From 515 initially retrieved studies, 111 studies published since 1980 were identified and included in this systematic review and meta-analysis (Figure S1
, Table S1
). Four studies overlapped with four of the 111 studies, and were therefore excluded. In the same time period, 85 reviews regarding neoadjuvant therapy in pancreatic cancer were published. The number of published original and review articles increased steadily within the last 15 years ().
The 111 reviewed trials included 4,394 patients. Seventeen centers published more than 1 study, altogether accounting for 73 of the 111 studies (Table S1
). Eight of the 17 centers published 2 studies. The University of Texas M.D. Anderson Cancer Center (Houston, TX) and the Fox Chase Cancer Center (Philadelphia, PA) published 12 and 11 studies, respectively (Table S1
). Other centers such as University of Osaka (Japan), the University of Marseille (France), and the Duke University (Durham, NC) published 8, 6, and 5 studies, respectively. The potentially overlapping patient populations were difficult to calculate since the information of which patients were included in which analysis could not always be retrieved. Using the study periods, study protocols, and a conservative estimate, there was a maximum of 17% overlapping patient populations.
Of the 111 included studies, 78 studies were prospective and 33 retrospective. There were 15 phase I, 13 phase I/II, and 28 phase II studies, as well as 14 cohort studies and 41 case series. No phase III trials have been published so far. A systematic search of clinical trial databases for pancreatic cancer trials identified 17 neoadjuvant trials (all phase I–II trials) and 23 trials for non-resectable but non-metastatic pancreatic cancer, i.e. potentially neoadjuvant trials (Table S2
The 111 analyzed studies reported on PDAC that generally included pancreatic head, corpus, and tail tumors without separate analysis regarding tumor localization. The studies included a median (IQR) of 31 (19–46) patients (). Ten of the 111 studies included in addition to pancreatic cancer a few patients with other periampullary tumors (i.e. ampullary, distal bile duct, and duodenal cancer), without separate analysis of the different entities. In 84 studies (76%), it was explicitly stated that histological or cytological tumor diagnosis was obtained before therapy. The age of the included patients varied, as did its reporting. The median of reported age of the patients in the 94 assessable studies was 62.5 y and was similar in the analyzed groups (group 1: 62 y, group 2: 62 y).
Summary of included studies in the different defined groups.
Chemotherapy was applied as neoadjuvant treatment in 107 of the 111 studies (96.4%). Different combinations of chemotherapies/agents and dosages were tested, as 56 of the studies were phase I–II trials. The main agents were gemcitabine, 5-FU (and oral analogues), mitomycin C, and platinum compounds (). In the trials that used only one regimen (n
79), 43 (54.4%) were performed using 5-FU or its oral analogues. 5-FU monotherapy was given in 14 (17.7%) of the studies. Thirty-six (45.6%) of the studies used a gemcitabine-based regimen, and of those, 18 (22.8%) studies applied gemcitabine monotherapy. 5-FU and gemcitabine combinations were used in 3 studies. Several studies compared different schemes or agents. Five studies were performed comparing gemcitabine with 5-FU or capecitabine, two studies comparing gemcitabine with cisplatin, two gemcitabine with 5-FU/cisplatin, and another three gemcitabine with 5-FU/mitomycin C. A further 16 studies included different agents and combinations (some for only few patients) (Table S1
). Twelve trials included taxanes (docetaxel/paclitaxel) in different combinations or as monotherapy (n
3). Five of the 107 studies included antibodies or tyrosine kinase inhibitors (bevacizumab, cetuximab, erlotinib) in the chemotherapeutic regimen. There were 44 studies using single agents (alone or in comparison) and 48 studies using combination therapies. In 15 studies both single agents and combination therapies were utilized.
Depiction of utilized chemotherapy and radiotherapy.
In 104 of the 111 studies (93.7%) patients received neoadjuvant radiotherapy. In three studies the exact radiation dose was not given. Doses applied ranged from 24 Gy to 63 Gy (). In 52 of the 104 studies that included radiotherapy the patients received doses between 45 and 50.4 Gy. In 14 studies different doses and radiation schedules were compared. Most patients received 1.8 Gy/fraction (50/104 studies), 2 Gy/fraction (15/104), or 3 Gy/fraction (10/104). In 13 studies intraoperative radiation (IORT) was applied with doses between 10 and 30 Gy. Since in most of those studies only few patients received IORT, this aspect was not further analyzed.
Data regarding treatment-related toxicity were available for 63 of 111 studies. For subsequent analysis, only severe (grade 3/4) toxicity (National Cancer Institute Common Toxicity Criteria; ctep.cancer.gov) was taken into account. Grade 3/4 toxicity for neoadjuvant therapy was estimated at 29.4% (CI 23.1%–36.1%) for all patients and was comparable for initially resectable (26.3%, CI 15.8%–38.3%) and patients with non- resectable tumors (“non-resectable tumor patients”) (31.1%, CI 22%–40.9%) (). Recent randomized controlled trials for adjuvant therapy report grade 3/4 toxicity rates for chemotherapy of 8.4%–22% (only neutropenia 
) and 14.7% (all toxicity 
). The reported grade 3/4 toxicity rates for radiochemotherapy were 9%–58% (only hematological toxicity 
) and 22.2%–79% (all toxicity 
Estimates of grade 3/4 toxicity of neoadjuvant treatment including the 95% confidence interval from the random effect model and number of assessable studies for each group (n).
Tumor response frequency for neoadjuvant chemo- and/or radiation therapy was evaluated in the different studies according to either radiographic or clinical response evaluation before exploration or histopathological response after resection. Six studies (5.4%) explicitly stated that the RECIST criteria 
were utilized. In 44 studies (39.6%) the criteria to assess tumor response were clearly stated, whereas in 61 studies (55%) criteria were either not clearly defined or not stated. For the whole study population the estimated fraction of patients with complete response was 3.9% (CI 3%–4.9%) () and with partial response 29.1% (CI 24.5%–34%) (). Stable disease was averaged to 43.9% (CI 37.9%–50%) in all patients and tumor progression under therapy occurred by estimation in 20.8% (CI 17.3%–24.6%) of the patients. Interestingly the pooled percentages did not vary much in the two groups of initially deemed resectable and non-resectable tumor patients (). Thus, complete/partial responses were 3.6%/30.6% and 4.8%/30.2% for groups 1 and 2, respectively; whereas progressive disease was estimated to 20.9% (CI 16.9%–25.3%) and 20.8% (CI 14.5%–27.8%) of primarily staged resectable and non-resectable tumor patients. Comparing tumor response frequencies for patients treated with mono chemotherapy (n
44) versus combination chemotherapy (n
48) revealed complete and partial responses of 2.2% (CI 1.3%–3.3%) and 25.8% (CI 20.2%–31.8%) versus 5.3% (CI 3.8%–7%) and 34.7% (CI 28.9%–40.9%) ().
Estimates of complete response percentages in patients following neoadjuvant therapy and re-staging including the 95% confidence interval from the random effect model and number of patients for each study (n).
Estimates of partial response percentages in patients following neoadjuvant therapy and re-staging including the 95% confidence interval from the random effect model and number of patients for each study (n).
Table 3 Estimates of exploration and resection percentages after neoadjuvant treatment and restaging, and estimates of patients with complete response/partial response, stable disease, and progressive disease including the 95% confidence interval from (more ...)
Table 4 Estimates of percentage of responses and resections in patients receiving mono chemotherapy versus combination chemotherapy groups including the 95% confidence interval from the random effect model and number of assessable studies for each group (more ...)
Exploration and Resection
Operations performed included explorative laparotomies, palliative bypass procedures, and curative resections, e.g. partial pancreatico-duodenectomies, distal pancreatectomies, and total pancreatectomies. Studies were analyzed for patients explored and resected after restaging. All 111 studies included data for resection.
Seven studies (6.3%) explicitly used the NCCN guidelines of resectability for non-metastatic pancreatic cancer 
. Forty-five studies (40.5%) clearly defined the resectability criteria assessing most often the vascular involvement or classified the resectability according to the maximal tumor dimension. In 59 studies (53.2%), resectability criteria were not clearly stated (e.g. judged by single surgeons or an interdisciplinary team) or not stated at all. In group 1 including the patients who were staged to be resectable before neoadjuvant treatment resectability estimated to 73.6% (CI 65.9%–80.6%) (, ), whereas in group 2 including the patients who were staged non-resectable before treatment the averaged probability for resectability was 33.2% (CI 25.8%–41.1%) (, ). As shown in , in the assessable studies the percentage of exploration for the entire group was 69.5% (CI 62.1%–76.4%) and 77.9% (CI 72.4%–82.9%) of these patients were resected. Of the patients deemed resectable before treatment, 88.1% (CI 82.9%–92.4%) were explored after restaging, and of those 85.7% (CI 78.9%–91.2%) could be resected. In group 2, 46.9% (CI 36.9%–57.1%) of the patients were explored. Of them, 69.9% (CI 61.2%–77.9%) could be resected successfully (). Interestingly the estimated fraction of R0 resections were comparable between patients in group 1 (82.1%; CI 73.1%–89.6%) and patients in group 2 (79.2%; 72.4%–85.2%) (). To analyze potential publication bias, funnel plots were created () that demonstrated heterogeneity (see below) but no considerable imbalance (no reasonable evidence for publication bias) neither for the group of patients with initially resectable tumors (“resectable tumor patients”) nor for the non-resectable tumor patients. There were three considerable outliers in the non-resectable group (). Omission of these trials in another supportive meta-analysis regarding resection rates demonstrated an estimated resection probability of 30% (CI 24%–36%), which was similar to the estimated proportion of 33% (CI 26%–41%) for the entire group of non-resectable tumor patients.
Estimates of resection percentages in patients following neoadjuvant therapy and re-staging including the 95% confidence interval from the random effect model and number of patients for each study (n).
Funnel plots for the resection rate for studies analyzing initially resectable (A) and non-resectable (B) tumor patients.
Analyzing resection frequencies for patients treated with mono chemotherapy versus combination chemotherapy revealed that in the group of initially resectable tumor patients, the averaged fraction of resections for patients receiving monotherapy was 80.8% (CI 66.1%–92.1%) and for combination chemotherapy 66.2% (CI 57.9%–74%). In contrast, in patients with locally advanced/unresectable tumors, resections were more frequent in the group of patients who received combination chemotherapy with 33% (CI 25.2%–41.3%) in comparison to monotherapy with 27.3% (CI 18.1%–37.5%) ().
Morbidity and Mortality
Data regarding morbidity and mortality following neoadjuvant treatment and pancreatic resection were presented in 50 and 85 of 111 studies, respectively. Perioperative morbidity was estimated at 34.2% (CI 28.3%–40.4%) for all patients (), which is within the range of reported morbidity data of 30%–55% for major pancreatic (head) resections 
. In-hospital mortality after neoadjuvant treatment and tumor resection was estimated at 5.3% (CI 4.1%–6.8%) for all patients (), which is at the upper limit of the 2%–5% mortality rates that have been reported in large series and surveys for major pancreatic resections at high volume centers 
. Interestingly, morbidity and mortality rates were estimated higher in the group of initially non-resectable versus resectable tumor patients (morbidity: 39.1% versus 26.7%, mortality: 7.1% versus 3.9%) ().
Estimates of morbidity and mortality in patients undergoing pancreatic resection following neoadjuvant therapy including the 95% confidence interval from the random effect model and number of assessable studies for each group (n).
Estimates of population median survival times were calculated as described and are provided with ranges from evaluable studies. Survival times for the individual studies were calculated from the time of diagnosis/start of neoadjuvant therapy in 47 trials and from surgery/resection in 4 trials. In 60 studies no detailed information regarding survival or survival calculations were provided. The longest median survival (23.3 mo, range 12–54 months) was estimated for the group of initially staged resectable tumor patients who were resected after neoadjuvant treatment (, ). The initially non-resectable staged patients reached an estimated median survival of 20.5 (range 9–62) mo following resection. The estimated median survival for the entire group of resected patients was 22.4 (range 9–62) mo. As expected, the median survival of the entire group of patients who did not undergo resection was shorter with 9.5 (range 6–21) mo. The patients who were initially classified as resectable and did not undergo resection after pretreatment survived an estimated median of 8.4 (range 6–14) mo, compared to 10.2 (range 6–21) mo of patients initially diagnosed as unresectable who did not undergo resection. Estimated 1- and 2-y survival probabilities for resected patients in group 1 were 77.9% and 47.4% and for group 2 79.8% and 50.1% ().
Estimates of median survival times (mp) in months and survival probabilities.
Summary overview of survival and resection percentages of different groups of patients with pancreatic cancer.
Analysis of Heterogeneity and Quality Assessment
The results of meta-regression, particularly the amount of explained heterogeneity (variance components), are summarized in . For the purpose of sensitivity analysis, particularly to investigate sensibility of results with regard to model considerations (prospective under- and/or over-fitting), both univariable and multivariable models (simultaneously including all potential explanatory factors) were employed. In total, 17 institutions could be identified which contributed more than one study to the total number of trials. There were two institutions with more than 10 trials considered in our systematic review and meta-analysis. The results of the multivariable meta-regression analysis revealed that the amount of heterogeneity could be explained from about 13% to 35% by differences between institutions (considered as random effect variable). The highest impact of centers was observed regarding toxicity (34.8%) despite simultaneous consideration of chemotherapy, which corresponds to the next highest component of total variability at least in the multivariable analysis. Study design showed some impact on response evaluation (explained variability: up to 11%) and morbidity (up to 9%). Mean age of patients and study period are considerable explanatory variables for heterogeneity in morbidity and in-hospital mortality with an estimated amount of explained heterogeneity of about 10%. Heterogeneity of resection and exploration rates could mainly be deduced to resectability as well to differences between institutions and there was no sufficient explanation supported by the other potentially influencing variables. In general, the results of univariable and multivariable heterogeneity analysis were quite comparable. However, some range in attributable source of outcome variability was apparent for resectability (concerning resection and exploration rates), institution (concerning response rates), and chemotherapy (concerning complete response rates and toxicity) (). Quality assessment according to the GRADEprofiler regarding toxicity analysis, response evaluation, resection and exploration rates, morbidity, mortality, and survival analysis is presented in .
Table 7 Multivariable meta-regression analysis for different variables as indicated and described in the Methods section.