Optimizing invasive strategies for necrotizing pancreatitis: A Bayesian network analysis of randomized controlled trials

Abstract

Background:

necrotizing pancreatitis is a severe complication of acute pancreatitis, often requiring invasive interventions to manage its high mortality and morbidity. The optimal timing and type of invasive procedures remain uncertain, necessitating a systematic evaluation to guide clinical decision-making.

Methods:

A systematic review and Bayesian network meta-analysis were conducted following the PRISMA guidelines. Relevant randomized controlled trials (RCTs) published up to November 2024 were retrieved from PubMed, EMBASE, and the Cochrane Library. The study assessed 10 invasive interventions, including early and delayed drainage, step-up approaches, and open surgeries, focusing on mortality and major complications. Statistical analysis employed random-effects models and Bayesian frameworks to synthesize direct and indirect evidence.

Results:

Fifteen RCTs involving 857 patients were included. Delayed step-up surgery (DSU) and early drainage (ED) with lavage (EDL) demonstrated significant survival benefits, with lower mortality rates and reduced complications. Conversely, delayed video-assisted surgery (DVS) was associated with the highest mortality. No statistically significant differences were observed between ED and EDL or ED and delayed drainage in direct comparisons. Subgroup analyses revealed no significant mortality difference between early and delayed interventions (OR = 1.15, 95% CI = 0.54–2.46), while EDL and DSU emerged as optimal strategies in early and delayed interventions, respectively.

Conclusion:

This review and network meta-analysis suggests that DSU and EDL may be promising options for treating necrotizing pancreatitis, though current evidence is inconclusive. Given varying risks, especially with DVS, treatment should be tailored to individual cases. More high-quality RCTs are needed to strengthen the evidence and guide practice.

1. Introduction

Necrotizing pancreatitis (NP), a severe consequence of acute pancreatitis, can lead to multiple organ failure and death. The global incidence of acute pancreatitis has risen by 3% annually over the past 50 years, now affecting approximately 1.5 million individuals annually.^[1,2] Unlike mild acute pancreatitis, which typically involves brief hospitalization, NP requires extended hospital and ICU stays, more interventions, and higher healthcare costs.^[3] While 90% to 95% of acute pancreatitis cases involve interstitial edematous pancreatitis, the remainder progress to NP.^[4] Local complications include acute necrotic collections within the first 4 weeks and walled-off necrosis after 1 month,^[4] also systemic complications may exacerbate preexisting conditions, such as heart or chronic lung disease.^[4] Furthermore, mortality following acute pancreatitis ranges from 6% to 20%.^[5,6]

Although most necrotic tissue remains uninfected, approximately 30% of patients develop infections, indicated by gas in the collection, positive cultures, persistent sepsis, or clinical deterioration.^[7,8] Infected necrosis, whether in the pancreas or surrounding tissue, significantly worsens outcomes and typically necessitates intervention.^[9] Treatment strategies for acute NP include early surgical debridement – via open surgery or minimally invasive retroperitoneal techniques – and delayed necrosectomy, performed after the acute phase. Percutaneous drainage and step-up approaches, beginning with endoscopic or percutaneous drainage and escalating to necrosectomy if required, are also widely utilized.^[9–23]

The American College of Gastroenterology recommends delayed necrosectomy as the primary treatment for clinically stable patients with infected NP, while early necrosectomy is advised for those with clinical instability.^[24] Evidence suggests that less invasive approaches, such as percutaneous drainage followed by necrosectomy or endoscopic transluminal drainage, yield better outcomes than surgical debridement.^[9,10] Therefore, comparing the effectiveness of various treatments is crucial. Network Meta-Analysis (NMA) facilitates this by evaluating multiple therapies simultaneously, even in the absence of direct head-to-head comparisons, offering valuable insights into their relative effectiveness.

2. Methods

2.1. Search strategy and selection criteria

A systematic review was undertaken to evaluate the available evidence on invasive treatments for NP, adhering to the guidelines outlined in the Cochrane Handbook for Systematic Reviews of Interventions and the PRISMA statement.^[25,26] The review encompassed studies indexed in PubMed, Embase, Cochrane Library, and Web of Science up to November 30, 2024. To ensure comprehensive coverage, the search strategy incorporated a range of keywords and Medical Subject Headings (MeSH and Emtree), including “NP,” “intervention,” “endoscopic,” “invasive,” “step-up approach,” “laparoscopic” and “surgery.” Logical operators (“AND” and “OR”) were applied to refine the search and target relevant outcomes, as illustrated in Figure 1.

Figure 1.

Flow diagram showing the selection of randomized controlled trials.

Only randomized clinical trials (RCTs) with adult patients (≥18 years) diagnosed with NP were included. All studies which compared different invasive treatments (included drainage, endoscopic debridement, surgical step-up approach, drainage + lavage, open or video-assisted surgery) or different timing (early or delayed) of invasive therapy with full texts or detailed abstracts were included, excluding conference proceedings or unpublished data. Interventions were categorized as early or late based on the timing of intervention (<4 weeks or ≥4 weeks from the onset of pancreatitis). Mortality outcome was the mandatory criterion for selection and all included studies must provide mortality outcomes. In addition, our study focused on the impact of different invasive treatments on NP, therefore, studies comparing invasive therapies with conservative treatments or focusing on fluid resuscitation techniques were excluded. Additional exclusions included retrospective and non-randomized prospective studies, as well as non-English publications.

2.2. Data extraction and methodological quality

The following information was extracted from the search results to assess comparability: author name, publication year, journal title, study title, inclusion and exclusion criteria, and primary and secondary outcomes. Data extraction and validation were independently conducted by 2 researchers, Fucai Yu and Nian Wu, who meticulously analyzed patient numbers, invasive procedures, and mortality. For studies from the same trial with varying follow-up durations, data from the most recent study were prioritized. The risk of bias (RoB) was evaluated using the Cochrane RoB 2 tool based on the Cochrane Handbook.^[26] Five domains were assessed, including deviations from intended interventions. Domains were rated as “Low RoB,” “Some Concerns,” or “High RoB.” Studies with one “Some Concerns” were classified as “Low RoB,” while 2 or more received “Some Concerns,” and any “High RoB” domain resulted in a “High RoB” rating. Two authors reviewed RoB independently, resolving disagreements through discussion, and any discrepancies were resolved through discussion. If necessary, a third party (Lu Huan) could be consulted for arbitration. Only existing protocol or statistical analysis documents were consulted.^[27] The majority of studies exhibited low bias in the randomization process and outcome measurement (Fig. 2). However, some studies showed a higher bias in the loss of outcome data, which could potentially lead to shifts in the estimation of effects.

Figure 2.

(A) Risk bias of summary. Judgments about each risk of bias item for each included trial. Green indicates low risk of bias. Yellow indicates unclear risk of bias. Red indicates a high risk of bias. (B) Risk bias of graph. Each risk of bias item presented as percentages across all of the included trials, which indicated the proportion of different level risk of bias for each item.

2.3. Data synthesis and analysis

Traditional pairwise meta-analysis was performed to provide direct evidence using the statistical software Review Manager 5.3. We chose random-effects models that assumed that the true effects were not identical in different studies. The results were presented as odds ratios (ORs) with 95% confidence intervals (CIs) for dichotomous data with 95% Cis. A Bayesian NMA was performed using WinBUGS 1.4 (MRC Biostatistics Unit) and R 4.4.2, enabling effectiveness estimates beyond direct comparisons. Non-informative uniform and normal prior distributions, we employed a Bayesian hierarchical model for NMA using the Markov Chain Monte Carlo method with 4 chains (20,000 tuning iterations, 50,000 sampling iterations, thinning interval of 10). Convergence was assessed using the Gelman-Rubin R-hat statistic (≤1.05), indicating good chain mixing. Posterior density plots demonstrated smooth, unimodal distributions, ensuring stable estimates. Node-split analysis showed no significant inconsistencies, confirming the consistency of direct and indirect evidence and supporting the robustness of our model. Furthermore, we set a normal (0, 10²) prior for the overall effect size and tested various priors for between-study heterogeneity, including half-normal (0.5) and uniform (0, 4), to assess sensitivity. Random-effects models were used to account for variability among studies. Agreement between direct and indirect comparisons was assessed to address inconsistencies.^[28] The Bayesian approach also provided overall ranking probabilities, making it possible to rank each outcome measurement from the best to the worst, and was then visualized by calculating the surface under the cumulative ranking curves (SUCRA) on the basis of the ranking profiles. Following NICE guidelines, deviance residuals and deviance information criteria were compared between consistency and inconsistency models, with posterior mean deviance plotted to identify potential treatment network inconsistencies.^[28,29] Moreover, to assess consistency between direct and indirect evidence within the NMA, a loop-specific consistency check was conducted, calculating the ratio of ORs (RORs) with their corresponding 95% CI for each loop. Random-effects models were also used to compute pooled effect estimates, accounting for statistical and clinical heterogeneity. Heterogeneity was classified as low (I² ≤ 25%), moderate (25% < I² ≤ 50%), or high (I² > 50%).^[30] Given the comparisons might engender concerns about an elevated risk of Type I errors, we would like to clarify that our study employed a Bayesian NMA framework. Given the risk of bias in the study due to the loss of outcome data and to investigate interventions, timing, and NP more precisely, we divided our studies into 3 subgroups: for direct comparisons between early and delayed surgical interventions (including drainage) on mortality (Considering that the majority of cases are caused by biliary tract stones, we conducted an exploratory meta-regression using the proportion of biliary pancreatitis as a continuous moderator); for direct comparisons of early interventions on mortality; and for direct comparisons of delayed interventions on mortality.

2.4. Data items

The extracted data were categorized as follows: general information: first author, publication year, country, type of invasive comparison, sample size, and age range. Intervention arms: early drainage (ED), ED + lavage (EDL), delayed drainage (DD), early endoscopic debridement (EED), delayed endoscopic debridement (DED), delayed video-assisted surgery (DVS), early video-assisted surgery (EVS), delayed surgical step-up (DSU), delayed open surgery, early open surgery. Endpoints was mortality.

2.5. Publication bias and sensitive assessment

In the primary analysis, publication bias was assessed using funnel plots, a visual method for detecting small-study effects or potential biases. Necessarily, Egger regression test was performed to quantitatively evaluate the asymmetry of the funnel plot when applicable. Furthermore, to assess the robustness of our findings, a leave-one-out sensitivity analysis was conducted. Each study was sequentially removed from the network, and the NMA was rerun to examine the impact of individual studies on the overall results. The primary outcome, expressed as ORs with 95% CIs, was recalculated at each iteration. In addition, for direct comparisons with sparse data (e.g., ED vs EDL), we applied a fixed-effects model as a sensitivity test. Concurrently, funnel plots were employed to examine potential publication bias within the subgroups where direct comparisons were made.

3. Results

3.1. Study characteristics

The search identified 2386 articles, of which 366 duplicates were removed. After screening titles and abstracts, 1928 irrelevant publications were excluded. A full-text review eliminated 73 additional references due to the following reasons: 15 were protocols or case reports, 11 were reviews or meta-analyses, 33 were not RCTs, and 14 addressed unrelated topics or reported different follow-up periods within the same cohort. Ultimately, 15 RCTs met the criteria for quality assessment and quantitative synthesis.

The 15 studies chosen for inclusion covered the time frame from 1984 to 2024. The research started as early as 1982 and as late as 2024 in 8 different nations. (Table 1) Each of the studies included in the analysis was carried out as a 2-arm randomized controlled experiment. A total of 857 patients with NP were randomly assigned to 1 of 10 interventions: 5 studies^{[12–14,16,23]} for ED, 3 studies^[12,14,23] for DD, 4 studies^{[13,16,18,21]} for EDL, 2 studies^[10,22] for DED, 1 study^[11] for EED, 1 study^[10] for DVS, 3 study^[11,15,20] for EVS, 4 studies^[9,17,21,22] for DSU, 5 studies^[15,17–20] for early open surgery and 2 studies^[9,19] for delayed open surgery. The average age of patients varied between 35.7 and 63 years. The cause of NP was investigated in 10 studies^{[9–14,16,19,21,22]} biliary pancreatitis was found to be the most common underlying cause, accounting for 52.8% (339 out of 642 cases). All the research documented morality. Nine studies^{[9–14,17,20,22,23]} reported major complications. And most studies demonstrated a low risk of bias across most domains, but certain areas, such as “missing outcome data,” showed a notable contribution of high risk; the summary view in panel (B) and the detailed breakdown in panel (A) together provided a clearer understanding of the overall bias distribution (Fig. 2).

Table 1.

Study characteristics were listed.

Author, year	Country	Study design	Comparison	Size	Age
Bakker 2012[10]	Netherlands	RCT	EU	10	64 (46–72)
			DVS	10	62 (58–70)
Bang 2019[11]	USA	RCT	EU	34	52.9 ± 14.2
			EVU	32	55.6 ± 14.2
Boxhoorn 2021[12]	Netherlands	RCT	ED	55	60 ± 14
			DD	49	59 ± 11
He 2019[13]	China	RCT	ED	41	53 (41–62)
			EDL	39	48 (38–61)
Ke 2021[14]	China	RCT	ED	15	38 ± 18
			DD	15	40 ± 16
Kivilaakso 1984[15]	Finland	RCT	ELL	17	39.7 ± 10
			EOS	18	38.4 ± 10.2
Kohli 2019[16]	India	RCT	ED	32	41.6 ± 14.8
			EDL	28	37.4 ± 12.9
Litvin A 2010[17]	Belarus	RCT	DSU	37	NA
			EOS	35	NA
Maroske 1982[18]	Germany	RCT	EDL	12	48 (23-73)
			EOS	12	46 (21-67)
Mier 1997[19]	Mexico	RCT	EOS	25	42 ± 16
			DOS	11	42 ± 12
Schröder 1991[20]	Finland	RCT	ELL	10	40.5 ± 7.5
			EOS	11	35.7 ± 6.8
Shenvi 2016[21]	USA	RCT	DSU	11	37.3 ± 13
			EDL	24	41.7 ± 13.8
van Brunschot 2018[22]	Netherlands	RCT	EU	51	63 ± 14
			DSU	47	60 ± 11
van Santvoort 2010[9]	Netherlands	RCT	DSU	43	57.6 ± 2.1
			DOS	45	57.4 ± 2.0
Van Veldhuisen 2024[23]	Netherlands	RCT	ED	47	NA
			DD	41	NA

DD = delayed drainage, DED = delayed endoscopic debridement, DOS = delayed open surgery, DVS = Delayed video-assisted Surgery, ED = early drainage, EDL = early Drainage + Lavage, EED = early endoscopic debridement, ELL = early laparotomy lavage, EOS = early open surgery, EVS = early video-assisted surgery, NA = not available, RCT = randomized controlled trial.

3.2. Mortality

All 15 studies included reported mortality outcomes, but no direct comparisons showed statistically significant differences. This NMA provided a comprehensive overview of surgical techniques, scheduling options, and associated sample sizes (Fig. 3A). Both conventional pairwise and Bayesian meta-analyses were used to integrate direct and indirect evidence, as shown in Table 2. The analysis revealed that patients undergoing DSU, EDL and DED had significant survival advantages, while DVS was associated with a higher mortality risk (Fig. 3B), and the results were stable by consistency and inconsistency effects standard deviation (0.8, 95% CI = 0.12–1.61; 0.8, 95% CI = 0.05–1.65, respectively). What’s more, there was no publication bias (Fig. 3C). However, direct comparisons showed no significant differences between ED and EDL or ED and DD (OR = 2.28, 95% CI = 0.37–14.12; OR = 0.84, 95% CI = 0.37–1.88, respectively). What’s more, we conducted a stratified analysis overall to further support this, we have presented both a forest plot (Fig. 3D) illustrating the treatment effects compared to the reference (DD), and a SUCRA ranking plot (Fig. 3E), which summarizes the cumulative ranking probabilities.

Figure 3.

(A) Mortality of the comparisons for the Bayesian network meta-analysis. The size of the nodes was proportional to the number of patients (in parentheses) randomized to receive the treatment. The width of the lines was proportional to the number of trials (beside the line) comparing the connected treatments. (B) The rank probability of mortality rate in patients with different treatments. The rankings of the 14 competing different treatments in terms of mortality were summarized. The analysis suggested that DVS had a higher likelihood mortality and DOS were more likely to experience greater survival benefits. (C) The funnel plot assessed publication bias. (D) Forest plot of treatment comparisons from the stratified analysis. (E) SUCRA rankings of the 10 interventions. DD = delayed drainage, DED = delayed endoscopic debridement, DOS = delayed open surgery, DVS = delayed video-assisted surgery, ED = early drainage, EDL = early drainage + lavage, EED = early endoscopic debridement, ELL = early laparotomy lavage, EOS = early open surgery, EVS = early video-assisted surgery, SUCRA = surface under the cumulative ranking curves.

Table 2.

Both traditional pairwise meta-analysis and Bayesian meta-analysis were employed to synthesize direct and indirect evidence for mortality.

DD
0.77 (0.02–30.80)	DED	–	–	–	–	–	–	–	–
0.78 (0.04–22.91)	1.02 (0.07–20.33)	DOS	–	–	–	–	–	–	–
1.13 (0.07–20.97)	1.50 (0.17–15.33)	1.47 (0.23–7.43)	DSU	–	–	–	–	–	–
0.08 (0.01–10.20)	0.12 (0.01–2.60)	0.11 (0.01–6.49)	0.08 (0.01–3.49)	DVS	–	–	–	–	–
1.25 (0.34–5.07)	1.62 (0.06–52.31)	1.57 (0.08–26.15)	1.10 (0.09–14.04)	15.22 (0.15–330.25)	ED	–	–	–	–
2.81 (0.37–26.07)	3.69 (0.20–84.27)	3.61 (0.32–38.68)	2.50 (0.38–18.49)	34.51 (0.49–257.79)	2.26 (0.45–12.41)	EDL	–	–	–
0.19 (0.00–18.72)	0.25 (0.00–22.48)	0.24 (0.00–11.08)	0.17 (0.00–7.40)	2.18 (0.01–135.19)	0.16 (0.01–11.44)	0.07 (0.01–3.52)	EED	–	–
0.41 (0.02–7.84)	0.54 (0.04–9.65)	0.53 (0.08–2.92)	0.36 (0.08–1.78)	4.75 (0.08–276.04)	0.32 (0.03–4.14)	0.14 (0.02–1.04)	2.13 (0.07–73.56)	EOS	–
0.30 (0.01–11.23)	0.38 (0.02–12.79)	0.38 (0.03–5.43)	0.26 (0.03–3.48)	3.63 (0.04–277.54)	0.24 (0.01–6.25)	0.10 (0.01–1.66)	1.56 (0.10–32.43)	0.73 (0.12–5.31)	EVS

DD = delayed drainage, DED = delayed endoscopic debridement, DOS = delayed open surgery, DSU = delayed surgical step-up, DVS = delayed video-assisted surgery, ED = early drainage, EDL = early drainage + lavage, EED = early endoscopic debridement, ELL = early laparotomy lavage, EOS = early open surgery, EVS = early video-assisted surgery.

The leave-one-out sensitivity analysis demonstrated that no single study, including those with small sample sizes such as DVS and DED, had a significant impact on the overall results (Fig. S1, Supplemental Digital Content, https://links.lww.com/MD/P499). Moreover, for the direct comparison between ED and EDL (2 studies), we applied a fixed-effects model to assess result stability. The OR remained similar (1.99; 95% CI = 0.81–4.94), indicating no significant difference. The wide CI reflects uncertainty, likely due to the small sample size. Furthermore, the OR estimates are consistent across priors with overlapping 95% intervals, suggesting robust conclusions and mitigating prior bias concerns. An additional half-cauchy (0.5) prior test yielded similar results, enhancing transparency (Fig. S2, Supplemental Digital Content, https://links.lww.com/MD/P499). Meanwhile, we evaluated Markov Chain Monte Carlo convergence using the Gelman-Rubin R-hat statistic, with all parameters demonstrating values ≤1.05, indicating good convergence (Fig. S3, Supplemental Digital Content, https://links.lww.com/MD/P499). Posterior density plots were consistent and free of multimodality. Node-Split analysis confirmed no inconsistencies between direct and indirect evidence, reinforcing the robustness of our NMA results (Figs. S4–S8, Supplemental Digital Content, https://links.lww.com/MD/P499). Finally, a loop-specific consistency check was conducted for 2 closed loops in the NMA: DSU-EOS-DOS and DSU-EDL-EOS. The RORs and heterogeneity (τ²) were found to be 1.629 (95% CI = 1.00–7.01) and 0.000 for DSU-EOS-DOS, and 1.204 (95% CI = 1.00–13.36) and 0.000 for DSU-EDL-EOS, respectively. The CIs included 1, indicating no significant inconsistency (Fig. S9, Supplemental Digital Content, https://links.lww.com/MD/P499).

3.3. Subgroups

In subgroup 1, we directly compared the effects of early versus late invasive therapy (including drainage) on NP. The overall pooled OR was 1.15 (95% CI = 0.54–2.46), indicating no significant difference between the 2 groups. This was further confirmed by the P-value for the overall effect (P = .71) and heterogeneity was moderate (I² = 42%, P = .12) (Fig. 4A). The points representing studies were roughly symmetrical around the vertical line (log [OR] = 0), suggesting no publication bias (Fig. 4B). The Egger test revealed no significant asymmetry in the funnel plot (P > .05), indicating no significant publication bias was observed (Fig. 5). To explore whether this heterogeneity could be explained by the proportion of biliary pancreatitis in each study, we performed a meta-regression. The analysis showed no statistically significant association between the proportion of biliary pancreatitis and treatment effect size (β = −4.91, 95% CI = −10.56 to 0.74, P = .1), suggesting that the variation in biliary etiology does not account for the heterogeneity across studies (Fig. 6).

Figure 4.

Analysis of early versus delayed interventions. (A) Forest plot of OR comparing the effects of early versus delayed interventions in 6 studies. Each blue square represents the study-specific OR with its size proportional to the study’s weight, and the horizontal lines denote 95% confidence intervals. The diamond represents the pooled OR (1.15, 95% CI: 0.54–2.46) under a random-effects model, with moderate heterogeneity (I² = 42%). (B) Funnel plot assessing potential publication bias. Study effect sizes (log [OR]) are plotted against their SE, showing a symmetrical distribution, indicating no significant publication bias. OR = odds ratios, SE = standard errors.

Figure 5.

Egger regression plot: X-axis: precision (1/SE), Y-axis: standardized effect size (TE/SE), dots: individual study effect sizes, shaded area: 95% confidence interval of the regression line, red dashed line: Egger regression fitted line. SE = standard errors, TE = treatment effect.

Figure 6.

Meta-regression of treatment effect by the proportion of biliary pancreatitis. This plot shows the association between the log odds ratio (early vs delayed intervention) and the proportion of biliary pancreatitis across included studies. The blue regression line indicates a decreasing trend, suggesting that early intervention may be less beneficial in populations with a higher proportion of biliary etiology. However, this trend did not reach statistical significance.

In subgroup 2, all early interventions (including drainage) were compared either directly or indirectly in terms of their effects on NP. ED and EDL seemed a lower mortality, inversely, EED had a higher mortality in early interventions (Fig. 7A).

Figure 7.

(A) Rank probability distributions of 5 early invasive treatments for NP. (B) Rank probability distributions of 5 delayed invasive treatments for NP. NP = necrotizing pancreatitis.

In subgroup 3, all delayed interventions (including drainage) were compared either directly or indirectly in terms of their effects on NP. DSU seemed to have a lower mortality, inversely, DVS had the highest mortality in delayed interventions (Fig. 7B).

4. Discussion

This NMA evaluated the effectiveness of 10 treatments for NP based on data from 14 trials, which included a total of 857 patients. Our research showed that the DSU technique was shown to have the lowest mortality. Furthermore, both DED and EDL had relatively low mortality rates. Generally speaking, NP was considered for EDL, and if necessary, DSU could be continued as needed in subsequent conditions.

In the past, conservative management was the most common approach for treating NP until the 20th century.^[31,32] However, approximately a decade ago, open necrosectomy emerged as a surgical treatment that gained substantial traction as a valid option for severe pancreatitis.^[33,34] In the 21st century, the introduction of minimally invasive techniques had prompted a shift in perspective, with numerous studies suggesting that these approaches were superior to open surgery.^[35–37] Furthermore, recent research has mounted considerable evidence corroborating the advantages of endoscopic procedures in the treatment of INP.^[38,39] Additionally, several meta-analyses had attempted to determine the best timing and most effective interventions for NP through comparisons, ultimately concluding that the endoscopic approach showed potential for a reduced occurrence of significant complications compared to open surgery. Moreover, it was found that the time of intervention played a crucial role in determining negative consequences.^[40,41] Nonetheless, these investigations were limited by the insufficient number of direct comparisons between different therapies. In contrast, a NMA largely focused on comparing different interventions simultaneously,^[42] but it overlooked the surgical strategy and specific methodologies while mainly considering the date of the surgery. Therefore, the optimal timing for implementing drainage intervention continued to be a subject of debate. In the current investigation, we utilized Bayesian modeling to compare various interventions simultaneously, and clarified the timing and approach of surgical intervention, resulting in more comprehensive outcomes compared to previous meta-analyses. The research indicated that a step-up approach, which might include EDL followed by DSU when necessary, could be a feasible therapeutic strategy.

The PANTER trial showed that using a minimally invasive step-up strategy reduced the occurrence of major complications and death compared to open necrosectomy.^[43] Another study found that an early minimally invasive step-up approach was linked to higher mortality compared to delayed intervention cases.^[44] One reason for this difference was that well-contained abscesses were better suited for open surgery and also made minimally invasive procedures like video-assisted retroperitoneal debridement or endoscopic transluminal necrosectomy easier.^[11] Additionally, a proactive medical treatment plan involving broad-spectrum antibiotics and fluid resuscitation might help control the inflammatory response and improve organ function, potentially leading to better outcomes,^[45] thereby potentially influencing outcomes favorably. Our investigation indicated that, in the subgroups, there was no significant difference between early and delayed interventions. However, based on the situation, further comparisons within 2 subgroups (one comparing early interventions and the other comparing late interventions) revealed that EDL might have the lowest mortality among early interventions, and DSU might have the lowest mortality among delayed interventions, besides, DVS was the worst, which aligns with our main findings. One potential reason was that some instances of NP might improve on their own without intervention, whereas DVS could worsen abdominal hypertension and lead to multiple organ failure.^[7]

The conservative management of NP encompassed therapies such as fluid resuscitation, analgesia, combined antimicrobial regimens, and nutritional support, potentially augmented by the drainage of infected fluid collections.^[46–48] Despite drainage being an invasive procedure, it was advisable to forego pancreatic debridement and necrosectomy unless absolutely necessary.^[48] Contrary to previous guidelines, which had suggested the superiority of these surgical interventions, several investigations revealed that conservative therapy could be efficacious in the treatment of NP.^[49,50] Nevertheless, our current study indicated that EDL followed by DSU remained a critical treatment option, irrespective of whether the patient had previously undergone conservative treatment.

There were some specific limitations in our meta-analysis. The small sample size of certain intervention groups (e.g., DVS and DED), which could lead to unstable effect estimates. To address this issue, we performed a leave-one-out sensitivity analysis and exchange the model for comparison to examine the influence of individual studies on the NMA results. In addition, the inclusion of studies comparing different endoscopic treatment procedures was not possible due to the differences in treatment methods and timing related with endoscopic interventions, such as endoscopic draining, stent types, and anastomosis. Ultimately, pancreatitis had various origins, and we did not possess precise data to examine the influence of different causes on clinical outcomes. Nevertheless, we performed a meta-regression analysis focusing on pancreatitis triggered by gallstones, and the outcomes remained consistent. These factors might be responsible for the observed fluctuations in clinical outcomes.

5. Conclusion

This systematic review and Bayesian NMA assessed the efficacy of invasive therapies for NP. While DSU and EDL appeared to be effective, with lower mortality and complication rates than other methods, the evidence for their superiority is not conclusive. DVS was associated with a high mortality risk, but direct comparisons did not consistently show statistically significant differences between various interventions. The results suggest that treatment strategies should be customized to patient-specific circumstances, with DSU and EDL considered as potential optimal choices. Additional high-quality RCTs are needed to corroborate these findings and to refine clinical guidelines.

Author contributions

Data curation: Nian Wu, Hua Liu, Fucai Yu.

Formal analysis: Nian Wu.

Investigation: Nian Wu, Lu Huan, Hua Liu, Zhiwei Wang.

Methodology: Nian Wu, Lu Huan, Fucai Yu.

Project administration: Hua Liu.

Resources: Lu Huan, Zhiwei Wang, Fucai Yu.

Software: Lu Huan, Zhiwei Wang, Fucai Yu.

Supervision: Hua Liu.

Validation: Lu Huan.

Visualization: Hua Liu.

Writing – original draft: Nian Wu, Zhiwei Wang, Fucai Yu.

Writing – review & editing: Nian Wu, Fucai Yu.