Recurrence Patterns for Pancreatic Ductal Adenocarcinoma after Upfront Resection Versus Resection Following Neoadjuvant Therapy: A Comprehensive Meta-Analysis

Background: Neoadjuvant therapy (NAT) represents a paradigm shift in the management of patients with pancreatic ductal adenocarcinoma (PDAC) with perceived benefits including a higher R0 rate. However, it is unclear whether NAT affects the sites and patterns of recurrence after surgery. This review seeks to compare sites and patterns of recurrence after resection between patients undergoing upfront surgery (US) or after NAT. Methods: The EMBASE, SCOPUS, PubMed, and Cochrane library databases were systematically searched to identify eligible studies that compare recurrence patterns between patients who had NAT (followed by resection) with those that had US. The primary outcome included site-specific recurrence. Results: 26 articles were identified including 4986 patients who underwent resection. Borderline resectable pancreatic cancer (BRPC, 47% 1074/2264) was the most common, followed by resectable pancreatic cancer (RPC 42%, 949/2264). The weighted overall recurrence rates were lower among the NAT group, 63.4% vs. 74% (US) (OR 0.67 (CI 0.52–0.87), p = 0.006). The overall weighted locoregional recurrence rate was lower amongst patients who received NAT when compared to US (12% vs. 27% OR 0.39 (CI 0.22–0.70), p = 0.004). In BRPC, locoregional recurrence rates improved with NAT (NAT 25.8% US 37.7% OR 0.62 (CI 0.44–0.87), p = 0.007). NAT was associated with a lower weighted liver recurrence rate (NAT 19.4% US 30.1% OR 0.55 (CI 0.34–0.89), p = 0.023). Lung and peritoneal recurrence rates did not differ between NAT and US cohorts (p = 0.705 and p = 0.549 respectively). NAT was associated with a significantly longer weighted mean time to first recurrence 18.8 months compared to US (15.7 months) (OR 0.18 (CI 0.05–0.32), p = 0.015). Conclusion: NAT was associated with lower overall recurrence rate and improved locoregional disease control particularly for those with BRPC. Although the burden of liver metastases was less, there was no overall effect upon distant metastatic disease.


Introduction
Patients who have undergone potentially curative treatment for pancreatic ductal adenocarcinoma (PDAC) typically suffer treatment failure with cancer recurrence [1], frequently within two years of resection [2]. Recurrence, including multi-site recurrence and liver metastases, are key prognosticators of overall survival [3]. Several risk factors for specific recurrence patterns have been identified. Positive resection margin and poor tumour differentiation [2,4] are risk factors for locoregional recurrence. Tumour size, elevated preoperative Ca19-9 serum level, and perineural invasion are risk factors for distant recurrence [5].
While upfront surgery followed by adjuvant therapy has been the standard of care for PDAC, this is being challenged by recent evidence with use of neoadjuvant chemotherapy (NAT) for borderline resectable pancreatic cancer (BRPC) [6] and resectable (RPC) [7] pancreatic cancer (LAPC) [8].
It is important to determine the impact of different treatment strategies on pattern of disease recurrence [9,10]. NAT is associated with a higher rate of achieving R0 margins [11,12] and may control micrometastatic disease; thus, it is important to determine whether NAT influences patterns of treatment failure in terms of the frequency and site of recurrent cancer.
While here have been meta-analyses examining the impact of NAT on overall survival [13,14], there is paucity of evidence examining the impact of NAT and upfront surgery (US) on disease recurrence and recurrence patterns with and without neoadjuvant chemotherapy.
This present study aimed to undertake a meta-analysis of available evidence to compare patients with PDAC undergoing upfront surgery versus neoadjuvant therapy followed by surgery to review sites of recurrence, recurrence-free survival patterns, overall recurrence rate, and time to recurrence in patients with resectable, borderline resectable and locally advanced pancreatic cancer.

Materials and Methods
The protocol of this review was registered on the Prospero database prior to inception (CRD42019159031).

Literature Search
The EMBASE, SCOPUS, PubMed and Cochrane library databases were systematically searched with the aid of a pre-devised literature search strategy that followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [15]. Details of the query string with accompanying Boolean operators are reported in Appendix A. Limits were set to English studies reported from January 2000 to February 2020 in human patients. Reference lists of included studies were then examined by two authors (C.B.B.R. and A.Y.S.) to identify eligible studies missed by the original search strategy.

Data Selection, Extraction, Risk of Bias Assessment and Publication Bias Assessment
Studies were included in the quantitative meta-analysis if they reported comparative recurrence outcomes specific to patients who underwent NAT compared to those undergoing US for PDAC. More specifically, if they reported overall recurrence, time to first recurrence, sites of recurrence and/or recurrence free survival (RFS). Studies were excluded if they were case reports/series, conference abstracts, study protocols, non-comparative observational cohorts, did not report recurrence or RFS data for either NAT or upfront surgery patients, did not employ NAT, and articles published prior to 2000. Title, abstract and full text screening were performed by two investigators (C.B.B.R. and A.Y.S.). Consistency of data extraction was compared, and disputes were adjudicated by a third author (S.P.). Baseline characteristic data including: age; gender; location of tumour; preoperative tumour stage and resectability; details of NAT data; preoperative radiotherapy; adjuvant therapy; pathological outcome data including; nodal status; tumour differentiation; tumour size; perineural/lymphovascular invasion; R0 resection; and patterns of recurrence including overall recurrence rate, time to first recurrence, site of recurrence, and RFS. The primary outcome was site-specific recurrence. The secondary outcomes were overall recurrence, time to first recurrence, and recurrence-free survival.

Assessment of Study Quality
The risk of bias and study quality assessment was performed through use of the methodological index for non-randomized studies (MINORS) grading criteria [16] for non-randomised studies. The Cochrane's risk of bias tool [17] was used for randomised controlled trials. Publication bias was identified through visual inspection of funnel plots whereby bias existed if individual studies lay beyond the funnel limits.

Terminology and Definitions
NAT included preoperative chemotherapy with or without radiotherapy. Sole radiotherapybased NAT regimens were not assessed in this meta-analysis. Fluoropyrimidine-based chemotherapies included S-1 (5-fluorouracil prodrug tegafur with oteracil, and gimeracil) [18], capecitabine and 5-fluorouracil; platinum-based chemotherapies included cisplatin and oxaliplatin; and plant alkaloids included docetaxel and paclitaxel. FOLFIRINOX was a combination regime of 5fluorouracil, leucovorin, oxaliplatin, and irinotecan. If an agent was used in combination with another, this was considered combination chemotherapy. Upfront surgery (US) included all forms of pancreatic resection (including pancreatoduodenectomy, distal pancreatectomy and total pancreatectomy) performed shortly after the diagnosis of a pancreatic cancer without preoperative NAT. The authors identified resectable (RPC), borderline resectable (BRPC), and locally advanced (LAPC) when reported as such. The definitions of these categories were variably reported and have changed over the course of the study with the publication of multiple definitions [19][20][21]. Nodal status was considered N0 when there were no positive nodes at the time of surgery. Tumour differentiation was based on histology of the specimen following resection [22] and tumour size by the largest diameter reported by the pathologist. Presence of malignant cells in neural tissue or lymphovascular tissue denoted perineural or lymphovascular invasion [23]. R0 resection was defined when there was no microscopic tumour present within 1 mm of the resection margin [24]. Recurrence patterns included overall recurrence rates, time to first recurrence and recurrence-free survival rates. Overall recurrence was defined by the first presentation of previously unknown disease activity during postoperative follow-up [25]. Locoregional recurrence was the presence of tumour recurrence at the site of previous resection or in surrounding lymph nodes (LN). Distant recurrence referred to recurrence present in distant organs. RFS rate was defined as the proportion of patients who had no confirmed recurrence of cancer and included the terms progression-free survival and disease-free survival [26]. Time to first recurrence represented the interval following resection that the first recurrence of disease was observed and was calculated by the mean RFS [27]. Disease-specific survival [28,29] or cancer-specific survival [8] were not included in this analysis as these terms focused solely on survival secondary to the cancer.

Statistical Analysis
The meta-analysis was conducted in its entirety with the packages: tidyverse [30], meta [31], metaphor [32], and MetaAnalyser (Jack Bowden and Christopher Jackson, UK) [33] in R project (R Foundation for Statistical Computing, Austria 2014). A Mantel-Haenszel model was utilised to perform the pairwise meta-analysis. The output was reported as odds ratios (OR) or standard mean difference (SMD) with 95% confidence intervals (CI). I 2 test allowed the authors to determine statistical heterogeneity whereby a threshold of 50% suggested moderate heterogeneity and 75% indicative of substantial heterogeneity [34]. Weighted means and recurrence rate estimates with their corresponding 95% CI were calculated for age, tumour diameter, follow-up interval, and recurrence outcomes. Weighted means were calculated by the inverse variance method. Weighted recurrence rate estimates were determined by a random intercept logistic regression model [31]. Weighted estimates were only calculated when three or more datapoints were available. WebPlotDigitizer (Ankit Rohatgi, California, USA 2019) [35] was employed to extract raw data from reported Kaplan Meier survival curves if articles failed to provide raw numbers for RFS or time to first recurrence. The median (1 st and 3 rd quartile) time to first recurrence was only calculable from a Kaplein Meier curve if both survival curves crossed the 75% and 25% survival threshold. Mean estimates were then derived from median and ranges or quartiles through Wan et al. [36].
Propensity score matched data was used in preference to raw unmatched data to improve homogeneity of the datasets in relevant studies [8,27,37,38]. However, if baseline characteristic data or outcome data were not reported for propensity score matching, the unmatched data was used [28]. Historic [25] or unpublished [39] US cohorts were utilised if baseline characteristics were reported. Data for baseline characteristics was commonly reported for the entire study population irrespective of those who subsequently underwent resection. However, the remaining outcomes including pathological, resectability and recurrence data are reported for the population who ultimately underwent surgery and were subsequently followed-up. When calculating NAT agent-specific time to first recurrence, the authors included situations where the agent was used as the base NAT, used in combination with another agent or used in more than 30% of the cohort [40]. Recurrence outcomes were reported with significantly fewer patients than in the pathological and resectability outcomes in another study [41] and this is reflected in the meta-analysis.

Time to First Recurrence
NAT was associated with a significantly longer weighted mean time to first recurrence (18.8 months, compared to US (15.7months, in nine studies [6,26,27,37,[39][40][41]45,47] (OR 0.18 (CI 0.05-0.32), p = 0.015, Figure 5B). Fewer than three studies reported resectabilityspecific time to first recurrence in both the US and NAT groups and no study reported organ-specific time to first recurrence. The weighted time to first recurrence for each specific NAT agent ranged from 18.5-22.9 months ( Figure 3D). The weighted time for each individual agent was calculated from studies using that specific NAT as the base or in combination with another agent. Two studies were excluded in the meta-analysis of time to first recurrence as outliers significantly skewed mean estimates for the US cohort [51] and the significant discrepancy in cohort numbers made the output uninterpretable [25].

Heterogeneity and Risk of Bias
Moderate heterogeneity was found in the outcomes to assess locoregional and distant recurrence rates. Moderate heterogeneity was further reported for liver recurrence and 2-and 5-year RFS. The pathological outcomes of tumour diameter, rates of lymphovascular involvement, and N0 nodal status also observed marked heterogeneity in the dataset. Overall, the non-randomised studies scored poorly in the MINORS criteria (Median 13 . Significant deficiency with regards to unbiased prospective data collection, blinded assessment of outcomes, and power calculations were seen globally (Table S5A). Many studies also failed in the domain of equivalent baseline characteristics (n = 12) [6,25,26,28,39,[41][42][43]46,50,52,54]. Cochranes risk of bias tool also identified obvious deficits in the three randomized controlled trials (RCTS) [7,44,47] in the domains of blinding of participants and outcomes, which is an obvious consequence of the nature of the disease (Table  S5B). Significant publication bias was observed on visual inspection of Forest plots for the outcomes of tumor diameter, N0 nodal status, and lymphovascular invasion ( Figure S8).

Discussion
Recurrence after pancreatic resection for PDAC is frequent with 20% of patients developing recurrence within 6 months and a further 40% in the first year of resection in spite of a margin-free resection [56]. Even among those who complete adjuvant therapy, median overall survival remains less than 20 months [57,58]. Recent chemotherapeutic regimens such as FOLFORINOX and Gemcitabine plus Nab-paclitaxel have shown favourable results in improving resectability even in patients with LAPC with resection rates ranging between 0-40% [59]. However, there is paucity of level 1 evidence regarding the role of NAT on recurrence patterns for resectable and BRPC. Two recent RCTs explored the role of NAT in resectable and BRPC. Jang et al. in a recent phase 2/3 RCT comparing neoadjuvant chemo radiation with gemcitabine versus upfront resection in patients with BRPC showed comparable over all recurrence rates and recurrence patterns. On the contrary, the Dutch PREOPANC phase III randomised controlled trial comparing preoperative chemo radiotherapy versus upfront surgery for resectable and BRPC showed improved disease-free survival in the NAT group although the time to recurrence and recurrence patterns were not fully evaluated. The present review of 26 studies including 4986 patients has shown that NAT was associated with lower overall recurrence rates (63% vs 74%), lower loco-regional recurrence rates (12 vs 27%), and particularly in patients with BRPC. NAT was associated with a lower liver recurrence rate (19.4 vs 30.1%) and improved recurrence-free survival at 2 years (40.5% vs 24.9%) and 5 years (22.2% vs 13.4%). Furthermore, on analysis of resection specimens, NAT appears to reduce rates lymphovascular (43% vs 52%) and perineural invasion (53% vs 80%) and increase the number of resections with N0 lymph node status (58% vs 29%). These results suggest the superiority of neoadjuvant therapy over upfront resection in reducing loco-regional recurrences, overall recurrence, time to recurrence, and improving recurrence-free survival.
In the present review, subgroup analysis of recurrence rates in BRPC patients showed comparable overall recurrence rates for BPRC in NAT and US. This is despite improved locoregional recurrence with NAT in BRPC patients. This is in direct contrast to RPC where a significant overall recurrence benefit was observed. In theory, the effect on RPC is surprising, however, this finding is similar to the preliminary overall survival results from the Prep-02/JSAP-05 trial [60] in which NAT (gemcitabine and S-1) performed better than immediate surgery for resectable PDAC with no significant survival benefit in the BRPC group. Previous reports have highlighted significant discrepancies in the outcomes following NAT for BRPC, and many authors attribute this finding to the failure to reach a consensus with regards to a standardised definition [61,62]. The recently concluded PREOPANC trial showed superior overall survival (OS) after preoperative chemo radiotherapy for BRPC and no significant difference for resectable PDAC. Although multiple societies have endeavored to provide standardised and, more importantly, validated definitions for PDAC. Assifi et al. reported that 40% of patients diagnosed as BRPC using the AHPBA/SSO or SSAT definitions could be reclassified as RPC [63]. Indeed, the degree of vessel involvement [54] and the individual vessel involved [62] (portal vein, hepatic artery, and or superior mesenteric artery) are on their own predictors of disease-free survival and so variations in definitions will undoubtedly influence rates of recurrence. BRPC is known to have higher rates of failure to achieve R0 resection, higher rates of occult metastases and requires resection of more tissue [64]. BRPC may therefore represent a spectrum of disease rather than a single entity, which may account for the conflicting results noted in these groups of patients in the literature. The inclusion of this heterogeneity in the cohort may have contributed significantly to the lack of a survival advantage with NAT. NAT is generally tolerated much better and morbidity following NAT is considered independent of surgical morbidity [65]. Eighty-nine percent of patients successfully completed NAT and underwent surgery in the review cohort, significantly higher than previously reported, which may be contributed to by the high rates of RPC undergoing NAT. This is also far superior to the 66% completed rate for adjuvant therapy reported in the literature [66].
Although chemotherapeutic subgroup analysis was limited, time to first recurrence and the distribution of NAT-specific rates of organ-specific recurrence did not differ between agents. Gemcitabine was the most frequently utilised NAT regimen in this cohort employed in nearly 60% of patients. However, following the results of the PRODIGE-24 trial [67] that demonstrated superior disease free survival (DFS) in patients with FOLFIRINOX in comparison to Gemcitabine for adjuvant therapy, FOLFIRINOX has become common practice. Indeed, combined NAT regimens including Gemcitabine and FOLFIRINOX have shown a high response rate, and there is increasing evidence to suggest that multi-agent NAT may suppress the recurrence of distant metastases [68,69]; however, to date, no trial has been published comparing FOLFIRINOX to other options. In the current review, combination therapy was only employed in 23% of patients and 13% of patients were administered FOLFIRINOX, suggesting the outcomes of NAT in this review may be more reflective of other chemotherapeutic agents. The efficacy of FOLFIRINOX is the topic of current investigation. The interim results of ESPAC-5F (ISRCTN89500674), Phase II RCT comparing immediate surgery with neoadjuvant gemcitabine plus capecitabine or FOLFIRINOX or chemoradiotherapy in patients with borderline resectable pancreatic cancer showed a significant survival benefit with NAT. The currently ongoing PREOPANC2 trial (NTR7292, 2018-06-19), NorPACT-1 trial (NCT02919787) [70], SWOG S1505 and PANACHE01-PRODIGE48 trial (NCT02959879) [71] may provide further evidence to support NAT first strategy. Furthermore, additional radiotherapy to NAT remains uncertain [72,73]. Although some reports suggest a significantly reduced rate of lymph node metastases and local recurrence with neoadjuvant chemo-radiotherapy (NACRT) [38], the subgroup analysis from this review showed no significant advantage with NACRT. Radiotherapy also has the added risk of increased postoperative morbidity and may compromise surgical outcomes [38].
In the present review, NAT was also associated with reduced tumour diameter, lower rates of perineural invasion, and lymphovascular invasion, leading to higher rates of R0 resection. This may have contributed to the lower loco-regional recurrence rates in the NAT group. However, there were comparable rates of distant recurrence between NAT and upfront resection groups, and this may be contributed to by the lack of significant differences in lung or peritoneal recurrence rates between the two groups. Whether the improved pathological parameters would translate into long term reduction in recurrence rates and survival benefit is still a matter of debate. Even in patients achieving complete pathological response after NAT, recurrence often occurs in half the patients [74].
There remain several limitations to the methodology of this review. This is a review of largely retrospective articles with limited outcomes available for pairwise comparisons. There were obvious discrepancies in the definitions of BRPC and LAPC that would have added significant heterogeneity to the cohort. There also remains significant variability in the assessment of outcomes over time, including methods of assessment for liver metastases following the increasing availability of magnetic resonance imaging (MRI). This study also reported outcomes on solely patients who underwent resection following NAT, thereby excluding those who failed to complete NAT from the final analysis. An intention to treat analysis was abandoned due to the failure of most studies to report on the original cohort. Furthermore, there remains some selection bias in favor of NAT therapy for improving rates of liver metastases in view of the exclusion of NAT patients who ultimately failed to undergo resection. This may have been due to multiple factors including local disease progression or distant metastases. Patients were also recruited from various institutions around the world with differing perioperative care protocols. The preoperative functional status of patients in each arm were also poorly reported and were therefore underutilised to compare the cohorts at baseline. Propensity score matched data was used where reported, however, this represented less than 20% of the cohort. The study data was limited in its ability to report chemotherapy agent-specific outcomes, and FOLFIRINOX-based chemotherapy regimens were uncommon despite there being a current trend towards its use in the current day. Furthermore, the recurrence rate was reported with varying definitions from when it was calculated. Significant deficiency was also identified regarding prospective data collection, blinded assessment of outcomes, and power calculations. However, this is the first review to quantitatively report the comparative recurrence outcomes including patterns of recurrence for this complex cohort of patients.

Conclusions
This is the first quantitative analysis of recurrence patterns following NAT for PDAC. The review found NAT improved rates of overall recurrence in RPC with lower locoregional and liver recurrence. However, NAT did not appear to impact peritoneal or lung recurrence. NAT was also associated with a reduced time to recurrence and recurrence free survival at both 2 and 5 years. The review is limited by its largely retrospective dataset, and FOLFIRINOX-based chemotherapy regimens are poorly reflected in the included studies. We await the results of currently ongoing RCTs and large prospective cohorts to confirm the findings of this review and compare the relative efficacy of various NAT regimens.
Supplementary Materials: The following are available online at www.mdpi.com/2077-0383/9/7/2132/s1, Figure  S1. Forest plot showing rate of locoregional (A) and distant (B) recurrence in studies reporting solely either borderline resectable (BRPC) or resectable (RPC) pancreatic cancer for Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Odds ratios (OR) are shown with 95 percent confidence intervals (CI). Figure S2. Forest plot showing recurrence-free survival at A) two-years and B) five-years postoperatively following Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Odds ratios (OR) are shown with 95 percent confidence intervals (CI). Figure S3. Forest plot showing overall recurrence rates in articles reporting A) preoperative resectability and B) preoperative chemoradiotherapy subgroups following Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Odds ratios (OR) are shown with 95 percent confidence intervals (CI). Figure S4. Forest plot showing A) Liver, B) Lung and B) Peritoneal recurrence rates following Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Odds ratios (OR) are shown with 95 percent confidence intervals (CI). Figure S5. Forest plot showing mean tumor diameter in mm following Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Standard Mean Differences (SMD) are shown with 95 percent confidence intervals. Figure S6. Forest plot showing A) Poor Tumor differentiation at the time of surgery B) perineural and A) Lymphovascular involvement rates following Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Odds ratios (OR) are shown with 95 percent confidence intervals (CI). Figure S7. Forest plot showing A) N0 Status and B) R0 Resection rates following Neoadjuvant therapy (NAT) vs Up-front Surgery (US). A Manel-Haenszel random effects model with a Hartung-Knapp adjustment was used for the meta-analysis of all outcomes. A Sidik-Jonkman estimator was utilised for tau 2 . Odds ratios (OR) are shown with 95 percent confidence intervals (CI). Figure S8. Funnel plots for all meta-analysis outcomes including A) Tumor Size, B) N0 Nodal Status, C) Poor tumor differentiation, D) Perineural invasion, E) Lymphovascular invasion, F) R0 Resection, G) Overall recurrence, and H) Recurrence free survival. Table S1. Patient characteristics among all studies included within the meta-analysis. Table S2. Pathological outcomes in all studies within the meta-analysis. Table S3. Resectability and eoadjuvant therapy of all patients who underwent resection included in the quantitative analysis within this meta-analysis. Table S4. Results of the mixed-effects meta-regression to identify potential predictors of overall recurrence in the A) neoadjuvant therapy (NAT) and B) up-front surgery (US) groups. Variables were included in the mixed effects model if they were significantly different between NAT and US where differences were identified on meta-analysis and with an incidence of ≥ three for categorical variables. Table S5. A) The MINORS criteria [27] scores for all non-randomised studies included in the review and B) Cochranes Risk of bias tool for randomised trials.

Conflicts of Interest:
The authors declare no conflict of interest.