A systematic search of PubMed, EMBASE and the Cochrane Library databases were conducted from inception til 7 July 2022 by two independent investigators (NWZH, SNL). The search terms used were “keyhole” or “robot” or “laparoscopic” or “minimally invasive” or “daVinci”, and “Pancreaticoduodenectomy” or “Whipple” or “pancreatectomy” or “pylorus-preserving pancreaticoduodenectomy” or “duodenopancreatectomy” or “jejunopancreatectomy” individually or in combination. Search terms used for this review are presented in Supplementary Table S1. A thorough manual search of reference lists in eligible studies was also performed.

Key eligibility criteria included: (1) studies reporting the comparison of surgical techniques in human subjects receiving distal pancreatectomy and (2) studies reporting oncological outcomes (overall survival, positive resection margins, number of lymph nodes examined, tumor recurrence) and (3) studies that included pancreatic ductal adenocarcinoma. Exclusion criteria were: (1) Conference abstracts, reviews, case reports; (2) studies where type of MIDP was not specified; (3) studies that included other types of pancreatic surgeries.

Two reviewers (NZHW and SNL) independently screened and selected potentially eligible studies based on title and abstract. Full-text evaluation was independently performed by two reviewers (NZHW and SNL). Any conflicts between authors were discussed and resolved by a third independent reviewer (DWTY).

As all included studies were observational, we used the Newcastle-Ottowa Scale (NOS) to evaluate the risk of bias. The studies were deemed to have high (<5 stars), moderate (5–7 stars) or low (≥8 stars) risk of bias (Supplementary Table S2).

A frequentist network meta-analysis was employed to compare ODP, LDP and RDP. The network meta-analysis is a statistical approach that combines both direct and indirect evidence to allow for comparison between 3 or more interventions. Relative effects estimates between pairs within the network are more precise than single direct and indirect estimates (18). Treatments were ranked using the P-score provided by the netmeta package (19–21). A probability of ranking of 0.9 was considered high enough to be confidently reported as the correct ranking position of a surgical approach (22). Funnel plots of treatment estimates were visually inspected. Evidence of asymmetry or points lying outside 95% pseudo-confidence limits was interpreted as publication bias. Network plots of treatments (nodes) and comparisons (lines) were generated (Figure 2). Networks were examined for the inconsistency by the fitting of net splitting models (23). A p value of <0.05 was deemed to represent significant inconsistency between the direct and indirect estimate. A separate meta-analysis with meta-regression was performed by considering the proportion of patients with vascular resection when comparing positive resection margins between ODP and MIDP (RDP or LDP).

Hazard ratios (HR) and odds ratios (OR) were reported for categorical outcomes, whilst weighted mean differences (WMD) were reported for continuous data. Mean and standard deviation estimates were derived from studies that reported medians using methods described by Wan et al. and Luo et al. (24, 25). The random effects, restricted maximum likelihood (REML) method was used for the meta-analysis of outcomes. Results were deemed to be statistically significant if the 95% confidence interval did not cross the no-effect line (1 for binary outcomes and 0 for continuous outcomes). A p value of <0.05 was considered to be statistically significant. Data analysis was performed using R Statistical software (R 4.1.3).

The primary outcome was overall survival. Secondary outcomes included positive resection margins, number of lymph nodes examined and tumor recurrence.

The electronic search returned 4,480 publications: 3,297 duplicates were excluded and 1,183 publications were screened. Of those, 1,128 were excluded after reviewing titles and abstracts and forty-four studies were excluded after reviewing full-text articles. An additional six studies were included from searching through reference lists. Nineteeen studies met the eligibility criteria. Two studies were further excluded due to overlapping inclusion periods in the National Cancer Database (NCDB) and one study was excluded due to double reporting of a study (26–28). Fifteen studies were included in the final analysis (Supplementary Table S2) (29–43).

All fifteen studies were retrospective observational studies (Table 1). One study compared all three approaches. Five studies utilised propensity score matching in their analysis (26, 30, 36, 42, 43). ODP and LDP were directly compared in ten studies. RDP and LDP were compared in three studies. ODP and RDP were compared in two studies. A summary of all comparisons made in the network meta-analysis are presented in Table 2.

A total of 9,301 patients were included in this analysis. 1,946 patients underwent LDP, 605 underwent RDP and 6,750 underwent ODP. Baseline characteristics of patients are included in Table 1.

Two studies reported data on Radical Antegrade Modular Pancreatosplenectomy (RAMPS) (36, 37). In both studies, RAMPS were performed more frequently in patients undergoing laparoscopic surgery.

Three studies reported data on vascular resection (37, 41, 42). There was a statistically significant difference in the rates of vascular resection between the three arms in one study with Chopra et al. reporting higher rates of vascular resection in the ODP group (41).

Overall survival was reported in five studies (29–32, 43). Both LDP (HR: 0.761, 95% CI: 0.642–0.901, p = 0.002) and RDP (HR: 0.757, 95% CI: 0.617–0.928, p = 0.008) were associated with better overall survival when compared to ODP. There was no statistically significant difference in overall survival between LDP and RDP (HR: 1.00, 95% CI: 0.793–1.27, p = 0.968) (Figure 3A: Overall survival). RDP and LDP were ranked first and second respectively for overall survival (Table 3). Two studies (29, 32) reported a follow-up duration of five years whilst three studies (30, 31, 43) reported a follow-up duration of three years. In this analysis, only three studies included information on tumor size of the PDAC. Qu et al. reported the median sizes of tumors to be 4.5 cm and 4.4 cm whilst Raoof et al. included tumours of 3.5 cm and 3.7 cm in the RDP and LDP groups respectively (31, 43). In another study by Raoof et al., the majority of tumors were ≥4 cm in both the ODP and LDP groups (30). All studies reported on tumour sizes found them to be comparable in the various arms of comparison.

Resection margins were reported in thirteen studies (29–31, 33–42). Both LDP (OR = 0.803, 95% CI: 0.635–1.02, p = 0.068) and RDP (OR = 0.811, 95% CI: 0.623–1.05, p = 0.115) were not associated with higher rates of positive resection margins. There was no statistically significant difference in positive margins between LDP and RDP (OR =  0.990, 95% CI: 0.730–1.34, p = 0.950) (Figure 3B: Resection Margins). LDP and RDP were ranked first and second respectively for resection margins (Table 3). Positive margins were defined as tumor extension within 1 mm of the margin in 2 studies (34, 41), microscopic evidence of invasion in three studies (33, 39, 42) and microscopic or gross evidence of invasion in two studies (35, 37). Positive margins were not defined in the remaining six studies (29–31, 36, 38, 40).

In a separate analysis of three studies reporting both resection margins and vascular resections, MIDP (LDP or RDP) was not associated with a higher rate of positive resection margins (OR = 0.625, 95% CI: 0.078–5.03, p = 0.434). Meta regression considering proportion of patients with vascular resection did not reach a level of statistical significance (p = 0.425) (Supplementary Figure S1).

Tumor recurrences were reported in six studies. Both LDP (OR = 0.948, 95% CI: 0.647–1.39, p = 0.783) and RDP (OR = 1.13, 95% CI: 0.622–2.06, p = 0.684) were not associated with statistically significant higher rates of tumor recurrence. There was no statistically significant difference in tumor recurrence between LDP and RDP (OR = 0.838, 95% CI: 0.463–1.51, p = 0.561) (Figure 3C: Tumor recurrence). LDP and ODP were ranked first and second respectively for tumor recurrence (Table 3).

The number of examined lymph nodes were reported in thirteen studies. Compared to ODP, LDP (WMD = −0.662, 95% CI: −2.31 to 0.989, p = 0.432) and RDP (WMD = 0.565, 95% CI = −1.90 to 3.03, p = 0.654) did not achieve a statistically significant difference in lymph node examined. Likewise, there was no statistically significant difference in lymph nodes examined between LDP and RDP (WMD = −1.23, 95% CI = −3.53 to 1.08, p = 0.300) (Figure 3D: Examined lymph nodes). LDP and ODP ranked first and second respectively for the number of examined lymph nodes (Table 3).

As some studies included skewed data as described by Shi et al, an additional sensitivity analysis was performed (44). Four studies with significantly skewed data were excluded. There were no statistically significant differences in examined lymph nodes between LDP (WMD = −0.591, 95% CI = −2.64 to 1.46, p = 0.572) and RDP (WMD = 0.526, 95% CI = −2.31 to 3.37, p = 0.719) when compared to ODP. Similarly, there was no statistically significant difference in lymph nodes examined between LDP and RDP (WMD = −1.12, 95% CI = −1.47 to 3.70, p = 0.397).