Robotic versus laparoscopic pancreaticoduodenectomy: a propensity score matching analysis of Taiwan’s national health insurance database

Abstract

Background

Pancreaticoduodenectomy (PD) is the gold standard treatment for localized pancreatic head tumors and has a high surgical morbidity rate. In recent years, minimally invasive approaches, such as laparoscopic pancreaticoduodenectomy (LPD) and robotic pancreaticoduodenectomy (RPD), have gained prominence. However, it remains controversial whether RPD has better clinical outcomes than LPD does.

Materials and methods

This nationwide cohort study used data from Taiwan's National Health Insurance Research Database to investigate patients who underwent RPD or LPD between 2017 and 2021. Full propensity score matching was used to estimate the average treatment effect between the two groups. The primary outcomes included reoperation, optimal outcomes, length of intensive care unit (ICU) stay, and hospital mortality. The secondary outcomes included length of stay (LOS) and 14-day readmission.

Results

In total, 854 patients were included in the study, 590 of whom underwent RPD and 264 of whom underwent LPD. After propensity score matching, RPD was significantly associated with a shorter ICU stay (mean difference [MD], -1 d, 95% confidence interval [CI]: -1.8 to -0.2 day, p = 0.020), lower reoperation rate (risk difference [RD] -6.2%, 95% CI: -9.3% to -3.1%, p < 0.001), and shorter LOS (MD, -3.5 d, 95% confidence interval [CI]: -5.5 to -1.6 day, p < 0.001) than was LPD. There were no statistically significant differences in optimal outcomes, 14-day readmission rates, or hospital mortality. Moderation analysis revealed that high-volume centers had higher rates of optimal outcomes and shorter hospital stays compared to low-volume centers.

Conclusion

Our results suggest that RPD has better postoperative outcomes than does LPD. Additionally, high-volume centers have better clinical outcomes than low-volume centers do.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12893-026-03626-4.

Introduction

Pancreaticoduodenectomy (PD), the definitive treatment for localized tumors of the pancreatic head, is a significant clinical challenge given its insidious presentation and associated high morbidity rate [1–3]. Minimally invasive PD (MIPD) emerged as an alternative to open PD (OPD) after the first introduction of laparoscopic PD (LPD) in 1994 by Ganger and Pomp [4]. Although LPD has been reported to reduce blood loss, lower postoperative morbidity rates, shorten the length of stay (LOS), and expedite recovery [5, 6], it has been performed in high-volume centers due to the extended learning curve. These limitations limit its widespread adoption in clinical settings.

Recently, the dawn of advanced robotic surgical systems, the da Vinci XI platform, has offered a revolutionary approach to minimally invasive complex surgical procedures. Robotic surgical system have several advantages over laparoscopic instruments, including improved three-dimensional visualization, increased dexterity of the Endo Wrist instruments, increased degrees of freedom of instrumental motion, and improved ergonomics [7, 8]. Through this instrumental improvement, Giulianotti first reported robotic PD in 2003 [7]. Since then, emerging evidence from high-volume robotic centers has indicated that robotic pancreaticoduodenectomy (RPD) may circumvent some of the challenges inherent to LPD, thus optimizing outcomes [9, 10].

Recently, cumulative reports have suggested that RPD is comparable to OPD in terms of morbidity and mortality but result in less blood loss and a shorter LOS [11]. In addition to these advantages over OPD, a recent systematic review revealed that RPD has lower conversion and transfusion rates than does LPD; however, no significant differences in postoperative complications, incidence of pancreatic fistulae, or R0 resection rates were noted between the two groups [12]. Using a pancreas-targeted ACS NSQIP dataset, Khachfe et al. recently reported that RPD adoption has increased compared with LPD adoption in recent years and that, compared with LPD, RPD is associated with decreased overall complications, decreased serious complications, and increased optimal outcomes from 2018—2019 [13]. Although a Western study revealed that RPD was associated with better composite outcomes than was LPD, whether RPD has better specific outcomes than does LPD remains an unanswered question. In the present study, we aimed to investigate whether RPD is superior to LPD in terms of the reoperation rate, optimal outcome, length of intensive care unit (ICU) stay, and hospital mortality rate via a nationwide database in Taiwan. In addition, we compared the clinical outcomes of RPD and LPD between high-volume centers and low-volume centers.

Materials and methods

Study design

We conducted a nationwide cohort study using data from Taiwan's National Health Insurance Research Database (NHIRD) for all patients who underwent robotic or laparoscopic PD. The NHIRD has information about the health of more than 23 million people, which is more than 99% of Taiwan's population. The accuracy of diagnosing major diseases via the NHIRD has been validated in many studies [14]. For monitoring and reimbursement purposes, Taiwan’s National Health Insurance Administration (NHIA) regularly samples and reviews data from all hospitals to ensure the accuracy and justification of the application fees. This study was approved by the Institutional Review Board of our institution. In addition, the work has been reported in line with the STROCSS criteria [15].

Study population

Although the NHIA reimbursement does not cover surgical fees for RPDs due to Taiwan's Health Technology Assessment (HTA) policy, the NHIA uses HTA codes to register these procedures. The HTA policy requires all hospitals to register RPDs. Therefore, we used the HTA code for RPD (N26014) to identify all hospitalized patients who had undergone RPD. Only 42 patients were identified in the registration database in 2016. Therefore, patients who underwent RPD between 2017/01/01 and 2021/12/31 were retrieved from the Taiwan NHIRD. In addition, we identified patients who underwent LPD via Taiwan’s Health Insurance Payment Codes (75412 B + 28014 C, 75417 B + 28014 C, 75427B or 75430B). Patients aged < 20 years; those who underwent multiple major surgical procedures, such as concomitant colectomy, hepatectomy, nephrectomy, or hysterectomy; those with a primary diagnosis of colon cancer, a primary diagnosis of gastric cancer, or cholangiocarcinoma; those with human immunodeficiency virus infection; and those with severe liver disease were excluded. Thus, the primary diagnoses of the patients in the study group were primary etiologies in the ampulla of Vater, distal common bile duct, pancreatic head, and duodenum. The diagnosis of malignancy was confirmed by both specific admission ICD-10 codes and inclusion in the Registry for Catastrophic Illness Patient Database (RCIPD) [14]. For patients with malignancies to be registered in the RCIPD, surgical pathology confirmation must be performed.

Outcome measurements

The primary outcomes included a) reoperation, b) optimal outcome, c) length of ICU stay, and d) hospital mortality. The optimal pancreatic surgery outcome was defined as a composite binary outcome including the absence of serious complications, prolonged stay (30 days), or 14-day readmission [13, 16]. Serious complications were defined as an ICU stay of > 48 h, percutaneous drainage, reoperation, and hospital mortality [13, 16]. We defined 30 days as a prolonged length of stay according to the administration's definition of ultralong-term hospitalization. The secondary outcomes were length of hospital stay (LOS) and 14-day readmission.

In addition to the overall reoperation rate, we performed a secondary analysis focusing on target reoperations that were likely attributable to complications following pancreaticoduodenectomy. These included procedures such as revision of anastomoses, repair of perforation, drainage of intra-abdominal abscess, and re-exploration for postoperative bleeding on the basis of a predefined set of procedure codes. The corresponding procedure codes included 73001B, 73002B, 73003B, 72019B, 72021B, 72033B, 73033B, 73039B, 73042B, 75401B, 75411B, 75609B, 75801 C, 75802B, 75805B, and 75816B.

Covariate assessment

In the present study, we aimed to estimate the average treatment effect (ATE) in the population—the average treatment effect among all eligible patients in the study population [17, 18]. The ATE is the difference in average outcomes, where all patients were treated via the robotic approach vs. all patients were treated via the laparoscopic approach. For the use of observational data, we calculated the propensity score via logistic regression to estimate the probabilities of assigning a patient to the treated cohort, given confounding variables such as age, sex, comorbidities, level of hospital accreditation (medical center vs. nonmedical center), disease etiologies, advanced stages of cancer, annual MIPD volume of the hospital, annual MIPD volume of the individual surgeon, and whether the patient was included in the RCIPD [17, 19, 20]. Before propensity score matching, we excluded patients treated in the RPD or LPD-only institutes, which might have violated the positivity assumption of the treatment. Confounding factors are common causes of exposure and outcomes. Therefore, we chose these confounders as controls on the basis of these rules [20].

The Charlson Comorbidity Index (CCI) score was used to evaluate the severity of comorbidities. The CCI was calculated via the ICD-10 Coding Algorithms for Charlson comorbidities [21]. The adjusted age-adjusted Charlson Comorbidity Index (AACCI) score was calculated as another severity score for comorbidities [22, 23]. The AACCI is a version of the NHIRD risk measure for perioperative mortality before cancer surgery and has been proven to be accurate [22, 23]. We defined four levels for the AACCI according to previous studies: score 0–3, 4–7, 8–11, and ≧12.

Furthermore, we defined disease etiologies as invasive carcinomas of the ampulla of Vater, distal common bile duct, pancreatic head, and other origins such as neuroendocrine tumors. In addition to invasive carcinoma, we attributed carcinoma in situ to its benign etiology. For advanced-stage of cancer, the discharge code for metastatic solid tumors is used for lymph node metastasis in patients with cancer. Finally, the level of hospital accreditation and its inclusion in the RCIPD were validated via data from the NHIRD with little chance of misclassification.

Statistical analysis

Continuous variables are summarized as medians with interquartile ranges (IQRs), and categorical variables are presented as frequencies and percentages. Baseline characteristics were compared between the robotic and laparoscopic groups via the chi-square or Wilcoxon rank-sum test, as appropriate. To estimate ATE, we used full matching without replacement to conduct propensity score matching. After matching, the covariate balance between the two groups was evaluated. The love plot displays the balance of each covariate, whereas the mirror histogram displays the propensity score distribution [24]. After evaluating the matching results, we conducted a regression of the outcome on the treatment and covariates in the matched sample (including the matching weights) and estimated the treatment effect via g-computation as implemented via the function of “comparisons” in the R contributed packages, “marginaleffects” [25, 26]. We included the covariates used in matching in the effect estimation and interaction with the treatment. This approach could provide additional robustness against slight imbalances after matching and improve the precision. Additionally, we used cluster-robust standard errors for all analyses after matching with pair membership as a clustering variable. We also performed a moderation analysis stratified by the hospital volume of MIPD (high-volume centers vs. low-volume centers) to examine whether the effect on outcomes differed between high-volume centers and low-volume centers [27]. We defined high-volume centers of MIPD as those with more than 20 MIPDs per year during the study period [13, 28]. All reported confidence intervals (CIs), and tests were 2-sided with a 5% significance level. All analyses were performed with R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria) with the contributed packages “MatchIt” [29], “tableone” [30] “cobalt” [24] and “marginaleffects” [26].

Results

We recruited 1132 patients between January 2017 and December 2021, while 854 were included in the final analysis on the basis of the defined inclusion and exclusion criteria. Overall, 590 RPDs and 264 LPDs were identified for propensity score matching analysis (Fig. 1). Since 2017, the annual number of RPDs has surpassed that of LPDs, reaching approximately 10% to 13% of the total PDs (SDCs, Fig. 1). Overall, RPDs and LPDs accounted for one quarter of the total number of PDs in 2021. The distributions of the RPDs and LPDs during the study period were quite different in the study sample (SDCs, Fig. 2). The RPDs were greater than the LPDs in the largest-volume center, whileas the other RPDs were different.

Fig. 1.

Flow diagram of study populations

Patient demographics and baseline characteristics

The baseline demographics and characteristics of the RPD and LPD groups are summarized in Table 1. Both groups were comparable in terms of age, sex, and comorbidities. The RPD group had a greater annual MIPD volume at the hospital (70.6 vs. 14.2, p < 0.001) and a greater annual MIPD volume at the surgeon (21 vs. 7.6, p < 0.001). In addition, the histological etiology differed marginally between the two groups. Compared with the LPD group, the RPD group had a significantly greater LN metastasis rate (31.4% vs. 21.2%, p = 0.003).

Table 1.

Characteristics of patients with RPD and LPD before and after propensity score matching

	LPD (n = 264)	RPD (n = 590)	P value	Unadjusted difference	Adjusted difference b
Age, y, median (IQR)	66 (59, 72)	66 (58, 73)	0.837	−0.0407	0.0532
Sex, male, (%)	146 (55.3)	314 (53.2)	0.624	0.0208	0.0052
CCI, median (IQR)	2 (2, 4)	3 (2, 8)	0.449	0.1429	0.0227
AACCI, median (IQR)	6 (4, 8)	6 (4, 11)	0.388
AACCI 3	64 (24.2)	156 (26.4)	0.331	0.0220	0.0049
AACCI	124 (47.0)	245 (41.5)		−0.0544	−0.0255
7 < AACCI	76 (28.8)	189 (32.1)		0.0325	0.0206
CPD, n (%)	11 (4.2)	14 (2.4)	0.223
Liver diseases, n (%)	19 (7.2)	52 (8.8)	0.511
Congestive heart failure, n (%)	2 (0.8)	8 (1.4)	0.684
Diabetes mellitus, n (%)	62 (23.5)	155 (26.3)	0.436
Renal diseases, n (%)	13 (4.9)	18 (3.1)	0.248
Medical center, n (%)	236 (89.4)	503 (85.3)	0.126	−0.0414	0.0047
Histology, n (%)			0.086
Pancreatic cancer	101 (38.3)	213 (36.1)		−0.0216	0.0289
Ampullary cancer	70 (26.5)	140 (23.7)		−0.0279	−0.0281
CBD cancer	31 (11.7)	50 (8.5)		−0.0327	−0.0097
Benign	48 (18.2)	131 (22.2)		0.0402	0.0292
Other cancers	14 (5.3)	56 (9.5)		0.0419	−0.0202
RCIPD, n (%)	210 (79.5)	446 (75.6)	0.239	−0.0395	−0.0310
Metastasis, n (%)	56 (21.2)	185 (31.4)	0.003	0.1014	0.0194
Annual MIPDs of the hospital, median (IQR)	14.2 (6.1, 17.2)	70.6 (21, 70.6)	< 0.001	1.0077	−0.0694
Annual MIPDs of the individual surgeon, median (IQR)	7.6 (3.4, 16.6)	21 (6.4, 23)	< 0.001	0.9522	0.0831
Propensity score a				1.4191	0.0009
LOS, median (IQR), day	29.50 (20.8, 41)	25 (17, 35)	< 0.001
ICU stay, median (IQR), day	2 (0, 5)	0 (0, 1)	< 0.001
14-day readmission, n (%)	28 (10.6)	63 (10.7)	1
Reoperation, n (%)	25 (9.5)	20 (3.4)	< 0.001
Hospital mortality, n (%)	7 (2.7)	11 (1.9)	0.63
Optimal outcome	83 (31.4)	301 (51.0)	< 0.001

RPD robotic pancreatoduodenectomy, LPD laparoscopic pancreatoduodenectomy, IQR interquartile range, ICU intensive care unit, LOS length of stay, CBD common bile duct, CPD chronic pulmonary disease, CCI Charlson Comorbidity Index, AACCI adjusted age-adjusted Charlson Comorbidity Index, RCIPD Registry for Catastrophic Illness Patient Database; MIPD, minimally invasive pancreatoduodenectomy

^aWe included age, sex, CCI, AACCI, histology, RCIPD, metastasis, medical center, the annual volume of MIPDs by the hospital, and the annual volume of MIPDs by the surgeon in the propensity score matching model

^bUnadjusted difference and adjusted difference indicate covariate balance before and after propensity score matching, respectively. Values of adjusted standardized mean differences close to zero indicate good balance

Postoperative and short-term outcomes before matching

Compared with the LPD group, the RPD group had shorter ICU stays (0 days vs. 2 days, p < 0.001), shorter LOSs (25 days vs. 29.5 days, p < 0.001), lower reoperation rates (3.4% vs. 9.5%, P = 0.001), and improved optimal outcomes (51.0% vs. 31.4%, p < 0.001). However, RPD was comparable with respect to the 14-day readmission rate and hospital mortality in the LPD group.

Covariate balance and distributions after matching

The covariate balance before and after propensity score matching is shown in Fig. 2 and Table 1, where the absolute standardized mean differences for each covariate were less than 0.1 (Fig. 2A, Table 1), and the distributional balance significantly improved after matching between the two groups (Fig. 2B). The propensity score, annual MIPD volume of the hospital, annual MIPD volume of the surgeon, and incidence of LN metastasis were greatly improved after full matching.

Fig. 2.

Covariate balance and distributional balance before and after propensity score matching. A Covariate Balance Before and After Propensity Score Matching. The red circles denote the original values, and the blue triangles represent the values after full matching. Figure 2 A shows the covariate balance before and after full matching in terms of the absolute standardized mean differences. The absolute standardized mean differences for each covariate were less than 0.1 after matching. CBD, common bile duct; CCI, Charlson Comorbidity Index; AACCI, adjusted age-adjusted Charlson Comorbidity Index; RCIPD, Registry for Catastrophic Illness Patient Database; MIPD, minimally invasive pancreatoduodenectomy, MIPD_H, the annual MIPD volume of the hospital; MIPD_S, the annual MIPD volume of the surgeon. B Distributional Balance Before and After Propensity Score Matching. The red and cyan bars denote the laparoscopic and robotic groups, respectively. The left panel of Fig. 2B shows the distribution balance before the full matching. The right panel of Fig.2B shows the distribution balance after the full matching. Distributional balance significantly improved after matching between the two groups

ATE after matching

After matching, RPD was associated with significantly shorter ICU stays (mean difference [MD], −1 d, p = 0.020), lower reoperation rates (risk difference [RD], −6.2%, p < 0.001), and a shorter LOSs (MD, −3.5 d, p < 0.001) than were found in the LPD group (Table 2). Nevertheless, the optimal outcomes, the 14-day readmission rates, and hospital mortality rates were not significant different after matching.

Table 2.

Outcome comparisons between the RPD and LPD groups after propensity score matching

Outcomes	RPD (ESS = 383.56) vs LPD (ESS = 70.48)	p value
ICU stay, MD (95% CI), day	−1 (−1.8, −0.2)	0.020
Reoperation, RD (95% CI), %	−6.2 (−9.3, −3.1)	< 0.001
Optimal outcome, RD (95% CI), %	5.4 (−2.5, 13.3)	0.182
Hospital mortality, RD (95% CI), %	0.4 (−1.1, 1.9)	0.608
LOS, MD (95% CI), day	−3.5 (−5.5, −1.6)	< 0.001
14-day readmission, RD (95% CI), %	1.5 (−2.7, 5.7)	0.493

ESS effective sample size, CI confidence interval, RR risk ratio, RD risk difference, MD mean difference, RPD robotic pancreatoduodenectomy, LPD laparoscopic pancreatoduodenectomy, ICU intensive care unit, LOS length of stay

Moderation analysis stratified by high-volume center versus low-volume center

Moderation analysis revealed that the outcome differences in optimal outcome (RD between subgroups 34.6%, p < 0.001) and LOS (MD between subgroups −5.4 days, p = 0.039) between the RPD and LPD groups were significantly better in high-volume centers than in low-volume centers (Table 3). () The outcomes of the RPD and LPD groups were not significantly different in terms of ICU stay, reoperation, hospital mortality, or 14-day readmission between the high-volume centers and low-volume centers. In high-volume centers, the RPD was better than the LPD in terms of ICU stay, reoperation, optimal outcome, and LOS. In low-volume centers, RPD was associated with a significantly lower rate of optimal outcomes compared with LPD (RD, − 18.9%; 95% CI, − 30.6% to − 7.1%).

Table 3.

Analysis of outcomes comparing RPD and LPD stratified by high-volume center versus low-volume center

Outcomes	RPD vs LPD	p value
ICU stay, MD (95% CI), day
High-volume center	−2.7 (−2.0, −3.4)	< 0.001
Low-volume center	−0.4 (−5.4, 4.7)	0.888
Subgroup difference	−2.3 (−7.4, 2.8)	0.376
Reoperation, RD, % (95% CI)
High-volume center	−4.3 (−0.6, −8.0)	0.021
Low-volume center	2.9 (−10.1, 15.8)	0.663
Subgroup difference	−7.2 (−20.7, 6.3)	0.296
Optimal outcome, RD, % (95% CI)
High-volume center	15.7 (8.3, 23.1)	< 0.001
Low-volume center	−18.9 (−30.6, −7.1)	0.002
Subgroup difference	34.6 (20.7, 48.5)	< 0.001
Hospital mortality, RD, % (95% CI)
High-volume center	−1.2 (−3.2, 0.8)	0.236
Low-volume center	3.1 (−8.9, 15.1)	0.612
Subgroup difference	−4.3 (−16.5, 7.9)	0.488
LOS, MD (95% CI), day
High-volume center	−3.2 (−1.1, −5.2)	0.003
Low-volume center	2.2 (−2.4, 6.9)	0.349
Subgroup difference	−5.4 (−0.3, −10.5)	0.039
14-day readmission, RD, % (95% CI)
High-volume center	−1.1 (−6.1, 3.8)	0.648
Low-volume center	6.0 (−4.2, 16.1)	0.249
Subgroup difference	−7.1 (−18.4, 4,2)	0.217

CI confidence interval, RD risk difference, MD mean difference, RPD robotic pancreatoduodenectomy, LPD laparoscopic pancreatoduodenectomy, ICU intensive care unit, LOS length of stay

Reoperation events may be attributable to pancreaticoduodenectomy

The reoperation rate attributable to complications directly related to pancreaticoduodenectomy was 4.24% in the RPD group and 12.12% in the LPD group. The reoperation rate in LPD group was significantly higher than in RPD group (RR = 2.86, 95% CI 1.69–4.83). Details of the reoperation events are summarized in Table 4.

Table 4.

Reoperation events may be attributable to pancreaticoduodenectomy

Group	Procedure Description	RPD (n=590)	LPD (n=264)
Drainage of Intra-abdominal Abscess		16	14
	Drainage of intraabdominal abscess for acute perforation peritonitis	12	7
	Drainage of pancreatic abscess or cyst or pancreatitis	0	3
	Drainage of subphrenic abscess	3	4
	Peritoneal lavage	1	0
Re-exploration for Postoperative Bleeding		1	0
	Re-exploration for postgastrectomy bleeding	1	0
Repair of Perforation		0	2
	Gastrorrhaphy	0	1
	Repair of intestinal perforation	0	1
Revision of Anastomoses		2	3
	Closure of enterostomy or colostomy anastomosis, by laparotomy	1	0
	Revision of colostomy or enterostomy complicated, deep	0	2
	Revision of gastrojejunostomy	1	1
Other		6	13
	Exploratory laparotomy	4	12
	Suture of abdominal wall for evisceration of dehiscence for secondary closure	0	1
	Total pancreatectomy (95%)	2	0
Total		25 (4.24%)	32 (12.12%)

Discussion

In the present nationwide study, we used Taiwan’s NHIRD to investigate the ATE between RPD and LPD. Our results demonstrated that the RPD was associated with a reduced reoperation rate, shorter ICU stay, and a shorter LOS. The optimal outcomes, hospital mortality rates, and 14-day readmission rates did not differ significantly between the two groups. The clinical outcomes of the high-volume centers were significantly better than those of the low-volume centers in terms of ICU stay, reoperation rate, and optimal outcomes. Notably, the number of reoperations attributable to complications directly related to pancreaticoduodenectomy also significantly differed between the groups, with a lower incidence observed in the RPD group than in the LPD group. On the basis of evidence, Taiwan’s NHIA has covered surgical fees for robotic pancreatic surgery since March 2023.

Recently, robotic surgical platforms have been used in various domains [31, 32]. Along with increased familiarity with the da Vinci surgical system, surgeons have reported the clinical outcomes of complex surgical procedures, such as RPD, using this platform. In Taiwan, the LPD program generally starts earlier than the RPD program in most institutes. To our knowledge, the RPD program was adopted in Taiwan from 2014 to 2016 in several centers. In the present study, RPD increased in 2016 and surpassed LPD in 2017. The overall percentage of MIPD to overall PDs reached approximately 20% to 25% from 2017–2021. This phenomenon has been observed in both American and Dutch studies [13, 33]. Thus, the overall trends of the RPD and LPD in Taiwan are consistent with the Western data. Considering the adoption period of the RPD program, we used data between 2017 and 2021 in the present study, excluding data from the earlier adoption period of the LPD and RPD programs.

The recently updated Miami guidelines revealed insufficient data to routinely advocate MIPD adoption in patients with pancreatic head tumors [34]. Increasing evidence has shown that RPD is comparable to OPD in terms of perioperative complications and overall survival but results in less blood loss and a shorter LOS [28, 35]. We hypothesized that RPD might have better clinical outcomes than LPD does on the basis of the inherent advantages of the robotic platform, including its stable enhanced three-dimensional camera, elimination of tremor signals, and 540-degree wrist articulation [36]. Our results are consistent with those of the American ACS NSQIP study for optimal outcomes; compared with RPD, LPD was less likely to achieve optimal outcomes (OR 0.56, 95% CI, 0.39 to 0.78, p = 0.001) from 2018–2019. From 2014 to 2019, the optimal outcomes in the American study significantly increased, from 48.2% to 57.8%. Overall, our optimal outcome was 51% in the RPD group, which is similar to that reported in the American ACS NSQIP study. Nevertheless, we first demonstrated that RPD was significantly associated with a lower reoperation rate than LPD (RD 6.2%, 95% CI 3.1% to 9.3%, p < 0.001) in this large nationwide study. This result may depend on the institutional learning curve.

There is consistent agreement among high-volume centers of MIPD (defined as > 20 MIPD per year) [13, 28]. In addition, the learning curve for RPD was defined as > 40 RPDs for the Chinese group [13, 34] and 20–40 RPDs for the international consensus [11]. Our current study revealed that high-volume centers (defined as those with > 20 MIPD per year) were associated with better outcomes than low-volume centers were in the moderation analysis. Compared with low-volume centers, high-volume centers in Taiwan had more promising effects on the optimal outcomes, and length of stay between the RPD and LPD groups. Our findings are reasonable and comparable to those of a Western study [13]. These findings raise the issue of standard training programs that surgical societies should consider. The LAELAPS-3 study revealed that a multicenter RPD training program with sufficient surgical volume was feasible without negatively impacting the learning curve of clinical outcomes [34]. Considering these issues, Taiwan’s NHIA requires robotic surgeons to obtain approval from the surgical society for RPD as part of the reimbursement criteria. When patients are discharged from the hospital, Taiwan’s NHIA requests that hospitals submit outcome data to the registered database for quality control, which affects the maintenance of reimbursement criteria to medical centers.

Earlier reports and meta-analyses indicated no significant differences in LOS, complication rates, or serious complication rates between the RPD and LPD groups [12, 36]. However, these studies did not consider the learning curve of RPD or its widespread adoption. Two large Asian studies revealed that RPD had advantages in terms of operative time and blood loss compared with OPD and similar overall survival among patients with pancreatic cancer between RPD and OPD after passing the learning curve [28, 35]. These findings were consistent with those of the American ACS NSQIP study [13]. The ACS NSQIP study revealed that the clinical outcomes of RPD were inferior to those of LPD from 2014–2015, but the results reversed from 2018–2019 after the trend of RPD increased with time. These findings were consistent with our moderation analysis. The ICU stay, and optimal outcomes of RPD were inferior to those of LPD in low-volume centers, which suggested their beginning adoption of the RPD program. Therefore, comparative studies investigating the clinical outcomes between the RPD and OPD groups or the RPD and LPD groups should consider the hospital volumes and learning curves of the surgical teams and the spread width of the RPD. In addition, such findings suggest that surgical societies may consider adopting the RPD programs such as the LAELAPS-3 study [33].

The present study has several strengths. First, our study used a population-based sample from Taiwan and full matching in the entire sample for estimating the ATE via the g-computation formula between the two groups [26, 29]. This approach allowed us to use our results as a representative sample in Taiwan. Our stakeholders can formulate policy interventions on the basis of these results. Second, in a large sample, our results revealed significant clinical outcomes, such as the reoperation rate. Third, our study used risk or mean differences as effect sizes. These findings can easily be interpreted in clinical scenarios. Finally, high-volume centers and institutional learning curves should be considered when stakeholders want to pay for an RPD. Otherwise, the certificates of the robotic surgeons should be accredited.

Our study also has several limitations. First, although this study used propensity score matching to account for confounders, it was not a randomized controlled trial. Therefore, we cannot exclude residual and unobserved founders. Second, we did not obtain perioperative data for analysis. These analyses were conducted via the American ACS NSQIP study [13]. Third, we did not have the cancer stages for each patient. Nevertheless, we believe that these results remain unchanged. Fourth, we did not obtain detailed information on clinical demographics. Therefore, we were unable to identify these factors. Finally, we did not analyze the economic outcomes. However, improved clinical outcomes from RPD are likely to reduce the health insurance burden compared with LPD. We will further explore the financial results in future studies using Taiwan’s NHIA-registered data.

Conclusion

Our results suggest that RPD has better postoperative outcomes than does LPD, including reduced reoperation rate, a shorter ICU stay, and a shorter LOS. In addition, high-volume centers have significantly better clinical outcomes than low-volume centers do. Further research is needed to validate these findings, investigate functional outcomes, analyze long-term survival, and explore their long-term economic consequences.

Abbreviations

PD

Pancreaticoduodenectomy

LPD

Laparoscopic pancreaticoduodenectomy

RPD

Robotic pancreaticoduodenectomy

MIPD

Minimally invasive pancreaticoduodenectomy

OPD

Open pancreaticoduodenectomy

NHIRD

National Health Insurance Research Database

NHIA

National Health Insurance Administration

ICU

Intensive care unit

LOS

Length of stay

CCI

Charlson Comorbidity Index

AACCI

Adjusted age-adjusted Charlson Comorbidity Index

RCIPD

Registry for Catastrophic Illness Patient Database

ATE

Average treatment effect

MD

Mean difference

RD

Risk difference

CI

Confidence interval

OR

Odds ratio

Authors’ contributions

(S.W.C, Y.J.W, Y.C.L., T.T.K., C.L.S, T.S.H.: conception and design; Y.C.L., T.T.K., C.L.S: acquisition of data; S.W.C, T.S.H.: analysis and interpretation of data; S.W.C, Y.J.W.: drafting of manuscript; Y.C.L., T.T.K., C.L.S, T.S.H.: critical revision of the manuscript for important intellectual content; Y.C.L., T.T.K., C.L.S, T.S.H.: administrative, technical, or material support; C.L.S, T.S.H: final approval to be submitted. All the authors reviewed the manuscript for important intellectual content and approved its final version.)

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

The data and statistical codes supporting the findings of this study are available upon request from the corresponding author. These data are not publicly available because of administrative restrictions.

Declarations

Ethics approval and consent to participate

Ethics approval and consent to participate This study was approved by the Institutional Review Board of the Chang Gung Medical Foundation (IRB No. 202201612B1). All methods were performed in accordance with the relevant guidelines and regulations, including the Declaration of Helsinki. The requirement for informed consent was waived by the Institutional Review Board of the Chang Gung Medical Foundation because the study utilized de-identified and encrypted claims data for medical research.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.