxel 80 mg/m² on Day 1, 8, 15 of 28‐day cycle and trastuzumab 8 mg/kg followed by 6 mg/kg on Day 1 of 21‐day cycle Arm B (45 participants): intravenous paclitaxel 80 mg/m² on Day 1, 8, 15 of 28‐day cycle
Outcomes	Primary endpoint: PFS Secondary endpoints: OS, RR, safety and translational biomarker research
Notes	Only abstract published UMIN000009297
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Unable to assess on the published abstract
Allocation concealment (selection bias)	Unclear risk	Unable to assess on the published abstract
Blinding of participants and personnel (performance bias) All outcomes	High risk	Open‐label study
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unable to assess on the published abstract
Imbalance in baseline characteristics across arms (confounding)	Unclear risk	Unable to assess on the published abstract
Incomplete outcome data (attrition bias) for outcomes other than QoL	Unclear risk	Unable to assess on the published abstract
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Unclear risk	All the relevant predefined efficacy outcomes were reported, however, safety outcomes were very limited
Other bias	Unclear risk	Unable to assess on the published abstract

*Study characteristics*
Methods	Multinational double‐blind placebo‐controlled phase 3 RCT
Participants	Disease: cytologically‐ or histologically‐confirmed advanced gastric or OGJ adenocarcinoma that is metastatic or locally advanced and unresectable Previous treatment status: progression on one prior line of therapy containing a fluoropyrimidine/platinum doublet Number of participants: 714 with enrolment commencing October 2014 Number of male participants: 72.1% Location of primary cancer: 25.4% OGJ and 74.6% gastric Study locations: 263 locations in 22 countries worldwide Other relevant key eligibility criteria were: patients with either measurable disease or non‐measurable evaluable disease were eligible ; ECOG PS 0‐1; adequate bone marrow, liver and renal functions. Participants with following conditions were excluded: prior taxanes in the neoadjuvant or adjuvant setting with progression occurring within 6 months of completion of taxane therapy or any taxanes in the metastatic setting; any known symptomatic brain metastases requiring steroids; peripheral neuropathy ≥ CTCAE Grade 2 at baseline.
Interventions	Arm A (estimated to be 357 participants): oral napabucasin 480 mg two times daily and intravenous paclitaxel 80 mg/m2 intravenously on Days 1, 8, and 15 of every 4‐week cycle Arm B (estimated to be 357 participants): oral placebo two times daily and intravenous paclitaxel 80 mg/m2 on Days 1, 8, and 15 of every 4‐week cycle Study treatment was continued until disease progression, death, intolerable toxicity or patient/investigator decision to stop treatment
Outcomes	Primary endpoint: OS Secondary endpoints: PFS, ORR, DCR and safety
Notes	Only abstract published NCT02178956
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Unable to assess on the published abstract
Allocation concealment (selection bias)	Unclear risk	Unable to assess on the published abstract
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Double‐blind placebo‐controlled trial, however, at interim analysis of 380 events, the data safety monitoring board recommended trial unblinding when meeting the primary endpoint at final analyses appeared unlikely, though no safety concerns of clinical significance were identified.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Double‐blind placebo‐controlled trial, however, at interim analysis of 380 events, the data safety monitoring board recommended trial unblinding when meeting the primary endpoint at final analyses appeared unlikely, though no safety concerns of clinical significance were identified.
Imbalance in baseline characteristics across arms (confounding)	Unclear risk	Unable to assess on the published abstract
Incomplete outcome data (attrition bias) for outcomes other than QoL	Unclear risk	Unable to assess on the published abstract
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Unclear risk	All the relevant predefined efficacy outcomes were reported
Other bias	Unclear risk	Unable to assess on the published abstract

*Study characteristics*
Methods	International open‐label phase 3 RCT Median follow‐up was 7·9 months (IQR 3·4 to 14·6) in the total population and 8·5 months (3·7 to 15·7) in the population with a PD‐L1 CPS ≥ 1.
Participants	Diagnosis: histologically‐ or cytologically‐confirmed adenocarcinoma of stomach or OGJ that was metastatic or locally advanced but unresectable Previous treatment status: disease progression per RECIST v1.1 after first‐line chemotherapy with a platinum and fluoropyrimidine (as well as with trastuzumab in participants with HER2‐positive tumours) Number of participants: 592 enrolled between 4th June 4 2015 and 26th July 2016 Mean age: 62·5 (54 to 70) years in Arm A and 60·0 (53 to 68) years in Arm B Number of male participants: 410 ECOG PS status: 164 PS0, 327 PS1 and 1 PS2 Extent of disease: 6 locally advanced and 586 metastatic Location of primary cancer: 185 OGJ and 407 gastric Study locations: 140 centres in 30 countries worldwide Other relevant key eligibility criterion was availability of a tumour sample for PD‐L1 assessment. Participants with following conditions were excluded: squamous or undifferentiated histology; previous treatment with any PD‐1, PD‐L1 or PD‐L2 inhibitor; active autoimmune disease that necessitated systemic treatment. Of 592 participants randomly assigned, 9 (6 with CPS ≥ 1) in pembrolizumab arm and 20 (13 with CPS ≥ 1) in paclitaxel arm discontinued treatment due to withdrawal of consent.
Interventions	Arm A (296 participants with 196 PD‐L1 combined positive score ≥ 1): intravenous pembrolizumab 200 mg on Day 1 of 21‐day cycle Arm B (296 participants with 199 PD‐L1 combined positive score ≥ 1): intravenous paclitaxel 80 mg/m² on Day 1, 8, 15 of 28‐day cycle Study treatment was continued until disease progression, intolerable toxicity, doctor decision or withdrawal of consent by participants.
Outcomes	Primary outcome: OS and PFS (assessed per RECIST by masked and independent central review) in PD‐L1 combined positive score ≥ 1 participants Secondary outcome: RR, duration of response, OS and PFS in total population, PFS (assessed per RECIST by local investigators and per irRECIST by masked and independent central review) in PD‐L1 combined positive score ≥ 1 participants and total population, TTP, safety
Notes	After 489 participants were enrolled, the independent data monitoring committee recommended that enrolment to be restricted to participants with PD‐L1 combined positive score of ≥ 1. Study was funded by Merck Sharp & Dohme, a subsidiary of Merck & Co., who participated in study design, data analysis and interpretation, and manuscript writing. NCT02370498
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Allocation schedule was generated by system vendor using a computerised random list generator
Allocation concealment (selection bias)	Low risk	Random allocation was performed using a central interactive voice‐response and integrated web‐response system
Blinding of participants and personnel (performance bias) All outcomes	High risk	Patients, treating doctors, the external data monitoring committee and sponsor representatives were not masked to treatment assignment.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Response assessment was performed by the central radiological reviewers who were masked to treatment assignment
Imbalance in baseline characteristics across arms (confounding)	Low risk	Enrolment of the first 125 patients was stratified by geographical region (Europe, Israel, North America, and Australia vs Asia vs rest of world) and ECOG performance status (0 vs 1). Following a protocol amendment, enrolment of the remaining 467 patients was stratified by geographical region (Europe, Israel, North America, and Australia vs Asia vs rest of the world), time to progression on first‐line therapy (< 6 months vs ≥ 6 months), and PD‐L1 CPS (< 1 vs ≥ 1). Baseline demographics and disease characteristics were generally balanced between groups in the total population and the population with a PD‐L1 CPS of 1 or higher.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Efficacy analyses were performed on intention‐to‐treat population (all randomised participants irrespective of whether they received treatment), while safety analyses were performed on safety population (all participants who received at least 1 dose of study treatment)
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported
Other bias	Low risk	No obvious potential source of bias

*Study characteristics*
Methods	Multicentre open‐label phase 3 RCT The participants were randomised between October 2002 and December 2006. Follow‐up was complete when the last participant died on the 10th of October 2007.
Participants	Diagnosis: measurable and/or evaluable histologically‐proven adenocarcinoma of stomach or OGJ which were metastatic or locally advanced disease with surgical incurability Previous treatment status: 1 palliative chemotherapy with documented objective imaging proven progression during or within 6 months after the end of first‐line chemotherapy. All participants in Arm A (irinotecan) and 90% of participants in Arm B (BSC) had fluoropyrimidine‐platinum containing chemotherapy for first line. Number of participants: 40 enrolled during October 2002 and December 2006 Mean age: 58 (43 to 73) years in Arm A and 55 (35 to 72) years in Arm B Number of male participants: 29 ECOG PS status: 31 PS0‐1 and 9 PS2 Extent of disease: 40 metastatic Location of primary cancer: 17 OGJ and 23 gastric Study locations: 10 centres in Germany Participants with following conditions were excluded: prior second malignancy; uncontrolled infection; CNS metastasis; other severe medical illnesses; major operation within the last 2 weeks; previous treatment with irinotecan; chronic diarrhoea; parallel treatment with any experimental or any other kind of antineoplastic therapy.
Interventions	Arm A (21 participants): intravenous Irinotecan 250 mg/m² in the first cycle which had to be increased to 350 mg/m² in subsequent cycles on Day 1 of 21‐day cycle Study treatment was continued until objective or clinical tumour progression, side effects, participants wish or maximum of 10 cycles. Arm B (19 participants): best supportive care
Outcomes	Primry outcome: OS Secondary outcome: ORR, TTP, toxicity
Notes	Study was closed prematurely due to poor accrual.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation blocks of 4 for each stratification arm were determined by using a coin
Allocation concealment (selection bias)	Low risk	Randomisation was performed centrally
Blinding of participants and personnel (performance bias) All outcomes	High risk	Open‐label study
Blinding of outcome assessment (detection bias) All outcomes	High risk	Open‐label study
Imbalance in baseline characteristics across arms (confounding)	Low risk	Randomisation was stratified according toECOG PS, pretreatment and timing of progression under first‐line treatment. Participants were well‐balanced between the 2 arms for localisation of the primary tumour, histological subtype, number of metastatic sites and response to first‐line chemotherapy, in addition to stratification factors.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	OS analysis was performed on the intention‐to‐treat population (all the randomised participants)
Incomplete outcome data (attrition bias) for QoL outcome	High risk	QoL was not reported due to poor completion rate of questionnaire
Selective reporting (reporting bias)	High risk	PFS, ORR and toxicity data were reported for irinotecan treated participants only and QoL was not published due to poor completion rate of questionnaire
Other bias	High risk	Study was closed prematurely due to poor accrual and the most frequent reason for low recruitment reported from different centres was patient refusal of randomisation.

*Study characteristics*
Methods	International open‐label adaptive phase 2/3 RCT Duration of median follow‐up was 17.5 months (IQR12.1 to 23.0) for trastuzumab emtansine 2.4mg/kg arm and 15.4 months for taxane arm
Participants	Diagnosis: centrally tested HER2‐positive, measurable and/or evaluable advanced gastric cancer including OGJ adenocarcinoma Previoius treatment status: disease progression either during or after first‐line treatment with combination of at least platinum and fluoropyrimidine given concurrently Number of participants: 345 enrolled during 3rd September 2012 and 30th June 2015 (pre‐planned interim was performed on 14th October 2013) Mean age: 62.0 (19 to 79) years in Arm A 62.0 (27 to 80) in Arm B Number of male participants: 272 ECOG PS status: 142 PS0, 201 PS1 and 1 PS2 Extent of disease: 14 locally advanced and 331 metastatic Location of primary cancer: 110 OGJ and 235 gastric Study locations: 107 centres in 28 countries worldwide Participants with following conditions were excluded: another malignancy within previous 5 years; brain metastasis; peripheral neuropathy Grade ≥ 2; cardiopulmonary dysfunction; concurrent serious uncontrolled or known infection with HIV, active Hepatitis B or C. Of 345 participants randomly assigned, 2 participants in trastuzumab emtansine 2.4 mg/kg arm and 3 participants in taxane arm withdrew consent before receiving treatment. Further 9 participants and 11 participants in each arm discontinued treatment due to withdrawal of consent.
Interventions	Arm A (228 participants): intravenous trastuzumab emtansine 2.4 mg/kg on Day 1 of 7‐day cycle Arm B (117 participants): intravenous docetaxel 75 mg/m² on Day 1 of 21‐day cycle or intravenous paclitaxel 80 mg/m² on Day 1 of 7‐day cycle Study treatment was continued until progressive disease, intolerable toxicity, initiation of another anti‐cancer therapy, participants' and/or physicians' decision to discontinue study treatment or study termination.
Outcomes	Primary outcome: OS Secondary outcome: PFS, ORR, duration of response, safety, patient‐reported outcome, pharmacokinetics
Notes	The phase 2 part of the trial examined two dosing regimens of trastuzumab emtansine (2·4 mg/kg weekly compared with 3·6 mg/kg every 3 weeks) to allow selection of the most appropriate dose for assessment in the phase 3 part, which compared the efficacy and safety of the selected dose of trastuzumab emtansine with taxane treatment. F Hoffmann‐La Roche funded the study; they provided the study drugs and was involved in study design, protocol development, regulatory and ethics approvals, safety monitoring and reporting, data management and data analysis and interpretation. NCT01641939
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using a permuted block randomisation procedure
Allocation concealment (selection bias)	Low risk	Randomisation was performed via an interactive voice response system or interactive web response system
Blinding of participants and personnel (performance bias) All outcomes	High risk	Neither participants nor investigators were masked to treatment assignment
Blinding of outcome assessment (detection bias) All outcomes	High risk	Response assessments were determined by investigators who were not masked to treatment assignment
Imbalance in baseline characteristics across arms (confounding)	Low risk	Randomisation was stratified for world region (Asia Pacific; western Europe, USA, and Canada; other [eastern Europe, Latin America, and remaining countries]), previous HER2‐targeted therapy (yes vs no), and previous gastrectomy (yes vs no). Baseline characteristics were generally well‐balanced.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Survival analyses were performed on intention‐to‐treat population (all participants assigned to treatment), while safety analyses were performed on safety population (all subjects who received at least 1 dose of study treatment). Response rate was assessed on participants with measurable disease at base line.
Incomplete outcome data (attrition bias) for QoL outcome	Unclear risk	Detailed result of patient‐reported outcome including the number of participants who completed the relevant questionnaires was not reported
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes except for patient‐related outcomes were reported
Other bias	Low risk	No obvious potential source of bias

PERMALINK

Salvage systemic therapy for advanced gastric and oesophago‐gastric junction adenocarcinoma

Yoko Tomita

Max Moldovan

Rachael Chang Lee

Amy HC Hsieh

Amanda Townsend

Timothy Price

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Description of the condition

Description of the intervention

How the intervention might work

Why it is important to do this review

Objectives

Primary objective

Secondary objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Types of interventions

Types of outcome measures

Primary outcomes

Secondary outcomes

Search methods for identification of studies

Electronic searches

Searching other resources

Data collection and analysis

Selection of studies

1.

Data extraction and management

Assessment of risk of bias in included studies

2.

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

Assessment of heterogeneity

Assessment of reporting biases

Data synthesis

'Summary of findings' tables

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

Reaching conclusions

Results

Description of studies

Results of the search

Included studies

Excluded studies

Risk of bias in included studies

Allocation

Blinding

Imbalance in baseline characteristics (confounding)

Efficacy and safety outcomes

Quality of life (QoL) outcomes

Selective reporting

Other potential sources of bias

Effects of interventions

Summary of findings 1. Chemotherapy versus placebo, best supportive care (BSC) or no treatment.

Summary of findings 2. Biological therapy versus placebo, best supportive care (BSC) or no treatment.

Summary of findings 3. Chemotherapy versus biological therapy.

Summary of findings 4. Chemmotherapy combined with biological therapy versus chemotherapy.

Summary of findings 5. Chemotherapy versus chemotherapy.

Chemotherapy versus placebo, best supportive care (BSC) or no treatment

Overall survival (OS)

1.1. Analysis.

Progression‐free survival (PFS)

1.2. Analysis.

Serious adverse events (SAEs)

1.3. Analysis.

Tumour response rate (TRR)

*Study characteristics*
Methods	Single centre open‐label phase 2 RCT
Participants	Diagnosis: unresectable or metastatic adenocarcinoma of stomach or OGJ progressed Previous treatment status: disease progression after a first‐line fluoropyrimidine‐platinum regimen Number of participants:105 enrolled during December 2008 and February 2011 Mean age: 54.0 (20 to 72) years in Arm A and 52 (36 to 70) years in Arm B Number of male participants: 73 ECOG PS status: 5 PS0, 71 PS1, 9 PS2 and 20 unknown Extent of disease: 11 locally advanced and 94 metastatic Study locations: single centre in South Korea Other relevant key criteria were: measurable or evaluable disease according to RECIST ver 1.1 and adequate organ function incl. bone marrow, liver and kidney. Participants with following conditions were excluded: severe comorbid illness and/or active infections; Grade ≥ 3 haemorrhage according to CTCAE v3.0 within prior 4 weeks; pregnant or lactating women; active CNS metastasis not controllable with radiotherapy or corticosteroids.
Interventions	Arm A (56 participants): intravenous docetaxel 60 mg/m² on Day 1 and oral sunitinib 37.5 mg once daily continuously of 21‐day cycle Arm B (49 participants): intravenous docetaxel 60 mg/m² on Day 1 of 21‐day cycle Study treatment was continued until tumour progression, unacceptable toxicity or consent withdrawal by participants.
Outcomes	Primary endpoint: TTP Secondary endpoints: OS, ORR, DCR, toxicity
Notes	Pfizer provided sunitinib, but was not involved in the accrual or analysis of the data or in the preparation of the manuscript NCT01238055
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using a random permuted block design
Allocation concealment (selection bias)	Unclear risk	Allocation method was not reported
Blinding of participants and personnel (performance bias) All outcomes	High risk	Open‐label trial
Blinding of outcome assessment (detection bias) All outcomes	Low risk	All treatment responses were centrally reviewed while blinded to treatment arm
Imbalance in baseline characteristics across arms (confounding)	Low risk	Random assignment was stratified by ECOG performance status (0, 1 vs 2). Demographics of the two arms, including median age, sex ratio, ECOG performance scale, laboratory findings (except serum alkaline phosphatase level), and sites of metastatic lesions, were not significantly different.
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Both efficacy and safety analyses were performed on all the randomised participants excluding 2 participants in Arm A (docetaxel‐sunitinib) who withdrew consent before receiving treatment.
Incomplete outcome data (attrition bias) for QoL outcome	Low risk	Not relevant
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported except for PFS, however, this is unlikely to affect the results of this review
Other bias	Low risk	No obvious potential source of bias

Study	Reason for exclusion
Ajani 2010	Included participants who received intervention as first‐line systemic therapy
Al‐Batran 2013	Included participants who received intervention as first‐line systemic therapy
Bando 2016	Included participants who received intervention as first‐line systemic therapy
Bang 2015	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Bang 2017b	Participants had stable disease on first‐line fluoropyrimidine‐platinum chemotherapy
Bang 2018	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Belloni 2015	Non‐RCT
Boku 2009	Included participants who received intervention as first‐line systemic therapy
Chon 2009	Non‐RCT
Cohen 2014	Included participants with oesophageal cancer other than OGJ adenocarcinoma
Cui 2015	Non‐RCT
Enzinger 2016	Included participants who received intervention as first‐line systemic therapy
Ford 2014	Included participants with oesophageal cancer other than OGJ adenocarcinoma
Fuchs 2014	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Fujii 2011	Included participants who received intervention as first‐line systemic therapy
Furue 1985	Included participants who received intervention as first‐line systemic therapy
Fushida 2016	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Goto 1989	Included participants who received intervention as first‐line systemic therapy
Guo 2019	Non‐RCT
Higuchi 2014	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Inokuchi 2007	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Janjigian 2018	Includedparticipants with oesophageal cancer other than OGJ adenocarcinoma
Kang 2012	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Kelly 2020	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Klein 1992	Included participants who received intervention as first‐line systemic therapy
Koizumi 2001	Non‐RCT
Kurokawa 2014	Included participants who received intervention as first‐line systemic therapy
Lee 2012	Non‐RCT
Lee 2018	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Li 2016	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Li 2016b	Included participants who received intervention as first‐line systemic therapy
Li 2018	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Loehrer 1994	Included participants who received intervention as first‐line systemic therapy
Lorenzen 2015	Included participants with oesophageal cancer other than OGJ adenocarcinoma
Luelmo 2008	Included participants who received intervention as first‐line systemic therapy
Maruta 2007	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Moehler 2016	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Morita 2007	Included participants who received intervention as first‐line systemic therapy
Murad 1993	Included participants who received intervention as first‐line systemic therapy
Murakami 2001	Non‐RCT
Nakanishi 2016	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Nakano 1999	Included participants who received intervention as first‐line systemic therapy
Nishikawa 2015	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Nishikawa 2015c	Included participants who received intervention as first‐line systemic therapy
Nishina 2016	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Ochiai 1992	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Ohtsu 2013	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Popov 2008	Included participants who received intervention as first‐line systemic therapy
Pyrhönen 1995	Included participants who received intervention as first‐line systemic therapy
Qiao 2019	Non‐RCT
Qu 2019	Eligibility for inclusion could not be assessed on the published full‐text article
Roy 2013	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Sadighi 2006	Included participants who received intervention as first‐line systemic therapy
Sang 2017	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Satoh 2014	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Satoh 2015	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Schnitzler 1986	Included participants who received intervention as first‐line systemic therapy
Shirao 2013	Included participants who received intervention as first‐line systemic therapy
Shitara 2017	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Sym 2013	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Tabernero 2017	Included participants who received intervention as first‐line systemic therapy
Tanabe 2015	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Tebbutt 2002	Included participants who received intervention as first‐line systemic therapy
Tebbutt 2010	Included participants who received intervention as first‐line systemic therapy
Tebbutt 2016	Included participants who received intervention as first‐line systemic therapy
Tsavaris 1996	Included participants who received intervention as first‐line systemic therapy
Unknown 1992	Included participants who received intervention as first‐line systemic therapy
Van Cutsem 2015	Included participants who received intervention as first‐line systemic therapy
Van Cutsem 2017	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Wu 2020	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Yamamura 1998	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Yoshino 2016	Included participants who received intervention as first‐line systemic therapy
Zhao 2009	Included participants who received intervention as first‐line systemic therapy
Zhong 2007	Included participants who received intervention as first‐line systemic therapy
Zhou 2015	≥10% or unknown proportion of participants received non‐fluoropyrimidine‐platinum chemotherapy for first‐line
Zhu 2016	Non‐RCT

Methods	Open‐label Phase 3 RCT
Participants	Asian patients with advanced PD‐L1–positive (combined positive score [CPS] ≥1) gastric cancer that progressed after platinum + fluoropyrimidine chemotherapy
Interventions	Arm A: pembrolizumab Arm B: paclitaxel
Outcomes	Primary endpoint: OS and PFS
Notes	Inadequte amount of data available for inclusion in the review on the published abstract. Study was discontinued after the result of KEYNOTE‐061 became available (pembrolizumab did not significantly prolong OS compared to paclitaxel). NCT03019588

Methods	Single‐blinded Phase 2 RCT
Participants	Patients with histologically‐proven non‐resectable or metastatic oesophagus, cardia or gastric adenocarcinoma, who have progressed after first‐line treatment with platinum containing chemotherapy
Interventions	Arm A: irinotecan Arm B: bevacizumab + irinotecan
Outcomes	Primary endpoint: PFS
Notes	No publication available

Methods	Phase 2/3 RCT
Participants	Patients with advanced or recurrent gastric cancer, who had not received any chemotherapy except one regimen (not including taxanes)
Interventions	Arm A: docetaxel plus S1 Arm B: 5FU plus CDDP
Outcomes	RR, TTF, PFS, OS and toxicity
Notes	Eligibility for inclusion in the review cannot be assessed on the published abstract.

Methods	Phase 2 RCT
Participants	advanced p53wt (≤ 20% nuclear IHC staining using D07 antibody) gastric adenocarcinoma and disease progression on non‐irinotecan chemotherapy regimen
Interventions	Arm A: irinotecan plus flavopiridol Arm B: irinotecan
Outcomes	Primary endpoint: PFS
Notes	Eligibility for inclusion in the review cannot be assessed on the published abstract.

Methods	Open‐label Phase 2 RCT
Participants	Inoperable or recurrent gastric cancer patients
Interventions	Arm A: TS‐1 based therapy + LNT Arm B: TS‐1 based therapy
Outcomes	Primary endpoint: TTF Secondary endpoints: OS, PFS, QoL, RR, safety and compliance
Notes	No publication available. The study was terminated.

Methods	RCT
Participants	HER‐2‐positive advanced gastric cancer patients
Interventions	Arm A: XELOX or FOLFOX4 + trastuzumab Arm B: XELOX or FOLFOX4
Outcomes	OS, PFS and ORR
Notes	Eligibility for inclusion in the review cannot be assessed on the published abstract.

Methods	International double‐blinded Phase 3 RCT
Participants	Patients with advanced/metastatic adenocarcinoma of the stomach or gastroesophageal junction after failure of ≥ 2 prior lines of chemotherapy
Interventions	Arm A: rivoceranib Arm B: placebo
Outcomes	Primary endpoint: OS Secondary endpoints: PFS, ORR, DCR, QoL, safety and tolerability
Notes	Eligibility for inclusion in the review cannot be assessed on the published abstract. NCT03042611

Methods	Multicentre, open‐label Phase Ib/II RCT
Participants	Patients with histologically‐ or cytologically‐confirmed recurrent or metastatic gastric or gastroesophageal junction adenocarcinoma who have progressed on, or are refractory to standard regimens
Interventions	Arm A: durvalumab and tremelimumab Arm B: durvalumab Arm C: tremelimumab
Outcomes	Primary endpoints: ORR, PFS Secondary endpoints: safety, tolerability, DCR, DOR and OS
Notes	Eligibility for inclusion in the review cannot be assessed on the published abstract. Study is ongoing. NCT02340975