Blumgart Anastomosis After Pancreaticoduodenectomy. A Comprehensive Systematic Review, Meta-Analysis, and Meta-Regression

Eligibility criteria were established using the "Participants, Intervention, Control, Outcomes, Study" (PICOS) approach: (1) the participants were patients having benign or malignant pancreatic head lesions; (2) the intervention was the open PD with reconstruction according to BA anastomosis; (3) the control arm was the PD with reconstruction according to non-BA DtoM anastomosis; (4) the primary outcome was the CR-POPF rate; and (5) prospective randomized controlled trials and non-randomized prospective or retrospective controlled studies. No other differences were considered about the technical variant of both BA and non-BA DtoM. Nonetheless, the impact of variants was considered in the adjustment of bias.

No language, publication date, or status restrictions were used. A systematic literature search was done through Pubmed, Scopus, and the ISI Web of Science. The last research was performed on November 9, 2020. The search string used was the following: "((pancreatojejunostomy OR pancreaticojejunostomy OR Blumgart anastomosis OR ((pancreas OR pancreatic) AND (anastomosis OR anastomoses OR anastomose))) AND (surgery OR resection OR pancreatoduodenectomy OR pancreaticoduodenectomy OR Whipple OR PPPD OR technique OR reconstruction))." We used all related articles to enlarge the systematic search, and the references of included studies were examined. The systematic search results were managed with Thomson Reuters Endnote version X7.

A PRISMA flowchart was plotted to report the conclusions obtained by the authors. The studies were de-duplicated, and subsequently, they were screened according to the title and abstract to remove the records not relevant for this study's aim. Thus, the eligibility was evaluated in the remaining full-text articles using the following inclusion and exclusion criteria. Briefly, the inclusion criteria were: (1) comparative, randomized or non-randomized, design; (2) as intervention arm, any BA technique after PD; as control arm, any non-BA DtoM after PD; (3) the presence of POPF rate in both arms according to the 2016 ISGPF definition [19] or according to 2005 ISGPF definition but reporting B and C grade [23]; and (4) humans only clinical study. We defined BA as any anastomosis using trans-pancreatic sutures in the external layer and duct to mucosa in the inner layer. In the comparator arm, we included any duct to mucosa using other external layer methods. In Supplementary Table 1, all technical variants were described.

The following criteria were used to exclude studies: (1) studies that did not report original data or reviews or meta-analyses; (2) studies with data reported in untractable form; (3) studies that did not report data about CR-POPF; and (4) others type of anastomosis such as pancreaticogastrostomy in the control arm. The selection was independently performed in a blinded manner using a standardized form by two different authors (C.I. and L.A.). All eligible papers were reviewed in full-text form, and all the studies that met all the inclusion criteria without exclusion were selected for the analysis. Any disagreement was resolved after a collegial discussion between the reviewers and the senior author (R.C.).

Two reviewers (C.I. and C.M.) performed the data extraction using a preformed excel spreadsheet^©. The following information was extracted to define each study's characteristics: authors, affiliation and country, year of publication, type of design (randomized or non-randomized), the technique of BA and non-BA DtoM anastomosis, the sample size of each arm, and the outcomes reported. On the contrary, the following data were extracted for the analysis: clinically relevant POPF, mortality, morbidity, B and C PPH, B and C DGE, reoperation rates, and LOS. All the disagreements were solved with a discussion between the reviewers and the last author (R.C.).

The studies' qualitative assessment was carried out based on the methodological index for non-randomized studies (MINORS) [24]. All categorical variables were reported as frequencies and percentages, while the continuous variables were described as means and standard deviations. A dedicated statistical algorithm was used to calculate the mean and standard deviation in studies that presented medians and interquartile ranges [25, 26]. The results were reported, for dichotomous variables, as risk difference (RD) with a 95% confidence interval (95% CI). The RD was used instead of the odds ratio (OR) to avoid an increase in the target event's rate. The study's primary endpoint was a relatively rare event, and some cells with zero value were expected [27]. Moreover, these results were reported as number needed to treat (NNT) calculated using the following formula \(NNT=\frac{1}{ARR}=\frac{1}{RD}\) in which ARR represents the absolute risk reduction, namely the RD of the target event. The NNT value means the number of subjects who would have to undergo the BA rather than the non-BA DtoM to prevent one more event (e.g., CR-POPF). When RD with 95% CI assumes a positive value, the NNT was not calculated because the BA anastomosis did provide any clinical advantage. The primary endpoint were also analyzed using the ORs of multivariate models predicting CR-POPF when reported. Thus, this result was calculated using directly the ORs derived from each study. In this case, the results we reported only overall OR and NNT were not calculated. For continuous values such as LOS, we reported the weighted mean difference (WMD) with a 95% CI. The meta-analysis was carried out in line with recommendations from the Cochrane Collaboration and Meta-analysis of Observational Studies in Epidemiology guidelines [28, 29], and the Mantel–Haenszel random-effects model was used to calculate effect sizes [30].

The risk of bias across included studies was tested, measuring both the "between-study heterogeneity" and publication bias. The heterogeneity was measured by testing both I² and Cochran's Q statistics [31]. The I² value reports the percentage of variation across the included studies related to heterogeneity rather than sampling error. The heterogeneity was interpreted as follows: If I² was < 50%, the risk of "between-study" heterogeneity was considered low–moderate, and if I² was ≥ 50%, it was judged high. The meta-regression analysis was carried out when the heterogeneity is high, and the result is statistically relevant [32]. The meta-regression was not planned for the primary endpoint based on ORs derived from multivariate analysis because it judged, in itself, not a risk. The meta-regression was based on the use of maximum residual likelihood (REML) [33, 34]. We calculated the distribution of confounding covariates among each arm in the first step, reporting the results as RR or WMD with an entire 95% CI. We also calculated the frequency of confounding variables that changed among the studies, such as MINORS score, type of BA, or non-BA DtoM. In the second step, β value with standard error (SE) and R² was reported. The β value ± SE was related to the change in the RD unit of the target event: a positive β value means that the covariate increased rate generates a positive RD modification. The R² measured the quote, in percentage, of the heterogeneity explained by the variable. A two-tailed P value < 0.05 was judged significant. The P values were also recalculated using Monte Carlo permutation to obtain robust results [35]. The Begg and the Egger test [36] was used to exploring the presence of the publication bias, and a P-value < 0.05 indicated a non-negligible "small-study effect." The statistical analysis was carried out using dedicated packages for STATA v14®.

The results of the systematic search of the literature following the PRISMA statement are reported in Fig. 1. The search identified 10,098 papers: 7083 from the Medline/PubMed database, 1127 from the ISI Web of Science, and 2848 from Scopus. Six thousand one hundred twenty-three titles were leftover after de-duplication. Of these, 7008 were excluded from evaluating the title and abstract because they were not pertinent to our study field. Seventy-four were reviewed in full-text form, and of these, 63 were excluded because: 53 were case series; 5 were reviews without original data; 2 were letters to the editor; and 3 contained unextractable data. Finally, twelve studies [5‐16] were available for quality assessment and quantitative synthesis. There was 100% agreement between the two reviewers. The characteristics of the studies selected are summarized in Table 1. All the studies were published in the twenty-first century and only four [5, 14‐16] in a Western country (33.3%). The majority of studies (8 out of 12, 66.7%) did not have a randomized design nor propensity score matching adjustment (PSM). In most studies (9/12, 75%), the original technique described by Blumgart [5] was modified. The main modification was the reduction of interrupted trans-pancreatic sutures from 4 or 6 to 2 or 3. There were 6 studies (50%) reporting non-BA DtoM according to the Cattel–Warren technique [37] and 6 (50%) reporting Kakita one in the control group [38]. A total of 2368 patients were recorded: 1075 (45.4%) undergoing BA and 1193 (54.6%) non-BA DtoM. A detailed description of all surgical techniques is reported in Supplementary Table 1. The other potential confounding variables, varying among the two arms, are summarized in Supplementary Table 2. Briefly, gender and age were extractable for all studies except Halloran et al. [14]; type of lesion and pancreatic texture were frequently available (10 and 9 studies, respectively), while using somatostatin analogs and Wirsung's dilatation rarely could be inferred (5 and 6 studies). The use of pancreatic stent was extremely variable within each study and not extractable. The two groups were well matched for gender, type of tumor (PDAC/CP vs. others), "soft pancreas," and not dilated Wirsung rate. The non-BA DtoM population was younger than BA one (WMD −1.59 years; P = 0.004), while the use of somatostatin analogs was more frequent in the BA group than non-BA DtoM one (RR 1.27; P = 0.002). Only in five studies [5, 7, 11, 13, 15], the surgeon's expertise was specified. The quality of included studies was reasonably good, with a median of 15 points (13–23) of MINORS score.

The funnel plot of each endpoint is plotted in Supplementary Fig. 1 Panel a–g. The meta-analytic data are summarized in Fig. 2 (primary endpoint) and Supplementary Fig. 2 Panel a–f (secondary endpoints) and synthesized in Table 2. The "between-study" heterogeneity and publication bias are also reported in Table 2.

The risk of CR-POPF, calculated by extracting each event rate, was significantly lower in the BA arm than in the non-BA DtoM one: The RD was -0.10 (95% CI: −0.16 to −0.04; P = 0.002). The NNT was 9 (every ten patients managed with BA instead of non-BA DtoM, one CR-POPF more could be prevented). The 95% CI indicated that clinical advantage could be more significant in the best scenario (NNT = 7) or more minor in the worst one (NNT = 12). The results did not change rerunning the analysis, using the ORs derived from multivariate models of each study: The odds that CR-POPF was lower in the BA group than in non-BA DtoM one is statistically significant and clinically relevant (OR 0.26; 95% CI: 0.09–0.70; P = 0.017). The heterogeneity of the primary endpoint was not negligible (51% and 75.1%). The "small-study effect" was discovered for RD values with Begg (P = 0.030) and Egger tests (P = 0.044). The regression line, plotted in Fig. 3, suggested the absence in the literature of studies with a small sample size and not favorable for the intervention arm (BA).

No significant differences were found between the two techniques about mortality, B and C DGE, and reoperation rate. Overall morbidity rate was significantly lower than BA with non-BA DtoM (RD = −0.10; 95% CI: −0.18 to −0.02; P = 0.007) with an NNT of 25. The B and C PPH rate was significantly lower in the BA group than the non-BA DtoM one (RD = −0.03; 95% CI −0.06 to −0.01; P = 0.022). However, the clinical relevance of this statistical difference was very small (NNT = 33). The LOS was significantly reduced in patients who received a BA than those undergoing non-BA DtoM (WMD =−4.2 days; −7.1 to −1.2 95% CI; P = 0.006). Among the secondary endpoints, overall morbidity, PPH, and LOS have a not negligible heterogeneity, while none of these results were affected by publication bias.

Meta-regression analysis for CR-POPF is reported in Table 3. The RD decreased when the included studies' quality increased (−0.02 ± 0.01; P = 0.028 ± 0.001). The remaining confounding parameters varying (gender, age, type of lesion (PDAC/CP), soft pancreas, use of somatostatin analogs, Wirsung not dilated type of BA, type of non-BA DtoM, and country) did not explain the heterogeneity. Meta-regression analysis for CR morbidity, PPH, DGE, and hospital stay is detailed in supplementary Tables 3–6.