The Role of Enhanced Recovery after Surgery in Pancreaticoduodenectomy: A Systematic Review and Meta-Analysis

Pancreatic cancer (PC) is currently the twelfth most common cancer in the world [1]. Unfortunately, the incidence of this cancer is increasing rapidly, and by 2030, it is projected to be the second leading cause of cancer-related mortality [2]. Pancreatic surgery is a highly complex and technically challenging procedure that can cure the disease. However, it is still considered a high-risk surgery despite significant advances in surgical techniques, training, and perioperative care that have reduced mortality rates from almost 30% three decades ago to less than 3% today [3, 4]. Unfortunately, such surgery-specific complications as delayed gastric emptying (DGE), postoperative pancreatic fistula (POPF), and post-pancreatectomy hemorrhage (PPH) can still occur, resulting in a high morbidity rate of up to 60% [5, 6]. These complications can significantly delay recovery and become life-threatening if not adequately treated.

The critical role of surgical stress response to the pathogenesis of postoperative morbidity and the importance of multimodal interventions to attenuate it and enhance outcomes was proposed more than 30 years ago [7]. The stress response to surgery is intricate, involving metabolic and inflammatory changes via activation of the sympathetic nervous system, the hypothalamic-pituitary-adrenal axis, and the renin-angiotensin-aldosterone system [8]. Although this response is a natural way for the body to handle stress, it can impede recovery. The heightened organ demands during this reaction contribute to postoperative morbidity, potentially causing complications even after a successful surgery [7].

Enhanced Recovery After Surgery (ERAS) pathways are a series of evidence-based interventions to minimize surgical stress throughout the perioperative period. To achieve this, ERAS initiatives prioritize preoperative optimization and patient education, early removal of catheters and drains, and multimodal opioid-sparing analgesia with early oral intake and mobilization. These approaches have improved recovery quality and reduced hospital stay, complications, and costs [9].

The ERAS Society released its first pancreatoduodenectomy (PD) guideline in 2012 [10]. In 2019, an updated guideline with 27 recommendations was published [11]. While implementing all these guidelines can be challenging, research indicates that achieving a compliance rate of 70% or higher can result in better outcomes [12]. This can prove challenging in PD, especially concerning managing postoperative drains, nasogastric tubes, and oral feeding.

In this systematic review and meta-analysis, the literature published following the first publication of ERAS guidelines for PD is critically appraised to compare the safety and short-term outcomes of ERAS with standard care for adult patients undergoing PD. The primary outcome of interest is the incidence of complications, with secondary outcomes including minor and major complications, DGE, POPF, PPH, readmission rates, length of hospital stay (LOS), time to start adjuvant chemotherapy, and overall postoperative 30-day mortality.

Table 1 summarizes the inclusion and exclusion criteria for our meta-analysis.

The study was designed following the Preferred Reporting Items for Systematic Review and Meta-analysis Protocol (PRISMA) to address the research questions [13]. Furthermore, it was registered in PROSPERO (CRD42023432293) [14]. The search methods, eligibility criteria, and data extraction process were designed prospectively. No patient informed consent or IRB/Ethics Committee approval was required, as the current meta-analysis was based on published records.

Two authors (D.L. and A.D.) conducted a thorough search across five databases, namely MEDLINE, Scopus, Web of Science, Cochrane, and EBSCO, to identify studies that reported on the safety and effectiveness of ERAS in patients undergoing PD. We did not perform a registry search, nor did we search multiple databases. We did not search the grey literature or the “health data” on Google. We used the following terms, including synonyms in all potential fields: “ERAS” OR “enhanced recovery after surgery” OR “fast track recovery” OR “accelerated recovery” AND “open pancreaticoduodenectomy” OR “duodenopancreatectomy” AND “complications” OR “length of stay” OR “time to chemotherapy” OR “delayed gastric emptying” OR “postoperative hemorrhage” OR “post pancreatectomy fistula” OR “readmissions” OR “deaths” OR “mortality” in any possible combination and form. The search period extended from 2013 until June 2023. The last search in all databases occurred on the 1st of July, 2023. No search filters were used. The search formulas are described in online supplementary file 1 (for all online suppl. material, see https://doi.org/10.1159/000539785). The references of eligible studies were searched for additional relevant citations, and duplicates were manually removed.

Two review authors (D.L. and A.B.) were individually involved in the quality assessment. Any disagreement between the review authors was resolved after discussion with the senior author (E.A.). The risk of bias was assessed according to the Cochrane risk-of-bias tools RoB-2 and ROBINS (I) for randomized and observational studies [15, 16]. The assessment was performed at both the study level and the meta-analysis level. The results were visualized in traffic-light plots and weighted bar plots of the distribution of risk-of-bias judgments within each bias domain for the primary outcome using the online app Robvis [17]. The overall body of evidence was graded according to the GRADE recommendations based on the study design of all eligible studies, the risk of bias, inconsistency, indirectness, publication bias, the magnitude of effect, dose-response relationship, and screening for confounding factors [18].

The event incidence for each arm was pooled after a proportion meta-analysis. The two treatment arms were compared using the relative risk (RR) and its 95% confidence interval as the pooled estimate. A fixed- or random-effects model was fitted to the data according to statistical heterogeneity. Heterogeneity was studied using the Q test and the Higgins I² statistic. We searched for potential sources of heterogeneity after eyeballing the Baujat plots. The sensitivity analysis of our results was carried out by re-running our meta-analysis, having excluded one study at a time. A meta-regression and a subgroup analysis studied the effect of moderators (continent, study design, and type of surgery) on the overall effect. We detected potential sources of publication bias using the Beggs test. We used the fragility index and a cumulative analysis to study the robustness of our results. Likewise, the net benefit of the intervention was calculated using numbers needed-to-treat (NNT) based on the RR. The results were plotted using forest and funnel plots. The statistical analysis was carried out using an R statistical environment [19]. When relevant summary data were provided in median and range, we estimated the mean and standard deviations according to Shi et al. [20] and Luo et al. [20‒22]. Statistical significance was set at 0.05, and we used a continuity correction of 0.5 for complications associated with zero events.

Our current literature search identified 455 unique articles. After reading the title and abstract, we excluded 408 articles and sought the full text of the remaining 47 studies. We could not retrieve five articles, and after reading the full text of the gathered studies, we excluded 25 other irrelevant studies. We have also identified five records from citation searching. Ultimately, 22 articles formed the basis of our systematic review and meta-analysis [23‒44]. The study selection process is outlined in a PRISMA flowchart according to the PRISMA 2020 statement (shown in Fig. 1) [13].

Twenty-two comparative studies with 2,063 patients in the ERAS group and 1,980 patients in the comparator arm fulfilled our eligibility criteria [23‒44]. There were four RCTs and 18 observational studies from 2013 to 2022. Asia, Europe, and the USA contributed ten, nine, and three studies. The reported surgical intervention was pylorus-preserved PD (PPPD) in three studies and Whipple in four, whereas it was either PPPD or Whipple in five studies and mixed (PPPD, Whipple, or Stomach preserved PD [SPPD]) in two articles and not specified PD in the remaining seven articles. The mean patients’ age ranged from 51 to 77 years across studies, and the male-to-female ratio was 1.26 and 1.3 for the ERAS and control groups. Table 2 summarizes the basic study characteristics of our eligible studies. ERAS items implemented in each study are depicted in the online supplementary file 4.

According to ROBINS-I, the overall risk of bias in the 18 observational studies was moderate (75%) to high (25%) (shown in online supplementary file 2). Regarding the RCTS, the RoB-2 tool identified serious concerns for risk of bias in 25% of the available evidence, mainly attributed to bias attributed to the potential deviations from the intended interventions (shown in online supplementary file 3).

Table 3 displays the meta-analysis results with a summary of the evidence.

Eighteen studies reported overall postoperative complications (shown in Fig. 2) [23‒31, 33, 35, 37‒41, 43, 44]. In total, 989 and 1,080 complications occurred in the ERAS and control groups, corresponding to a proportion incidence of 52% (44–59%) and 66% (57–73%), respectively. In the presence of significant heterogeneity (63%), the random-effect models showed that the ERAS pathway was associated with fewer complications (RR 0.83; 95% CI: 0.75–0.91). The Beggs test did not identify significant concerns about publication bias (p = 0.198). The studies by Takchi et al. [44] and Dai et al. [33] contributed the most to the statistical heterogeneity according to the Baujat plots [44]. After excluding each study and re-running the meta-analysis, the pooled RR were 0.82 (0.77–0.86) and 0.86 (0.81–0.91), respectively. The meta-regression showed that the effect varied according to the continent of the study origin (p = 0.047). Indeed, the RR was more profound in studies from Asia (0.74; 0.66–0.84), marginal in studies from Europe (0.91; 0.83–0.99), and not significant in studies from the USA (0.95; 0.58–1.56). The present results are robust and originate from high-quality evidence with a fragility index as high as 29 (Table 3). At the same time, the NNT to avoid a single complication using ERAS is as low as 9 (6–18).

Clavien-Dindo (CD) grade 1 and 2 complications were reported in 11 studies (shown in Fig. 3) [23‒25, 28‒31, 33, 35, 37, 39]. In total, 252 and 345 CD 1 and 2 complications occurred in the ERAS and control groups, corresponding to a proportion incidence of 35% (29–42%) and 43% (31–56%), respectively. In the absence of significant heterogeneity (3%), the fixed-effect models showed that the ERAS pathway was associated with fewer CD 1 and 2 complications (RR 0.82; 95% CI: 0.72–0.92). The Beggs test did not identify significant concerns about publication bias (p = 0.436). The studies by Abu Hilal et al. [23] and Dai et al. [33] mainly contributed to the statistical heterogeneity according to the Baujat plots. After excluding these studies and re-running the meta-analysis, the pooled RR was 0.80 (0.71–0.91) and 0.85 (0.74–0.97), respectively. The meta-regression did not identify any effect modifier among the studied parameters. The present results are robust and originate from moderate-quality evidence with a fragility index as high as 10. At the same time, the NNT to avoid a single CD 1 and 2 complication using ERAS is as low as 12.

CD grade 3 and 4 complications were reported in 12 studies (shown in Fig. 4) [23‒25, 28‒31, 33, 35‒37, 39]. In total, 150 and 173 CD 3 and 4 complications occurred in the ERAS and control groups, corresponding to a proportion incidence of 18% (11–28%) and 19% (14–25%), respectively. In the presence of significant heterogeneity (56%), the random-effect models showed that the ERAS pathway was not associated with fewer CD 3 and 4 complications (RR 1.00; 0.72–1.38). The Beggs test did not identify significant concerns about publication bias (p = 0.945). The studies by Abu Hilal et al. [23] and Dai et al. [33] contributed the most to the statistical heterogeneity according to the Baujat plots. After excluding these studies and re-running the meta-analysis, the pooled RR was 0.88 (0.73–1.08) and 1.05 (0.87–1.28), respectively. The meta-regression did not identify any effect modifier among the studied parameters. The present results are relatively robust and originate from high-quality evidence with a fragility index as high as 26. At the same time, the NNT to avoid a single CD 3 and 4 complication using ERAS is as high as 3,333.

DGE was reported in 20 studies (shown in Fig. 5) [23‒36, 38‒40, 42‒44]. In total, 254 and 369 cases presented DGE in the ERAS and control groups, corresponding to a proportion incidence of 14% (1–20%) and 24% (16–35%), respectively. In the presence of significant heterogeneity (73%), the random-effect models showed that the ERAS pathway was associated with fewer DGE (RR 0.69; 0.52–0.93). The Beggs test did not identify significant concerns about publication bias (p = 0.112). The study by Van der Kolk et al. [36] contributed the most to the statistical heterogeneity according to the Baujat plots. After excluding this study and re-running the meta-analysis, the pooled RR was 0.78 (0.67–0.91). The meta-regression did not identify any effect modifier among the studied parameters. The present results originate from low-quality evidence and are characterized by a fragility index of 7. At the same time, the NNT to prevent a patient from DGE using ERAS is as low as 12.

POPF was reported in 21 studies (shown in Fig. 6) [23‒35, 37‒44]. In total, 260 and 339 cases presented POPF in the ERAS and control groups, corresponding to a proportion incidence of 13% (11–17%) and 16% (13–21%), respectively. In the absence of significant heterogeneity (23%), the fixed-effect models showed that the ERAS pathway was associated with fewer POPF (RR 0.76; 0.66–0.89). The Beggs test did not identify significant concerns about publication bias (p = 0.319). The studies by Kobayashi et al. [26] and Aviles et al. [32] contributed the most to the statistical heterogeneity according to the Baujat plots. After excluding these studies and re-running the meta-analysis, the pooled RR was 0.80 (0.69–0.93) and 0.75 (0.64–0.87). The meta-regression identified that the type of surgery (p = 0.05) and the continent of study origin (p = 0.02) were among the effect estimate modifiers. Indeed, the beneficial effect of ERAS was evident in studies with mixed interventions (0.60; 0.48–0.76) but not with Whipple PD (0.96; 0.62–1.44) or PPPD (0.86; 0.64–1.28). Likewise, studies from Asia reported fewer cases of POPF using ERAS (0.68; 0.52–0.82). However, this beneficial effect of ERAS was not observed in studies from Europe (0.81; 0.62–1.05) and the USA (1.77; 0.94–3.22). The present results originate from a moderate quality of evidence and are characterized by a fragility index of 4. At the same time, the NNT to prevent a patient from POPF using ERAS is 24.

PPH was reported in 16 studies (shown in Fig. 7) [23‒27, 29, 31, 33‒35, 37, 39‒43]. 104 and 101 cases presented PPH in the ERAS and control groups, corresponding to a proportion incidence of 6% (5–7%) and 7% (6–8%), respectively. In the absence of significant heterogeneity (0%), the fixed-effect models showed that the ERAS pathway was not associated with fewer PPH (RR 0.88; 0.67–1.14). The Beggs test did not identify significant concerns about publication bias (p = 0.829). The meta-regression did not identify any effect modifier among the studied parameters. The present results originate from very low-quality evidence and are characterized by a fragility index of 11. At the same time, the NNT to prevent a patient from PPH using ERAS is as high as 114.

Twenty-one articles studied the readmission rate after pancreatic resection surgery (shown in Fig. 8) [23‒27, 29‒44]. In total, 208 and 204 cases required re-admission in the ERAS and control groups, corresponding to a proportion incidence of 6% (3–10%) and 7% (5–11%), respectively. In the absence of significant heterogeneity (0%), the fixed-effect models showed that the ERAS pathway was not associated with fewer readmissions (RR 1.01; 0.84–1.21). The Beggs test did not identify significant concerns about publication bias (p = 0.528). The meta-regression did not identify any effect modifier among the studied parameters. The present results were extracted from low-quality evidence and characterized by a fragility index 12. At the same time, the NNT to prevent a patient from readmission using ERAS is as high as 275.

All studies provided data on LOS (shown in Fig. 9) [23‒44]. However, 17 studies described the LOS in terms of median and range or IQR values. Among them, in 15 studies, the data were significantly skewed away from normality, and thus, it was not appropriate to apply the normal-based method for data transformation. Only seven articles provided data amenable to quantitative synthesis on the LOS after pancreatic resection surgery [23, 26, 31, 34, 39, 42, 43]. The pooled mean hospital stay was 16.6 days (95% CI: 12.2–19.9) and 19.7 (95% CI: 16.5–22.9) using ERAS and the standard of care, respectively. According to our meta-analysis, and in the presence of significant statistical heterogeneity (95%), the standardized mean difference (SMD) in LOS was −0.56 (95% CI: −1.12 to 0.01) and did not differ between the two groups at a significant level. After eyeballing the funnel plot, we concluded that there was no significant risk of publication bias. According to the Baujat plots, the study by Kim et al. [42] contributed the most to the statistical heterogeneity. Notably, after excluding this study and re-running the meta-analysis, the pooled SMD was −0.76 (95% CI: −0.91 to −0.60) and changed in favor of the ERAS group. The meta-regression did not identify any effect modifier among the studied parameters. The present results were extracted from very low-quality evidence.

Two articles studied the time to chemotherapy after pancreatic resection surgery (shown in Fig. 10) [38, 40]. The pooled mean time to chemotherapy was 53.6 days (95% CI: 50.7–55.4 days) and 67.9 days (95% CI: 65–70.87 days) using ERAS and the standard of care, respectively. According to our meta-analysis and in the absence of significant statistical heterogeneity (0%), the standardized mean difference in the time to chemotherapy was −0.68 days (95% CI: −0.88 to 0.48 days) in favor of the ERAS group. Due to the small number of eligible studies, no further analysis took place. The present results were extracted from very low-quality evidence.

Twenty-two articles studied the mortality after pancreatic resection surgery [23‒44]. In total, 39 and 47 cases occurred in the ERAS and control groups, corresponding to a proportion incidence of 2% (1–3%) and 2% (1–3%), respectively. In the absence of significant heterogeneity (0%), the fixed-effect models showed that the ERAS pathway was not associated with lower mortality (RR 0.81; 0.54–1.22). The Beggs test did not identify significant concerns about publication bias (p = 0.36). According to the Baujat plots, the studies by Tackhi et al. [44] and Li et al. [40] contributed the most to the statistical heterogeneity. After excluding these studies and re-running the meta-analysis, the pooled RR was 0.73 (0.48–1.13) and 0.91 (0.58–1.41). The meta-regression did not identify any effect modifier among the studied parameters. The present results originated from very low-quality evidence and are characterized by a fragility index of 9.

Our recent systematic review has found twenty-two studies that compared the ERAS pathway with the standard of care for patients undergoing PD. Serial meta-analyses showed that ERAS can reduce overall and minor complications, DGE, POPF, and time to chemotherapy. Nonetheless, we have found no significant impact on the incidence of severe complications, PPH, re-admission rates, and associated mortality. Additionally, most of the studies we reviewed indicated that ERAS could reduce the duration of hospital stay after PD.

These results align with prior systematic reviews examining the effects of ERAS on morbidity following PD [45‒51]. Implementing ERAS principles lowers overall and minor complications while not causing an increase in major complications. Notably, this meta-analysis is the first to calculate the NNT for each outcome, revealing that the NNT to avoid a single complication using ERAS can be as low as 9. This finding further supports the beneficial effect of ERAS on overall morbidity. Furthermore, the study suggests that ERAS protocols can safely reduce the incidence of complications, including DGE and POPF. Such results reinforce the safety of ERAS interventions, such as early oral feeding and the prompt removal of NGT and drains, which have been controversial.

Additionally, this systematic review is the first to explore the role of ERAS in the timing of adjuvant chemotherapy following PD. Interestingly, ERAS patients were found to initiate chemotherapy 14 days sooner than patients who received conventional perioperative care. While it is essential to exercise caution in interpreting this finding due to the limited number of studies, it is of paramount clinical significance as 30% of patients do not receive adjuvant therapy following PD due to postoperative complications, early metastases, and decreased performance status [52].

The ERAS society has established 27 guidelines for PD [11]. However, the protocols of the studies reviewed in our investigation included only some of these items. Some of the studies implemented their ERAS programme several years before the first ERAS guidelines for PD was published in 2012. That is the main reason why some studies implemented as low as 9 ERAS items. The most commonly implemented guidelines were preoperative counseling, antithrombotic and antimicrobial chemoprophylaxis, prevention of PONV and hypothermia, as well as multimodal analgesia. Postoperatively, the most frequently implemented items were removal of NGT, early oral feeding, early and scheduled mobilization, and drain and urinary catheter removal plans. Nevertheless, not all studies reported compliance with ERAS pathways. A recent meta-analysis of studies on patients undergoing PD revealed that the median overall compliance with ERAS guidelines was 65.7%, with postoperative compliance as low as 44% [49]. It was observed that morbidity was lower when compliance levels were above 50% [49]. Interestingly, low compliance in the early postoperative period, particularly poor tolerance of early oral feeding, was often linked to complications. Early identification of such patients offers a chance to intervene early and prevent further deterioration [24, 49].

Implementing ERAS programs for pancreatic surgery is a challenging task due to the complexity of the surgical procedure and the high risk of complications that may hinder the change in traditional clinical practice. Nevertheless, the findings of this review strengthen the existing evidence that ERAS pathways for PD improve safety and short-term outcomes, thus becoming the standard of care. It is crucial to encourage local initiatives to establish modern and evidence-based perioperative care models, and successful strategies should be shared across centers to foster a culture of learning from one another. To this end, we are sharing our locally adapted ERAS protocol for PD, which we recently introduced in our practice (shown in Fig. 11). Changing the culture in the complex healthcare environment is undoubtedly challenging. Nonetheless, strong leadership, teamwork, and continuous audits based on the PDSA (plan-do-study-act) theoretical framework should be the way to promote a culture of quality improvement and evidence-based practice.

While ERAS protocols have shown promise in patients undergoing PD, questions remain about their application, given the high complication rate associated with this procedure. To better understand which ERAS items lead to positive outcomes, further research is needed, focusing on functional recovery, time for adjuvant chemotherapy, and the impact on long-term outcomes and survival.

Timely postoperative chemotherapy is crucial in increasing the chances of cure after surgery. However, the postoperative stress response leads to immunosuppression, which creates a vulnerable window of opportunity for the expansion of minimal residual disease. As a result, the patient becomes more susceptible to tumorigenesis after removing the primary tumor [53]. It is reasonable to assume that strategies such as ERAS principles that suppress stress response may protect patients against perioperative tumor growth. Another critical issue for future research is whether compliance with ERAS programs may affect oncological outcomes.

Well-designed multicenter prospective cohort studies may be more appropriate compared to RCTs. Not only does the number of interventions that comprise the ERAS pathway make randomization and blinding not feasible, but it might also be unethical to randomize patients to the control group and deny them evidence-based interventions. In addition, future research should examine human factors and models of education, teamwork, and leadership support to facilitate the application of new perioperative care models that target the whole patient journey rather than just one intervention.

It should be noted that the current study has certain limitations. First, the analysis is based on relatively few studies, particularly those focusing on the LOS and time to chemotherapy. Second, the quality of evidence varies significantly among the parameters studied. Most studies were case-control studies, which may result in selection bias. Additionally, most of these studies were conducted retrospectively, which means that the accuracy of process indicators may have affected some patients. The quality of the randomized controlled trials (RCTs) was moderate due to the lack of blinding, which could introduce bias in implementation and measurement. However, applying blinding methods for the ERAS protocol is problematic. Third, we identified significant statistical heterogeneity in some parameters during the meta-analysis, such as overall complications and DGE. We used several techniques, such as random-effect models and sensitivity analysis using the leave-out-one method, to search for sources of statistical heterogeneity and overcome the problem. Fourth, the LOS and time to chemotherapy were reported using median and range values. Whenever possible, we transformed them into mean values and standard deviation. Lastly, not all studies implemented the same ERAS protocol, and the number of ERAS items used in each study varied between 9 and 25, potentially causing clinical heterogeneity. However, this is inevitable due to how clinical pathways are devised based on local clinical practices and sociocultural needs.

According to this review and meta-analysis, implementing ERAS principles in pancreatic surgery can lower the occurrence of overall and specific complications such as DGE and POPF while not posing any more significant risk of major complications, readmission, or mortality. ERAS is a secure and practical approach to pancreatic surgery, and it may even enhance oncological outcomes by hastening recovery and decreasing the time needed for chemotherapy. In future research, emphasis should be placed on implementation strategies and considering human factors and cultural context to ensure the successful application of new perioperative care models.

No patient informed consent or IRB/Ethics Committee approval was required, as the current meta-analysis was based on published records.

The authors have no conflicts of interest to declare.

This study was not supported by any sponsor or funder.

Despoina Liotiri: conceptualization, methodology, formal analysis, investigation, resources, data curation, writing – original draft, writing – review and editing, and visualization.

Alexandros Diamantis: methodology, investigation, and writing – review and editing.

Ismini Paraskeva: methodology and writing – review and editing.

Alexandros Brotis: methodology, formal analysis, investigation, resources, and data curation.

Dimitrios Symeonidis: methodology and writing – review and editing supervision.

Eleni Arnaoutoglou: methodology, writing – review and editing, and supervision.

Dimitrios Zacharoulis: conceptualization, methodology, writing – review and editing, and supervision.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants but are available from the corresponding author [D.L.] upon reasonable request.