Survival benefits of neoadjuvant chemo(radio)therapy versus surgery first in patients with resectable or borderline resectable pancreatic cancer: a systematic review and meta-analysis

Background Pancreatic adenocarcinoma is a highly lethal malignancy. Neoadjuvant chemo(radio)therapy [NAC(R)T] is recommended to use for borderline resectable pancreatic cancer (BRPC) and high-risk resectable pancreatic cancer (RPC), but no high-level evidence exists. Methods We searched PubMed, EMBASE, Web of Science, and Cochrane library to identify trials comparing survival data of NAC(R)T with SF for RPC or BRPC. Overall survival (OS) was synthesized in analysis of all the patients (intention-to-treat [ITT] analysis) and resected patients respectively. Results The meta-analysis included 17 trials with 2286 participants. For BRPC, NAC(R)T improved OS both in ITT analysis (HR, 0.49; 95% CI, 0.37–0.65; P < 0.001) and in analysis of resected patients (HR, 0.66; 95% CI, 0.51–0.85; P = 0.001) in comparison to SF, accompanied with comparable overall resection rate [odds ratio (OR), 0.69; 95% Cl, 0.41–1.16; P = 0.159]. Disease-free survival, R0 rate, and recurrence were also in favor of NAC(R)T. For RPC, OS in analysis of resected patients was higher with NAC(R)T (HR, 0.75; 95% CI, 0.63–0.89; P = 0.001), but OS in ITT analysis was similar (HR, 1.02; 95% CI, 0.85–1.22; P = 0.818). The overall resection rate (OR, 0.50; 95% Cl, 0.25–0.99; P = 0.048) was lower, but R0 rate was higher with NAC(R)T. No differences in disease-free survival and recurrence between NAC(R)T and SF. Survival benefits of NAC(R)T basically persisted across sensitivity and subgroup analyses. Conclusions This meta-analysis demonstrates that NAC(R)T can provide survival benefits in BRPC patients and a subgroup of RPC patients compared with SF. Future research should focus on investigating the potential biomarkers to screen the subgroup of RPC patients who can benefit from neoadjuvant therapy. Trial registration CRD42018103086.


Introduction
Pancreatic ductal adenocarcinoma (PDAC) portends an overall poor prognosis and is expected to become the second lethal malignancy in the USA by 2030 [1,2]. Although surgery remains the only curative-intent treatment for PDAC, the management based on surgery first (SF) has not substantially improved the survival of patients with potentially resectable disease over the past two decades, even after the effort of adjuvant therapy (AT) [2][3][4]. The main reason is the early recurrence caused by micrometastases that were not undetected before surgery [3,5,6]. Based on these clinical evidence together with other preclinical evidence, PDAC even in early stage, analogous to breast cancer, should be recognized as a systemic disease [2,7,8]. Recently, neoadjuvant chemo(radio)therapy [NAC(R)T] is proposed as a new therapeutic strategy for early systemic treatment to increase completeness of resection (R0 rate) and control systemic micrometastases [3,9]. The newest National Comprehensive Cancer Network (NCCN) guidelines, version 2.2018, recommended NACRT for the management of borderline resectable pancreatic cancer (BRPC). Also, NACRT is considered to be used in high-risk resectable pancreatic cancer (PRC). However, the recommendation of NCCN guidelines lacks high quality evidence [10,11]. It is controversial for the application of NAC(R)T to RPC or BRPC in the real world, particularly in RPC, which is still intensely discussed at the European Society for Medical Oncology (EMSO) World Congress on Gastrointestinal Cancer 2019. Although there are several randomized controlled trials (RCTs) indicating NACRT increases survival in resectable or borderline resectable PDAC, the trials are limited by small sample sizes [9,12]. It is still necessary to pool the existing studies to perform a meta-analysis. Indeed, some scholars have done relevant meta-analyses, but most of them are single-arm meta-analyses, such as a recent meta-analysis by Versteijne et al. that lack direct comparison and ignore interstudy heterogeneity [11,13,14]. Other published meta-analyses did not focus on survival benefits [15]. Additionally, it is a fact that the definition of RPC and BRPC has undergone several changes over time, which leads to the existence of mixture of RPC and BRPC in the population of included studies according to current standard of resectability status. From this point of view, interstudy heterogeneity exists in all previous meta-analyses.
Hence, we only included comparative studies and reclassified the population as RPC, RPC/BRPC, and BRPC in each study on a basis of the criteria of resectability status in the NCCN guidelines version 2.2018 and conducted this meta-analysis to compare survival benefits of neoadjuvant chemotherapy with or without radiotherapy [NAC(R)T] to SF with or without AT for patients with RPC or BRPC.

Material and methods
This meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [16]. The protocol for this metaanalysis is registered at PROSPERO (CRD42018103086).

Search strategy
A systematic literature search of online database including PubMed, Web of Science, EMBASE, and Cochrane library was performed for published articles from the inception dates to January 10, 2019. The combination of heading terms and keywords were used to search comprehensively and precisely. The relative terms were as follows: "pancreatic neoplasms," "surgery," "resection," and "neoadjuvant." The language of articles is limited to English. A detail description of the search is available in Additional file 1: Table S1. Besides, we also reviewed the references of included studies to identify additional literatures.
Study selection, data extraction, and quality assessment Two independent investigators (L.P., J.F.) screened articles according to the inclusion and exclusion criteria (Additional file 1: Table S2). The same two researchers independently extracted data and evaluated methodological quality of articles, using a Microsoft excel database to record all available data. For quality assessment, RCTs and nonrandomized comparative trials (NRCTs) were respectively evaluated by utilizing the Cochrane risk of bias and the modified Methodological Index for Non-randomized Studies (MINORS) score (Additional file 1: Table S3) [17,18]. Any disagreement was resolved by another investigator (Y.F.W.).

Study definition and outcomes of interest
The definition of "borderline resectable" has changed over time and varies in the published literature. In present study, we use the definition of RPC and BRPC in the NCCN guidelines version 2.2018 (Additional file 1: Table  S4) to reclassify the study population in included trials as RPC, BRPC, and RPC/BRPC based on the detailed description in the included articles. In RPC/BRPC, the study population in trials mixed with RPC patients and BRPC patients. Stratified analysis (RPC + BRPC, RPC, BRPC) was performed, and RPC + BRPC group contains RPC patients, BRPC patients, and RPC/BRPC patients. The resectability status of PDAC in each of the articles included is discussed and confirmed by all the authors.
The primary outcomes were OS. The hazard ratio (HR) with 95% confidence intervals (CIs) for OS were obtained directly based on data from multivariate Cox proportional hazards regression models in the included literatures. If studies did not offer HRs and 95% CIs, the method provided by Tierney et al. was used to calculate HRs from Kaplan-Meier curves [19]. The second outcomes include 1-, 3-, and 5-year survival rates (1-, 3-, and 5-YSRs), disease-free survival (DFS), recurrence rate, overall resection rate, R0 rate, and pathological positive lymph node (pN+) ratio. The 1-, 3-, and 5-YSRs were acquired from Kaplan-Meier curves, if the studies did not present these data.

Statistical analysis
HRs and 95% CIs were estimated for OS and DFS using an inverse variance model to pool the data. The pooled odds ratios (ORs) with 95% CIs were estimated for dichotomous outcomes. Between-study heterogeneity was calculated using Higgins' I 2 and I 2 > 50% indicated significant heterogeneity [20]. A random-effects model was used to pool data when I 2 > 50%, while a fixed-effects model was chosen when I 2 < 50% [21]. The 1-, 3-, and 5-YSRs for NAC(R)T and SF were calculated by single-arm metaanalysis and were presented graphically using bubble plots. Various sensitivity analyses were conducted to observe the stability of results along with extraction of matched baseline characteristics of included trials: (1) matched patient factors, (2) matched tumor size, (3) matched vascular resection, (4) matched initial CA19-9 level, (5) matched tumor factors, (6) matched patient and tumor factors, (7) pancreatic head cancer (≥ 80% of patients), (8) matched AT, and (9) Asians. χ 2 tests and independent t tests were respectively used to identify matched baseline factors for dichotomous and continuous variables, if included studies did not provided relevant P value. To assess the effects of covariates on the pooled estimates, subgroup analysis and meta-regression analysis were conducted respectively. Publication bias was detected using funnel plots, Begg's tests, and Egger's tests [22]. Twosided P < 0.05 was considered as statistical significance. All statistical analyses were performed using STATA/SE version 15.0 (StataCorp LP, College Station, TX).

Overall survival
Firstly, we did an intention-to-treat (ITT) pooled analysis, which means both NAC(R)T and SF groups included patients who did not undergo surgery. Nine studies (11 data sets) [3,9,12,23,24,[33][34][35][36] (Fig. 3).  Figure 4b showed the mean 1-, 3-, and 5-YSRs following NAC(R)T and SF, in which the size of circles represents the number of cases in each study. For resected RPC + BRPC, the mean 1-, 3-, and 5-YSRs after NAC(R)T were 89%, 45%, and 24% and those after SF were 71%, 22%, and 13%, respectively. As for resectability status, similar trends were observed in RPC and BRPC.  were consistent with the evidence from primary outcomes analysis, except in RPC or BRPC with matched tumor sizes, matched tumor factors, and matched patient and tumor factors (P > 0.05). The eligible data sets with matched relevant factors above were insufficient (≤ 3), inevitably reducing the reliability of results.
Meta-regression analysis indicated that additional neoadjuvant radiotherapy had little effects on the pooled HR for OS comparing NAC(R)T with SF in all the patients and resected patients (all P > 0.05, Fig. 5b, c).

Discussion
This meta-analysis with 2286 (1082 vs 1204) patients only included comparative trials from 2011 to 2018 and mainly focus on survival outcomes between NAC(R)T and SF for resectable or borderline resectable PDAC. In ITT analysis, BRPC patients who underwent NAC(R)T have increased OS in comparison to SF while similar OS was observed between NAC(R)T and SF in RPC patients. In resected patients, NAC(R)T markedly increase OS, and 1-, 3-, and 5-YSRs compared to SF regardless of patients with RPC or BRPC.
Recently, there is one single-arm meta-analysis published in 2018 by Versteijne et al. [11] containing several single-arm trials besides comparative trials, which found that neoadjuvant treatment improved median OS by ITT analysis in resectable or borderline resectable PDAC (RPC + BRPC, 18.8 vs 14.8 months; BRPC, 19.2 vs 12.8 months). However, comparing with their study, our study only included comparative trials using HR to analyze the survival benefits between SF and NAC(R)T and found that NAC(R)T has no significant advantages in resectable PDAC in comparison to SF by ITT analysis (HR = 1.02, P = 0.818), which was consistent with their results (median OS in RPC, 18.2 vs 17.7 months).
For BRPC patients, a higher OS was shown in NAC(R)T group regardless of the analysis of all patients (HR = 0.49, P < 0.001) or resected patients (HR = 0.66, P = 0.001). Besides, patients who underwent NAC(R)T had higher DFS, lower recurrence rate, higher R0 rate, and similar overall resection rate compared with patients For RPC patients, NAC(R)T has a similar OS in ITT analysis but a higher OS in the analysis of resected patients compared with SF (HR = 0.75, P = 0.001). Moreover, RPC patients who underwent NAC(R)T had higher DFS and lower recurrence rate than those who underwent SF, although advantages did not reach statistical significance (DFS: HR = 0.80, P = 0.137; recurrence rate: OR = 0.77, P = 0.131). Also, R0 rate in NAC(R)T is higher than SF (NAT, 89%; SF, 78%; OR 1.95, P < 0.001), but overall resection rate in NAC(R)T is lower than SF (NAT, 66%; SF, 81%; OR 0.50, P = 0.048). Our study indicated that there may exist a subgroup of RPC patients who are sensitive to chemo(radio)therapy and can obtain survival benefits from neoadjuvant therapy. Therefore, looking for potential biomarkers Furthermore, it is a pity that the eligible data sets are so insufficient that we are unable to compare OS of patients who received NAC(R)T followed by resection with those who received SF followed by AT (SF + AT). Mokdad et al. [10] using a national cohort from National Cancer Database (2006Database ( -2012 found that the survival benefits were maintained in the NAC(R)T group in comparison with SF + AT for resected RPC patients (HR, 0.83 [95% CI, 0.78-0.89]). Similar result was also found by Parmar et al. [38] using the data from Surveillance, Epidemiology, and End Results database for resected RPC patients without vascular invasion (HR, 0.54 [95% CI, 0.40-0.72]). However, the only one RCT performed by Jang et al. [9] has reported that there was no significant difference in OS between NAC(R)T group and SF + AT group for resected BRPC patients (HR 0.67 [95% CI, 0.30-1.52]) with a total of 29 patients (17 vs 12). Given that the small sample size in the study by Jang et al. [9], we consider the trend is the same but the 95% CI is wide. Moreover, although the additional chemotherapy after surgery has been shown to improve OS, the implementation of AT is limited by performance status of patients, postoperative complications, and early disease progression [40][41][42]. Of course, AT is still recommended after NAC(R)T followed by resection as long as patients can tolerate postoperative chemotherapy [40].
For patients with RPC or BRPC, vascular resection with concomitant reconstruction is widely used to attain negative margins during the pancreatic resection. Currently, pancreaticoduodenectomy combined with venous resection is proved to be safe and feasible and has the same long-term survival if R0 resection can be achieved   [43][44][45]. Our subgroup analysis further found RPC + BRPC patients with a higher baseline proportion of vascular resection tended to show more survival benefits for NAC(R)T over SF (> 75% of vascular resection vs ≤ 35% of vascular resection; HR, 0.57 vs 0.77, respectively). Lee et al. [29] also found NAC(R)T achieved better survival outcomes than SF in RPC + BRPC with vascular resection. Therefore, NAC(R)T should be considered as a preferred therapeutic strategy for patients who may require vascular resection in the preoperative evaluation, especially in BRPC patients.
There are various chemoradiotherapy regimens in this meta-analysis, including multiple-agents chemotherapy (4 trials), combined single-agent chemotherapy and radiotherapy (8 trials), and combined multiple-agents chemotherapy and radiotherapy (5 trials), which is inherent heterogeneity in our study. Accordingly, the outcomes should be explained cautiously. At present, a number of RCTs are ongoing comparing survival benefits between neoadjuvant therapy based on more effective regimens and immediate surgery, which will provide more evidence about the role of neoadjuvant therapy in the treatment of RPC (NCT02172976, NCT02047513, and NCT02919787).
As for variation in the dose of radiotherapy, metaregression analyses were used to assess the effect of additional preoperative radiotherapy on the survival benefits and the result showed, relative to neoadjuvant chemotherapy alone, no significant differences in OS were found both in all patients and resected patients (all P > 0.05). Besides, the study by Cloyd et al. [46] also indicated that a high-dose (50.4 Gy) radiotherapy combined with chemotherapy was associated with similar OS in comparison with a standard dose (30 Gy) chemoradiotherapy or chemotherapy alone in patients undergoing pancreatectomy for PDAC in multivariate cox regression analysis. Meanwhile, several RCTs are in progress to investigate the survival benefits of different neoadjuvant regimens for the treatment of BRPC or RPC, contributing to determining the optimal chemoradiotherapy regimens (NCT02562716 and NCT03777462).
This study has several limitations. First, the majority of evidence in favor of NAC(R)T are based on NRCTs that increase the risk of potential selection and publication bias. However, considering that NRCTs usually have large sample sizes, a meta-analysis of RCTs is not necessarily superior to well-designed NRCTs in terms of evidence level [47]. In our study, all the included literatures were relatively high quality (modified MINORS score ≥ 12) indicating a low risk of bias. Besides, the between-study heterogeneity on most outcomes was low. Also, elaborate sensitivity, subgroup, and metaregression analyses had demonstrated the stability of outcomes. Second, heterogeneity exists in chemotherapy regimen and radiotherapy dose, as discussed previously, so results should be interpreted with caution. Therefore, more large-scale and well-designed RCTs with more effective regimens are needed to investigate survival outcome between NAC(R)T and SF in resectable PDAC.

Conclusions
This meta-analysis uses stratified analysis as well as sophisticated subgroup and sensitivity analyses to demonstrate that NAC(R)T can provide survival benefits in patients with BRPC and a subgroup of RPC in comparison with SF. Future researches should look for potential biomarkers to screen the subgroup of RPC patients who can benefit from neoadjuvant therapy.