S-1-Based Chemotherapy versus Capecitabine-Based Chemotherapy as First-Line Treatment for Advanced Gastric Carcinoma: A Meta-Analysis

Background Although both oral fluoropyrimidines were reported effective and safe, doubts exist about whether S-1 or capecitabine is more advantageous in advanced gastric carcinoma (AGC). Herein, we performed a meta-analysis to comprehensively compare the efficacy and safety of S-1-based chemotherapy versus capecitabine-based chemotherapy as first-line treatment for AGC. Methods PubMed/Medline, EmBase, Cochrane library, and China National Knowledge Infrastructure databases were searched for articles comparing S-1-based chemotherapy to capecitabine-based chemotherapy for AGC. Primary outcomes were overall response rate (ORR), time to progression (TTP), overall survival (OS), progression-free probability, and survival probability. Secondary outcomes were toxicities. Fixed-effects model were used and all the results were confirmed by random-effects model. Results Five randomized controlled trials and five cohort studies with 821 patients were included. We found equivalent ORR (38.3% vs. 39.1%, odds ratio [OR] 0.92, 95% conﬁdence interval [CI] 0.69-1.24, P = 0.59), TTP (harzad ratio [HR] 0.98, 95% CI 0.82-1.16, P = 0.79), OS (HR 0.99, 95% CI 0.87-1.13, P = 0.91), progression-free probability (3-month OR 1.02, 95% CI 0.62-1.68, P = 0.94; 6-month OR 1.34, 95% CI 0.88-2.04, P = 0.18) and survival probability (0.5-year OR 0.90, 95% CI 0.61-1.31, P =0.57; 1-year OR 0.97, 95% CI 0.70- 1.33, P = 0.84; 2-year OR 1.15, 95% CI 0.61-2.17, P = 0.66). Equivalent grade 3 to 4 hematological and non-hematological toxicities were found except hand-foot syndrome was less prominent in S-1-based chemotherapy (0.3% vs. 5.9%, OR 0.19, 95% CI 0.06-0.56, P = 0.003). There’re no significant heterogeneity and publication bias. Cumulative analysis found stable time-dependent trend. Consistent results stratified by study design, age, regimen, cycle, country were observed. Conclusion S-1-based chemotherapy was associated with non-inferior antitumor efficacy and better safety profile, compared with capecitabine-based therapy. We recommended S-1 and capecitabine can be used interchangeably for AGC, at least in Asia.


Introduction
Gastric carcinoma ranks second among the most common causes of cancer deaths worldwide, with especial high prevalence in Asia [1][2][3]. A large number of gastric cancer patients present with advanced disease (unresectable, recurrent or metastatic disease) precluding surgery and chemotherapy becomes the most effective treatment [4][5][6]. However, a globally accepted standard regimen has not been established, among which fluoropyrimidines comprise the backbone of chemotherapy for advanced gastric carcinoma (AGC) and the optimization was established by extensive research [7,8]. Oral fluoropyrimidines (capecitabine and S-1) have opened new perspectives for treatment for AGC with their simplicity and convenience over the traditional 5-FU [9][10][11].
However, doubts exist about whether S-1 or capecitabine is more advantageous in first-line treatment for AGC. Several clinical trials and cohort studies, comparing S-1 with capecitabine in mono or combined therapy, have published no completely consistent results. Some slightly favored S-1 on efficacy [20], some slightly favored capecitabine [21,22], while some reported equivalent results [23,24]. No consensus on toxicity profiles of these two chemotherapies were reached especially on hand-foot syndrome, thrombocytopenia, stomatitis and diarrhea [20,[22][23][24][25]. These allowed no definite conclusions about the efficacy and safety of these two chemotherapies with limited number of individuals assessed. In addition, the non-uniform study design, regimen, chemotherapy cycle, patient age and country all made people assailed with doubts. Meanwhile, there has been no meta-analysis to detect the difference of these two oral fluoropyrimidines in any cancer.
Evaluation of the efficacy and safety of these two oral fluoropyrimidines will provide necessary and important information for making clinical decision. Therefore, we conducted a meta-analysis with greater power of statistical comparisons to comprehensively compare S-1-based chemotherapy versus capecitabine-based chemotherapy as first-line chemotherapy for AGC.
Corresponding authors were contacted for further details if necessary. Discrepancies were resolved by the third party (Ruihua Xu, Feng Wang) adjudication. To limit publication bias, no language limitation, time limitation or other restrictions such as study design were imposed [27].

Selection Criteria
The inclusion criteria were as follows: (1) studies aimed to compare efficacy or safety between S-1-based chemotherapy and capecitabine-based chemotherapy as first-line chemotherapy for patients with advanced gastric adenocarcinoma (unresectable, recurrent or metastatic gastric cancer); (2) data for calculating the efficacy or safety of these two therapies were provided; (3) randomized controlled trials (RCTs), cohort studies. The exclusion criteria were: (1) studies with no data for efficacy and safety including protocols and phase 1 clinical trials; (2) studies based on overlapping patients; (3) Case reports, abstracts, reviews , conference reports and experiments.

Data Extraction and Outcomes
We extracted data for demography information and potential confounding factors. Primary outcomes were overall response rate (ORR), time to progression (TTP), overall survival (OS), progression-free probability, survival probability. Secondary outcomes were toxicities. Figures were electronically digitized and Kaplan-Meier curves were downloaded by an appropriate software GetData Graph Digitizer (http://getdata-graphdigitizer.com). The data collection was in accordance with the Quality of Reporting of Meta-Analyses statement. We used the modified Jadad Scale [28] and the modified Newcastle-Ottawa scale [29] to assess the quality of RCTs and cohort studies, respectively.

Statistical Analysis
All analyses were performed using the STATA 11.0 package (StataCorp, College Station, TX, USA). HR with 95% CI was used for TTP and OS as demonstrated by Parmar MK et al [30]. For binary data, including ORR, progression-free probability, survival probability and toxicities, the odds ratio (OR) with 95% CI was used. HR > 1 reflects more deaths or progression in the S-1-based arm. OR > 1 reflects a favorable outcome in the S-1-based arm for response, survival probability or an unfavorable outcome for toxicities. Fixed-effects model was used and we then used random-effects model to confirm all the results. Cumulative meta-analysis is performed to sort out the time-tendency of outcomes and meta-regression is performed to explain some heterogeneity. Subgroup analyses were conducted by potential confounding factors selected by reviewing the characteristics of included studies. Sensitivity analyses were conducted to assess the stability of the result. P < 0.05 was considered significant. Heterogeneity was assessed by I 2 inconsistency test and χ 2 -based Cochran's Q statistic test in which I 2 >50%, or P < 0.05 indicated significiant heterogeneity. Publication bias was detected by graphical funnel plots. Asymmetry of the funnel plot was tested by Begg's test and Egger's test [31,32]. P < 0.05 was considered as significant. This article follows the QUORUM and the Cochrane Collaboration guidelines (http://www.cochrane.de) for reporting meta-analysis and accords with the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines (Checklist S1).

Eligible Studies
Detailed steps of the search are shown (Figure 1). After the selection procedure, five RCTs [22,23,[33][34][35] and five cohort studies [20,21,24,25,36] were included, with a total of 423 patients in S-1-based arm and 398 patients in capecitabinebased arm (Table 1). There were no significant differences in the baselines between S-1-based arm and capecitabine-based arm in these studies, as reported. All the RCTs were considered to be of high quality (Table 2) and the included cohorts showed satisfactory quality with reasonable selection criteria, comparable patient characteristics and adequate follow-up of the subjects (Table 3).
Subgroup analysis according to study design found consistent result in the RCTs (OR 0.97, 95% CI 0.62-1.51, P = 0.88; I 2 = 0%) with the overall effect. Similarly, consistent result
In detail, meta-analysis of 3 RCTs also showed no significant differences between two arms (3-month OR 0.84, 95% CI high † There are four items in the Jadad scale: randomizations, allocation concealment, double blinding, withdrawals and dropouts. If the item was not described in the study, the score would be 0; otherwise it was 1. And if the method of the item was described and it was appropriate, the score would reach to 2 except for the item of withdrawals and dropouts. Randomized control trials (RCTs) were considered to be of high quality if the score was 4-7, of low quality if the score was 1-3.

Overall Survival
Median OS was demonstrated in eight studies which ranged from 7.8 to 13.8 months for S-1-based arm, and from 8.1 to 13.5 months for capecitabine-based arm. The pooled HR for OS of 6 studies showed no significant difference between the Table 3. Modified Newcastle Ottawa quality assessment scale for cohort studies † .

Study Selection Comparability Outcome
Representa-tiveness of the exposed cohort Selection of the non-exposed Ottawa scale is for case-control study and cohort study.
Selection: (1) Representativeness of the exposed cohort: A, truly representative of the average patient with S-1 regimen; B, somewhat representative of the average patient with S-1 regimen; C, selected group; and D, no description of the derivation of the cohort (2). Selection of the non-exposed cohort: A, drawn from the same community as the exposed cohort; B, drawn from a different source; and C, no description of the derivation of the non-exposed cohort (3). Ascertainment of exposure: A, secure record; B, structured interview; C, written self-report; and D, no description (4). Demonstration that outcome of interest was not present at the start of the study: A, yes; B, no; C, no description.
Comparability: Comparability of cohorts on the basis of the design or analysis: A, study controls for co-morbidities; B, study controls for additional risk factors (such as age, or severity of illness, etc,); C, not done.  In detail, meta-analysis of 3 RCTs showed no significant difference of OS between two arms (HR 0.97, 95% CI 0.84-1.13, P = 0.71; I 2 = 0%). Meta-analysis of 3 cohort studies also showed no significant difference (HR 1.09, 95% CI 0.79-1.49, P = 0.61; I 2 = 0%).

Heterogeneity, Regression Analysis and Publication Bias Assessment
No significant heterogeneity was found for all analyses (I 2 < 50%, P > 0.05), When fixed-effects model changed to randomeffects model for all comparisons, all the results remain.
Meta-regression analysis further found patient median age of either group was not significant contributor to between-study heterogeneity (P ranging from 0.283 to 0.876), without enough information for other string variables.
3-month progression-free probability showed borderline publication bias by Egger's test (P = 0.048), however, no publication bias by Begg's test (P = 0.089). There's no publication bias for other results, with a symmetrical appearance on funnel plot analysis and P ranging from 0.221 to 1 given by Begg's test and P ranging from 0.102 to 0.803 by Egger's test (Figure 7).

Subgroup Analysis and Sensitivity Analysis
Although no significant heterogeneity was observed in all the comparisons, we probe into detail results in subgroup analyses stratified by study design (RCTs or cohort studies), patient median age (patient median age ≤ 65 or > 65), chemotherapy regimen (single drug, combined with oxaliplatin, cisplatin or docetaxel), median of chemotherapy cycles (S-1< capecitabine, S-1≥ capecitabine) and country (Japan, Korea and China). All subgroup results were quite consistent with the overall results. The subgroup analyses according to study design were showed in forest plots, while all the other subgroup analyses were summarized in Table 5. No significant heterogeneity were found in all subgroup analyses, except in patient median age > 65 subgroup analysis of 3-month progression-free probability (I 2 =74%, P = 0.05). We observed Seol YM recruited oldest patients among the included studies. By excluding this study, heterogeneity was reduced and the conclusion remained the same.

Cumulative Meta-Analysis
Provided time span of the available studies was considerable (from 2008-2013), a cumulative meta-analysis was encouraged to identify the time-tendency of outcomes by successively adding studies to the given result. For ORR, TTP, OS, progression-free probability and survival probability, cumulative meta-analysis consistently and stably showed equivalent effects of S-1-based chemotherapy versus capecitabine-based chemotherapy since the several initial studies were pooled, which also showed the range of 95% CI became narrower and the HR and OR were gradually closer to 1 (Figure 8).

Discussion
As the first meta-analysis to compare S-1-based chemotherapy and capecitabine-based chemotherapy in the field of all cancers, this current meta-analysis demonstrated equivalent efficacy of the two chemotherapies as first-line treatment for AGC. A cumulative meta-analysis supported this result and suggested that the findings were robust with time. Meanwhile, equivalent tolerance was observed between two chemotherapies with regard to all the grade 3 to 4 toxicities except that hand-foot syndrome was significantly less prominent in the S-1-based chemotherapy.
As an important measure of anti-tumor efficacy, ORR saw equivalence of S-1-based chemotherapy and capecitabinebased chemotherapy, which had a high degree of consistency with each included study. The pooled ORR (38.3%) and the ORR (20.0% to 50.0%) of S-1-based chemotherapy in the included studies were within the range of the results in the onearm phase 1/2 clinical trials of S-1-based chemotherapy (40.0% to 54.0%) [37][38][39][40][41][42], the S-1-based arm of RCTs including FLAGS trial (29.1%) [17], SPIRITS trial (31%, 54%) [43] and the S-1-based arm of the meta-analysis comparing S-1 and 5-FU (31.3%) [19]. The dose intensity of S-1 in our included studies ranged from 168 to 560 mg/m 2 /week, between that of FLAGS trial (262.5mg/m 2 /week) and SPIRITS trial (doublet arm, 336 to 504 mg/m 2 /week) in most cases, conforming to that the ORR here is between that of FLAGS trial and SPIRITS trial (doublet arm). Similarly, the pooled ORR (39.1%) and ORR of capecitabine-based therapy (13.3% to 55.0%) were consistent with those in the one-arm phase 1/2 clinical trials of capecitabine-based chemotherapy (23.5% to 62.2%) [44][45][46][47], the capecitabine-based arm of RCT including ML17032 trial and REAL-2 trial (35% to 48%) [12,13], and the capecitabine-based arm of the meta-analysis comparing capecitabine and 5-FU (45.6%) [14,15]. The regimen of the REAL-2 trial was capecitabine (7000mg/m 2 /week) plus platinum (cisplatin 20 mg/m 2 /week or oxaliplatin 43.3 mg/m 2 / week) plus epirubicin. Although none of the 10 included studies in our meta-analysis were triplet therapy, the intensity of capecitabine-based chemotherapy in our meta-analysis was within the range of previous studies. Eight included studies had same dose intensity of doublet partner or single drug in the two arms, while 2 studies has less dose intensity of cisplatin as S-1 partner than as capecitabine partner (Lim Do H et al [20] and Shitara K et al [24]). Both studies published equivalent ORR (42% vs. 38.6%; 43.2% vs. 50%), median OS (13.3m vs. 11.2m; 13.8m vs. 13.4m) even though the dose for S-1 partner was relative less. A meta-analysis of the two studies and the sensitivity analysis after excluding the two generated same results.
The pooled HR showed comparable OS of these two therapies (HR 0.99, 95% CI 0.87-1.13) which slightly favored S-1-based therapy. As the most clinically meaningful measure of treatment effect for cancer, the impact of first-line therapy on OS may be confounded by the second-line therapy. We further accessed the chance of second-line chemotherapy and found patients with first-line S-1-based regimen received second-line chemotherapy more frequently than those with first-line capecitabine-based regimen (OR 1.82, 95% CI 1.18-2.82; I 2 = 0%). When we omitted the study with the high odds ratio of second-line chemotherapy in S-1-based regimen versus  capecitabine-based regimen by Kim GM et al [22], or the study by Shitara K [24] with the high proportion of second-line chemotherapy, the comparable OS remained. Another important factor influencing OS is follow-up time. By reviewing the included studies, we found most of patients had passed away when follow-up ended and it indicated the follow-up was enough. TTP and PFS were surrogate measures for efficacy. The pooled HR for TTP and the pooled HR for TTP and PFS (TTP_PFS) [48] showed equivalent results of the two chemotherapies, with HR = 0.98 for both analyses slightly favoring S-1-based therapy. Further complementally, all equivalent 0.5-year, 1-year, 2-year overall survival probabilities, and 3-month, 6-month progression-free survival probabilities which were consistent with the case in each included study reinforced the comparable efficacy of the two therapies. Whether the hand-foot syndrome, thrombocytopenia, and stomatitis were more frequent for capecitabine is controversial, so is whether diarrhea occurred more for S-1 [20,23,24,33,35,36]. Our study did find a significant prominent of grade 3 to 4 hand-foot syndrome in capecitabine-based therapy versus S-1-based therapy, however, equivalent tolerance was found with regard to other grade 3 to 4 hematological and non-hematological toxicities. All 10 included studies reported the toxicities of both chemotherapies were relatively tolerable and manageable. Four studies reported decreased dose of capecitabine was largely due to hand-foot syndrome [22,23,25,35] and two studies due to hematological toxicity [22,24], however, demonstrated most patients could continue capecitabine-based therapy until progression after dose modification. The rate of 5.6% here for grade 3 or 4 handfoot syndrome for capecitabine-based chemotherapy (based on Asian studies) was relatively lower than previous report for Westerners (11-17% in Westerners) and ethnic differences may help explain [49]. Grade 3 or 4 diarrhea was uncommon and no difference was between two chemotherapies in this meta-analysis (3.2% vs 3.6%, P = 0.68). Dose modification occurred unusually for S-1 and 3 studies did report due to hematological toxicity and none due to diarrhea [21,22,24]. Literature showed diarrhea was the main dose-limiting toxicity of S-1 in Westerners due to higher activity of cytochrome P-450 2A6 enzyme systems [43]. Although the dose of S-1 was reduced mainly due to diarrhea in West compared to the dose in Asia, global phase 3 FLAGS trial reported non-inferior results regarding OS between S-1 plus cisplatin and 5-FU plus cisplatin [50]. That indicated the promising value of S-1 for Westerners after careful evaluation and adjustment. In subgroups stratified by potential confounders (regimen, age, median cycles, study design, country), equivalence of efficacy was found and was quite consistent with the overall results. This meta-analysis included 10 studies that were 3 studies of single drug, 2 studies with oxaliplatin, 4 studies with cisplatin, and 1 study with docetaxel according to regimen. Based on SPIRITS trial [43], phase 2 trials of S-1 plus oxaliplatin [39], study of capecitabine plus cisplatin and REAL-2 study [12,51], combination chemotherapy with an oral fluoropyrimidine (S-1 or capecitabine) plus a platinum (cisplatin or oxaliplatin) showed advantage over monotherapy and has been recognized as standard chemotherapy for advanced gastric cancer all over the world [52,53]. Although triplet therapy, which contains a fluoropyrimidine, a platinum, with an anthracycline or a taxane in the West, has demonstrated better or non-inferior efficacy than doublet therapy, its usage was restricted because of substantial toxicities [54,55]. That all studies included in our meta-analysis were either doublet chemotherapy or monotherapy reflected the above current situation, especially in Asia. In this background, our results delivered quite meaningful value for non-inferior S-1 plus platinum versus capecitabine plus platinum. In the subset of patients with median age > 65, 3 studies used recommendedintensity single drug (S-1, 373.3, 466.7, 560 mg/m 2 /week; capecitabine, 11666.7 mg/m 2 /week), 1 study used reducedintensity S-1/capecitabine plus normal-intensity oxaliplatin (43.3 mg/m 2 /week). The intensity was relatively less compared to that of the median age ≤ 65 subset in which all used combination regimen. S-1/capecitabine as monotherapy or platinum plus reduced S-1/capecitabine are popular options for old patients with consideration for tolerance, especially on the basis of equivalent efficacy. Docetaxel is another promising combination partner of fluoropyrimidine, with advantage for S-1 plus docetaxel over S-1 showed in a phase 3 clinical trial [56]. A new clinical trial showed the non-inferior efficacy and toxicity of docetaxel plus S-1 than cisplatin plus S-1 [57]. Only one study about docetaxel was included in our meta-analysis, in which ORR, progression-free and survival probability were compared, but TTP and OS were not. Two meta-analyses showed docetaxel-containing palliative chemotherapy improved ORR with or without OS prolongation than non-taxanecontaining for AGC [2,58]. More studies about taxane (docetaxel/pacilitaxel) combined with S-1/capecitabine are expected.
Median number of cycles of first-line chemotherapy is one of prognostic factors for survival [59]. The median cycle ratio of S-1-vs. capecitabine-based chemotherapy ranged from 0.25 to 1.25 in the included studies. However, we noted the number of the days in a cycle differed across the included studies. Thus, when only the studies with same number of the days in one cycle for S-1-and capecitabine-based chemotherapy were considered, we still found the overall less cycles of S-1-vs. capecitabine-based therapy, which strengthened the noninferior efficacy of the former versus the latter. In the study by Lim do H et al (S-1 > capecitabine) [20], the dose intensity of S-1 (S-1, 168 mg/m 2 /week; cisplatin, 8.6~14.3 mg/m 2 /week) was relatively less than that of the capecitabine (capecitabine, 9333.3 mg/m 2 /week; cisplatin, 8.6~33.3 mg/m 2 /week). The above two factors guarantee the comparability of two chemotherapies and resulted in equivalent efficacy and tolerance.
Five cohort studies were included and they were of good quality with evaluation. In order to determine whether study design impact the results, we took subgroup analysis and found both RCT and cohort subsets demonstrated consistency     Strengths of the current meta-analysis are that it was a systematic retrieval and review of the medical literature, with comprehensive exploration in subgroup analysis and cumulative analysis. All heterogeneities were insignificant. Both the fixed model and random model were used and all the results remained. However, there're limitations in our analysis. First, as with any meta-analysis, the results were impacted by the quality of the included studies. Second, there included five RCTs and five cohort studies. The data from cohort studies might be biased and more RCTs are warranted. Third, only 1 RCT reported the study was designed by a 'pick the winner' format [22], however, none of the remaining 4 RCTs and 5 cohort studies reported whether it was conducted as a noninferiority or superiority study. Efficacy was studied as the primary endpoint and toxicities as the secondary endpoint in all 10 studies. Quality of life was only referred to in 1 RCT and 1 cohort study which reported no difference as while [22,36] and none studies talked about economic costs. Only 1 RCT demonstrated the sample size (129) was of the statistical power for primary endpoint [22], and whether the sample size  had enough power to capture the endpoints for the remaining 9 studies was unclear. That was a major weakness. Therefore, on the basis of non-inferiority efficacy, more RCTs should be expected as non-inferiority trials, looking differences in terms of toxicities, quality of life or economic costs, with adequate number of patients and statistic power to capture these aspects. Forth, this meta-analysis was based on clinical studies, not the translational research. Literature showed S-1 was better in patients with high dihydropyrimidine dehydrogenase, while capecitabine was reported to be more effective in high thymidine phosphorylase gastric cancer [60,61]. A randomized study of capecitabine plus cisplatin versus S-1 plus cisplatin for AGC is ongoing in Japan, focusing on translational research [62]. This trial is expected to provide more information for choosing S-1 or capecitabine considering different translational characters. Finally, the current results are based on Asian studies, which need confirmation in the West.
In conclusion, this meta-analysis indicated the S-1-based chemotherapy was associated with non-inferior antitumor efficacy and better safety profile, compared with capecitabinebased chemotherapy. We recommended S-1 and capecitabine can be used interchangeably for advanced gastric carcinoma, at least in Asia. Meanwhile, more high-quality randomized controlled trials and Western studies are needed to provide more information.