An evidence-based model for predicting conversion to open surgery in minimally invasive distal pancreatectomy

Abstract

Background

Intraoperative conversion to open surgery is an adverse event during minimally invasive distal pancreatectomy (MIDP), associated with poor postoperative outcomes. The aim of this study was to develop a model capable of predicting conversion in patients undergoing MIDP.

Methods

A total of 352 patients who underwent MIPD were included in this retrospective analysis and randomly assigned to training and validation cohorts. Potential risk factors related to open conversion were identified through a literature review, and data on these factors in our cohort was collected accordingly. In the training cohort, multivariate logistic regression analysis was performed to adjust the impact of confounding factors to identify independent risk factors for model building. The constructed model was evaluated using the receiver operating characteristics curve, decision curve analysis (DCA), and calibration curves.

Results

Following an extensive literature review, a total of ten preoperative risk factors were identified, including sex, BMI, albumin, smoker, size of lesion, tumor close to major vessels, type of pancreatic resection, surgical approach, MIDP experience, and suspicion of malignancy. Multivariate analysis revealed that sex, tumor close to major vessels, suspicion of malignancy, type of pancreatic resection (subtotal pancreatectomy or left pancreatectomy), and MIDP experience persisted as significant predictors for conversion to open surgery during MIDP. The constructed model offered superior discrimination ability compared to the existing model (area under the curve, training cohort: 0.921 vs. 0.757, P < 0.001; validation cohort: 0.834 vs. 0.716, P = 0.018). The DCA and the calibration curves revealed the clinical usefulness of the nomogram and a good consistency between the predicted and observed values.

Conclusion

The evidence-based prediction model developed in this study outperformed the previous model in predicting conversions of MIDP. This model could contribute to decision-making processes surrounding the selection of surgical approaches and facilitate patient counseling on the conversion risk of MIDP.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s00464-024-11216-9.

Distal pancreatectomy (DP), also known as left pancreatectomy, is a procedure for pancreatic lesions located in the body or tail of the pancreas. Amidst the mounting emphasis on reducing surgical trauma, minimally invasive distal pancreatectomy (MIDP) has emerged as the prevalent modality for this procedure, which encompasses both laparoscopic distal pancreatectomy (LDP) and robotic distal pancreatectomy (RDP). A recent international survey highlighted that 80% of distal pancreatectomies are now conducted using minimally invasive techniques [1]. Against this backdrop, identifying suitable candidates for MIDP and devising strategies to avert unnecessary conversions to laparotomy are of significant importance in ensuring optimal patient outcomes.

Characterized by their location in the retroperitoneum and close proximity to adjacent organs and major vessels, pancreatic tumors are often found to infiltrate the surrounding structures as they grow, thereby complicating minimally invasive pancreatic resection. The reported conversion rate for MIDP spans from 13.0% to 32.7%, with the principal reasons being vascular involvement, oncological concern, and massive bleeding [2]. The MIDP technique gains notable popularity due to its benefits of reduced hospital stays and accelerated functional recovery compared to open distal pancreatectomy (ODP) [3–5]. Conversely, when MIDP attempts necessitate conversion to laparotomy, patients encounter greater blood loss, extended operation durations, and an increased risk of postoperative complications compared to those undergoing scheduled ODP [6]. Moreover, such conversions to open surgery notably escalate the financial costs associated with hospitalization, relative to successfully executed minimally invasive procedures [7].

For patients for whom minimally invasive surgery was considered, preoperative evaluation of the risk of conversion provides insight into the selection of a proper surgical approach. In endoscopic surgeries such as video-assisted thoracic surgery (VATS) [8], laparoscopic colorectal surgery [9], and laparoscopic biliary surgery [10], models for predicting conversion surgery have been reported. In MIDP, a predictive tool for LDP conversion was proposed by Casadei et al. [11] in 2021, but its performance and stability have not yet been validated. This model comprised four risk factors: sex, BMI, type of pancreatic resection, and tumor location. Notably, other potential risk factors like surgeon experience and anatomical intricacies of the tumor were not included in this model, suggesting a need to incorporate such variables for a more holistic assessment of the conversion risk.

To date, many preoperative risk factors related to conversion in patients undergoing MIDP have been proposed, including male sex [11, 12], large tumor size, smoking [13], close relationship between tumor and major vessels [6, 14], which provide the evidence for model development. However, the associations of several factors with open conversion of MIDP are still ambiguous. For example, large tumor size was found to be a significant risk factor for conversion in the study of Jiang et al. [7], while the results of Liu et al. [15] indicated that tumor size was not related to the risk of conversion. Of note, the multivariate analyses conducted in these two studies differed considerably in the variables they included, and some previously reported risk factors were not integrated into their analyses. On this premise, the effect of confounders could not be properly adjusted, and the results may indicate an unreliable relationship between risk factors and the odds of conversion [16].

Therefore, the present study aims to: (1) reevaluate the reported risk factors associated with open conversion of MIDP; (2) develop and validate a novel predictive model based on the identified independent predictors; and (3) compare the performance of this new model against that of the existing model.

Methods

Patients

We retrospectively reviewed consecutive patients who underwent planned MIDP at Fujian Medical University Union Hospital between January 2016 and October 2023. Patients who underwent planned diagnostic laparoscopy followed by an ODP was not considered as patients scheduled for MIDP. Exclusion criteria were as follows: (1) patients previously receiving pancreatic resection; (2) patients with incomplete data regarding conversion; and (3) conversions due to malfunction of minimally invasive instruments. The cohort was further divided into training cohort and validation cohort at a 7:3 ratio using random numbers generated from the statistical software. The flow chart is shown in Fig. 1. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Ethics Committee of Fujian Medical University Union Hospital. Written informed consent was waived due to the retrospective nature, with the approval of the ethics committee. We reported this study in line with the TRIPOD checklist [17].

Fig. 1.

Flow chart of the present study. MIDP minimally invasive distal pancreatectomy, MIDPS minimally invasive distal pancreatectomy with splenectomy, ROC receiver operating characteristic

Literature review and data collection

A systematic literature search was conducted to identify reported risk factors related to converted MIDP on PubMed, Embase, and Cochrane Library databases between inception and December 2023. The search string is “((laparoscop* OR robot*) OR (minimally invasive)) AND ((distal pancreatectomy) OR (left-sided pancreatectomy) OR (left pancreatectomy)) AND (conversion)”. Studies involving the investigation of preoperative risk factors related to open conversion in MIDP were reviewed; editorials or conference abstracts were excluded. Figure 1 depicts the literature screening details. Only preoperative factors that were identified by multivariate analysis or randomized controlled trials were included. Data on the final selected factors were extracted from the hospital database for model construction. Moreover, the demographic and clinical variables were also obtained, including age, sex, height, weight, body mass index (BMI), history of smoking, hypertension, diabetes mellitus, abdominal surgery history, laboratory examination (albumin, total cholesterol, triglyceride, platelet count, prothrombin time—international normalized ratio [INR], activated partial thromboplastin time [aPTT], fibrinogen, and hemoglobin), perioperative outcomes (conversion, operative time, red blood cell transfusion, reoperation, intraabdominal infection, postoperative pancreatic fistula [POPF], days until diet start, postoperative hospital stay, and unplanned readmission).

Definitions

Conversion was defined as an attempt at minimally invasive resection followed by an unplanned laparotomy for other reasons rather than specimen extraction. Tumor close to major vessel was demonstrated by preoperative contrast-enhanced computed tomography or magnetic resonance imaging, including: (1) a tumor invaded or pressed the splenic artery or splenic vein; (2) a tumor located less than 2 cm away from the origin of the splenic artery. The MIDP experience was quantified based on the case number of MIDP performed. Multidimensional CUSUM analysis incorporating both intraoperative and postoperative metrics was performed to depict the learning curve of MIDP and determine the case number needed for “high MIDP experience”. The formula of the learning curve is as follows: $\text{CUSUM}={\sum }_{(i=1)}^n[(X{i}_{\text{time}}-u_1)+(X{i}_{(\text{blood loss})}-u_2)+(X{i}_{\text{conversion}}-u_3)+(X{i}_{\text{reoperation}}-u_4)+(X{i}_{\mathit{POPF}}-u_5)+(X{i}_{\text{infection}}-u_6)+(X{i}_{\mathit{LOS}}-u_7)]$ where LOS represents the length of postoperative hospital stay. For continuous parameters (operative time, blood loss, and LOS), Xi = 1 indicates that the individual value exceeded the mean overall value; otherwise, Xi = 0 (the mean operative time was 261.8 min; the mean blood loss was 249.7 ml; the mean LOS was 11.1 days). For binary events (conversion, reoperation, POPF, and intra-abdominal infection), Xi = 1 indicates the occurrence of event; otherwise, Xi = 0. “u” represents the percentage of cases in which the individual value was more than the mean value (u₁ = 0.460, u₂ = 0.427, u₇ = 0.300) or represents the incidence rate of events (u₃ = 0.180, u₄ = 0.013, u₅ = 0.113, u₆ = 0.140). POPF was defined in accordance with the International Study Group of Pancreatic Surgery (ISGPS) classification [18].

Surgical methods

Every case was discussed preoperatively by a multidisciplinary team to assess the indication for surgery, the type of surgery and the feasibility of minimally invasive surgery. If MIDP was deemed feasible, the differences between the robotic system and conventional laparoscopy were discussed with patients, and the choice of robotic surgery was determined based on patient preference, as robotic surgery was not yet covered by health insurance. All the operations were done by a pancreatic surgery team at our institution. Laparoscopic and robotic DP with or without splenic preservation was performed as described in previous literature [19]. In patients with benign or premalignant tumors, spleen preservation was initially attempted with the splenic vessel preservation technique (Kimura technique [20]). If the Kimura technique failed, the Warshaw technique was subsequently employed [21]. Patients with pancreatic cancer require procedures combined with a radical splenectomy. Subtotal pancreatectomy refers to the removal of the neck, body, and tail of the pancreas, and the pancreas is transected on the right of the portal vein. Robotic surgeries were completed utilizing the Da Vinci Surgical System (Intuitive Surgical, Inc., USA). After the procedure was done, a silicone tube was routinely placed beside the stump of the pancreas for drainage. The drains are passive drains and will evacuate fluid by gravity. In case of splenectomy, another drainage tube was placed in the former splenic fossa.

Statistical analysis

The continuous variables were reported as medians with interquartile range or mean ± standard deviation (SD), and analyzed using Mann–Whitney U or Student’s t-test, depending on the variable distribution. Categorical variables were described as frequencies and percentages and compared using chi-square test or Fisher’s exact test where appropriate. Factors significantly associated with conversion in the univariate analyses (P < 0.1) or gleaned from the literature were further included in multivariate logistic regression analyses to identify independent factors for model development. The receiver operating characteristic curve (ROC) analysis was applied to assess the discrimination ability of prediction models by calculating the area under the ROC curve (AUC). The AUC values of our model and the previous model were compared by Delong test [22]. Model calibration was assessed with the Hosmer–Lemeshow goodness-of-fit test and calibration plots. The decision curve analysis (DCA) was conducted to estimate the clinical utility of the nomogram. Statistical analysis was performed using SPSS 26.0 software (SPSS Inc., Chicago, IL, USA) and R software, version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria). The nomogram and calibration curves were plotted using the rms package, and decision curve analysis was performed with the rmda package in R. A significant result was indicated by a two-tailed P value < 0.05.

Results

Patients

A total of 353 patients who underwent scheduled MIDP were reviewed, and 1 case with missing conversion information was excluded. Overall, indications were pancreatic cancer in 101 (28.6%), serous cystic neoplasm in 59 (16.7%), solid pseudopapillary neoplasm in 50 (14.2%), and mucinous cystic neoplasm in 44 (12.5%). Laparoscopic distal pancreatectomy was attempted in 191 (54.3%). The conversion rate was 14.2%. The reasons for conversion to laparotomy were as follows: (1) vascular involvement (30%); (2) oncological concerns (28%); (3) bleeding (26%); (4) adhesions (12%); (5) poor visualization tumor (2%); (6) unrecorded (2%). The demographics of the training and validation cohorts were summarized and compared in Table 1. The baseline characteristics of the two cohorts were similar (P > 0.05).

Table 1.

Baseline characteristics of patients undergoing MIDP in the training and validation cohorts

Parameters	Training cohort (N=246)	Validation cohort (N=106)	P
Age (year), median (IQR)	53.0 (41.0–64.0)	54.0 (43.0–62.5)	0.580
Sex, n (%)			0.747
Male	102 (41.5)	42 (39.6)
Female	144 (58.5)	64 (60.4)
Height (m), median (IQR)	1.62 (1.58–1.68)	1.63 (1.58–1.67)	0.674
Weight (Kg), median (IQR)	60.0 (53.0–66.3)	60.0 (53.0–69.3)	0.662
BMI (kg/m2), n (%)			0.737
<18.5	19 (7.7)	9 (8.5)
18.5-24.9	166 (67.5)	67 (63.2)
≥25	61 (24.8)	30 (28.3)
Hypertension, n (%)			0.544
Yes	62 (25.2)	30 (28.3)
No	184 (74.8)	76 (71.7)
Diabetes mellitus, n (%)			0.104
Yes	42 (17.1)	26 (24.5)
No	204 (82.9)	80 (75.5)
Smoker, n (%)
Yes	46 (18.7)	19 (17.9)	0.864
No	200 (81.3)	87 (82.1)
Abdominal surgery history, n (%)			0.392
Yes	81 (32.9)	30 (28.3)
No	165 (67.1)	76 (71.7)
Preoperative labs
ALB (g/L), median (IQR)	44.2 (42.0–46.2)	44.1 (41.4–45.7)	0.287
TC (mmol/L), median (IQR)	4.51 (3.61–5.23)	4.39 (3.11–5.31)	0.439
TG (mmol/L), median (IQR)	1.21 (0.89–2.25)	1.37 (0.96–2.85)	0.199
PLT, median (IQR)	227.5 (193.0–272.0)	225.0 (179.8–265.8)	0.274
INR, median (IQR)	0.91 (0.88–0.97)	0.92 (0.88–0.96)	0.972
APTT, median (IQR)	35.2 (33.1–37.8)	35.4 (33.4–38.1)	0.610
Fibrinogen, median (IQR)	3.32 (2.88–3.82)	3.39 (2.96–3.40)	0.104
Tumor location, n (%)			0.219
Neck	34 (13.8)	8 (7.5)
Body	106 (43.1)	46 (43.4)
Tail	106 (43.1)	52 (49.1)
Imaginga, n (%)			0.471
not close to major vessel	135 (54.9)	62 (59.0)
close to major vessel	111 (45.1)	43 (41.0)
Suspicion of malignancy	92 (37.4)	40 (37.7)	0.952
Approach, n (%)			0.910
Laparoscopic	133 (54.1)	58 (54.7)
Robotic	113 (45.9)	48 (45.3)
Type of pancreatic resectiona, n (%)			0.682
Left pancreatectomy	204 (85.7)	90 (87.4)
Subtotal pancreatectomy	34 (14.3)	13 (12.6)
Conversion, n (%)			0.327
Yes	32 (13.0)	18 (17.0)
No	214 (87.0)	88 (83.0)
Spleen treatment, n (%)			0.918
Splenectomy	139 (56.5)	58 (54.7)
Kimura technique	91 (37.0)	40 (37.7)
Warshaw technique	16 (6.5)	8 (7.5)
Pathology, n (%)			–
PC	65 (26.4)	36 (34.0)
MCN	31 (12.6)	13 (12.3)
SCN	49 (19.9)	10 (9.4)
IPMN	8 (3.3)	1 (0.9)
SPN	32(13.0)	18 (17.0)
MFCP	11 (4.5)	3 (2.8)
Pseudocyst	9 (3.7)	5 (4.7)
pNETs	27 (11.0)	11 (10.4)
Others	14 (5.7)	9 (8.5)
Diameter of tumors (cm), median (IQR)	3.6 (2.5, 5.2)	3.5 (2.5, 5.0)	0.668
Operative time (min), median (IQR)	240.0 (205.0–300.0)	262.0 (223.8–320.0)	0.073
MIDP experience, n (%)			0.884
Low (≤58 cases)	41 (16.7)	17 (16.0)
High (>58 cases)	205 (83.3)	89 (84.0)
Perioperative Hb loss (g/L), median (IQR)	19.0 (12.0–27.0)	20.0 (11.0–28.0)	0.908
Blood transfusion, n (%)			0.910
Yes	20 (8.1)	9 (8.5)
No	226 (91.9)	97 (91.5)

^aMissing data: type of pancreatic resection: 8 cases in the training cohort and 3 cases in the validation cohort; imaging: 1 case in the validation cohort

Hb hemoglobin, ALB albumin, TC total cholesterol, TG triglyceride, PLT platelet, INR international normalized ratio, aPTT activated partial thromboplastin time, MIDP minimally invasive distal pancreatectomy, PC pancreatic cancer. SCN serous cystic neoplasms, MCN mucinous cystic neoplasms, IPMN intraductal papillary mucinous neoplasm, SPN solid pseudopapillary neoplasm, MFCP mass-forming chronic pancreatitis. pNETs pancreatic neuroendocrine tumors, IQR interquartile range

MIDP versus converted MIDP regarding intra- and postoperative outcomes

Patients who underwent converted MIDP had significantly longer operative time (medians, 240 vs. 336 min, P < 0.001), more perioperative hemoglobin loss (medians, 18 g/L vs. 29 g/L, P < 0.001), and a higher transfusion rate (9% vs. 20%, P < 0.001) than those with completed MIDP. Short-term outcome analyses revealed that open conversion increased the POPF rate, day of diet start, and length of stay (Table 2). The incidences of reoperation and unplanned readmission were comparable between the two groups.

Table 2.

Comparison of perioperative outcomes between MIDP and unplanned conversion

Parameters	MIDP (N = 302)	Conversion (N = 50)	P value
Operative time, min, median (IQR)	240.0 (200.0, 281.3)	336.5 (293.8, 370.8)	<0.001
Perioperative Hb loss, g/L, median (IQR)	18.0 (10.0, 25.0)	29.0 (22.8, 38.5)	<0.001
Transfusion, n (%)	9 (3.0)	20 (40.0)	<0.001
Diet start (day, mean±SD)	3 (2, 3)	4 (3, 6)	<0.001
POPF (grade B/C), n (%)	36 (11.9)	14 (28.0)	0.003
Reoperation, n (%)	7 (2.3)	0	0.277
Intra-abdominal infection, n (%)	40 (13.2)	11 (22.0)	0.103
Postoperative hospital stay, day, median (IQR)	7 (6-10)	13 (10-19)	<0.001
Unplanned readmission, n (%)	37 (12.3)	7 (14.0)	0.729

Bold values indicate P < 0.05.

ALB albumin, TC total cholesterol, TG triglyceride, PLT platelet, INR international normalized ratio, aPTT activated partial thromboplastin time, MIDP minimally invasive distal pancreatectomy, IQR interquartile range

Acquisition of risk factors for open conversion in MIDP through the literature review

Nine studies reported meaningful preoperative risk factors for converted MIDP, including eight retrospective studies and one randomized controlled trial. Details are shown in Fig. 2. A total of 11 risk factors were reported, including sex, body mass index (BMI), preoperative serum albumin, smoker, size of lesion, tumor close to major vessel, operative approach (RDP or LDP), type of resection (subtotal pancreatectomy or left pancreatectomy), surgery experience, suspicion of malignancy, and treatment at academic facility. The data of these variables were collected accordingly for analysis. The variable “treatment at academic facility” is poorly defined and was not applicable in our study.

Fig. 2.

Summary of risk factors collected from published studies. a, Casadei et al. b, Stiles ZE, et al. c, Nassour et al. d, Jiang, et al. e, Partelli et al.; f, Lof, et al. g, Liu et al. h, Hua et al. I, Korrel et al. MIDP minimally invasive distal pancreatectomy, LDP laparoscopic distal pancreatectomy RDP robotic distal pancreatectomy

Impact of MIDP learning curve on the conversion rate

MIDP experience was defined using learning curve analysis (Fig. 3).The learning curve consists of two parts: an uphill slope from the 1st to the 58th case, indicating outcomes above average, and a downhill slope after the 58th case, indicating outcomes below average. The 58th case marked the turning point, representing “high experience” in MIDP. The conversion rate was 14/58 (24.1%) during the “low experience” phase and 13/92 (14.1%) following the attainment of “high experience” (P = 0.018). The detail baseline characteristics of patients before and after the 58th case can be found in Supplementary Table 1.

Fig. 3.

CUSUM learning curve analysis to determine the number of procedures required for “high MIDP experience”. The trend of CUSUM change versus case number was depicted by the black trend line generated from polynomial regression, which consisted of an up-slope and a down-slope, with the transition point occurring at the 58th procedure

Identifying independent preoperative predictors for open conversion in MIDP

Underlying factors related to conversion were evaluated by univariate analysis in the training cohort (Table 3). Age, sex, hypertension, INR, fibrinogen, tumor location, suspicion of malignancy, relationship between lesion and major vessels, operative approach (LDP or RDP), type of pancreatic resection (subtotal pancreatectomy or left pancreatectomy), and MIDP experience were significantly associated with conversion. Multivariate analyses further identified the independent predictors as follows: male sex (OR 8.108, 95% CI 2.458–26.738, P < 0.001), tumor close to major vessel (OR 5.851, 95% CI 1.753–19.530, P = 0.004), suspicion of malignancy (OR 16.292, 95% CI 3.542–74.950, P < 0.001), subtotal pancreatectomy (OR 3.372, 95% CI 1.071–10.622, P = 0.038), high MIDP experience (> 58 cases, OR 0.150, 95% CI 0.036–0.624, P = 0.009).

Table 3.

Univariate and multivariate analyses of risk factors associated with open conversion in the training cohort

Parameters	Univariate analysis			Multivariate analysis
	MIDP n=214	Conversion n=32	P value	OR	95% CI	P value
Age, [median (IQR), year]	51 (40, 63)	63 (51, 72)	0.001	–	–	0.913
Sex, male, n (%)	78 (36.4)	24 (75.0)	<0.001	8.108	2.458–26.738	<0.001
Height, m, n (%)	1.62 (1.58, 1.68)	1.65 (1.58, 1.70)	0.302
Weight, kg, n (%)	60.0 (53.0, 65.5)	57.0 (52.0, 68.0)	0.849
BMI, kg/m2, median (IQR)	22.6 (20.6, 25.0)	22.4 (20.0, 24.8)	0.629	–	–	0.953
Hypertension, n (%)	49 (22.9)	13 (40.6)	0.031	–	–	0.959
Diabetes mellitus, n (%)	35 (16.4)	7 (21.9)	0.439
Smoker, n (%)	38 (17.8)	8 (25.0)	0.327	–	–	0.512
Abdominal surgery history, n (%)	69 (32.2)	12 (37.5)	0.555
Preoperative labs
ALB, g/L, median (IQR)	44.3 (42.3, 46.3)	43.6 (40.2, 45.7)	0.229
TC, mmol/L, median (IQR)	4.47 (3.57, 5.22)	4.69 (3.76, 5.77)	0.181
TG, mmol/L, median (IQR)	1.21 (0.85, 2.34)	1.23 (1.01, 1.71)	0.971
PLT, median (IQR)	228.5 (193.8, 274.3)	225.0 (174.8, 265.3)	0.440	–	–	0.253
INR, median (IQR)	0.91 (0.88, 0.97)	0.94 (0.90, 1.01)	0.041	–	–	0.112
APTT, sec, median (IQR)	35.2 (33.2, 37.8)	35.2 (32.0, 37.5)	0.494
Fibrinogen, g/L, median (IQR)	3.26 (2.86, 3.79)	3.68 (3.10, 4.01)	0.031	–	–	0.451
Tumor location, n (%)			0.005			0.076
Neck	30 (14.0)	4 (12.5)		–	–
Body	84 (39.3)	22 (68.8)		–	–
Tail	100 (46.7)	6 (18.8)		–	–
Diameter of tumors, cm, n (%)	3.6 (2.4, 5.1)	3.6 (2.5, 6.5)	0.264	–	–	0.181
Imaging, n (%)			<0.001			0.004
Not close to major vessels	128 (59.8)	7 (21.9)		1	(reference)
Close to major vessels	86 (40.2)	25 (78.1)		5.851	1.753–19.530
Suspicion of malignancy	62 (29.0)	30 (93.8)	<0.001	16.292	3.542–74.950	<0.001
Operative approach, n (%)			<0.001			0.482
Laparoscopic	103 (48.1)	30 (93.8)		–	–
Robotic	111 (51.9)	2 (6.3)		–	–
Type of pancreatic resection, n (%)			<0.001			0.038
Left pancreatectomy	186 (89.4)	18 (60.0)		1	(Reference)
Subtotal pancreatectomy	22 (10.6)	12 (40.0)		3.372	1.071–10.622
MIDP experience, n (%)			0.018			0.009
Low (≤58 cases)	31 (14.5)	10 (31.3)		1	(Reference)
High (>58 cases)	183 (85.5)	22 (68.8)		0.150	0.036–0.624

Bold values indicate P < 0.05

ALB albumin, TC total cholesterol, TG triglyceride, PLT platelet, INR international normalized ratio, aPTT activated partial thromboplastin time, MIDP minimally invasive distal pancreatectomy, IQR interquartile range

Establishment and validation of a nomogram to predict converted MIDP preoperatively

In the training cohort, a nomogram was constructed based on the five risk factors identified by multivariate logistic regression model, as shown in Fig. 4. Each subtype of the variables in this nomogram model was assigned a score, and summing these points yields the total score, reflecting the cumulative impact of all predictors on the outcome. By placing the total points on the designated 'total points' axis, the corresponding estimated conversion risk can be readily ascertained. The discriminatory power of our established model versus the one proposed by Casadei et al. [11] was evaluated and contrasted using ROC curves depicted in Fig. 5. In the training cohort, our model achieved an AUC of 0.921 (95% CI 0.872–0.971), performing better in predicting converted MIDP than the previous model (AUC, 0.921 vs. 0.757, P < 0.001 [Delong lest]). To mitigate the potential effects of overfitting in the training cohort, the models were further compared using an independent validation dataset. The validity of the model was corroborated with data from the validation cohort (AUC, 0.834, 95% CI 0.737 to 0.931), demonstrating its sustained superiority in discriminative power compared to the previous model (AUC, 0.834 vs. 0.716, P = 0.018 [Delong lest]). Hosmer–Lemeshow test suggested a satisfactory goodness-of-fit, with a P value of 0.115, indicating that the model fits the data well. Figure 6 A and B display the calibration curves of the MIDP conversion nomogram, illustrating a commendable agreement between predicted and observed outcomes in both cohorts. The DCA curves depict the net benefits of decision-making guided by our predictive model. Figure 6 C and D showed that our model yielded greater net benefits compared to scenarios of either no treatment or universal treatment when the threshold probability ranged from 0.03 to 0.97 in the training cohort and from 0.08 to 0.49 in the validation cohort, revealing the practical significance and clinical relevance of our model.

Fig. 4.

A nomogram for predicting conversion to open surgery in MIDP based on preoperative risk factors. MIDP, minimally invasive distal pancreatectomy

Fig. 5.

Diagnostic accuracy of the nomogram for the estimation of conversion to open surgery and comparisons between our model and an existing model using receiver operating characteristic curve (ROC) in the training and validation cohorts. A Training cohort; B Validation cohort. Model A, the nomogram proposed by Casadei et al. AUC area under the ROC curve

Fig. 6.

Calibration curve and decision curve analysis of the established nomogram in the training cohort (A, C) and validation cohort (B, D)

Discussion

In this study, we conducted a literature review to identify risk factors associated with conversion to open surgery in MIDP, thereby providing an evidentiary foundation for the development of a prediction model. A total of 10 reported risk factors, along with additional 5 clinical variables, were subjected to multivariate analysis. This process aimed to adjust for the effects of confounding variables and to derive independent risk factors crucial for constructing the prediction model. Ultimately, male sex, suspicion of malignancy, tumor close to major vessels, subtotal pancreatectomy, and low MIDP experience were found to be independently related to conversion requirements. Conversely, factors such as smoker [13], BMI [11, 13], size of tumor [7], serum albumin [13], and surgical approach of MIDP (RDP or LDP) [12, 13, 15], which had been previously reported as significant risk factors, did not exhibit independent associations in our analysis. The efficacy of the established model was validated using a distinct cohort and demonstrated improved performance compared to the existing model.

MIDP was initially utilized for the treatment of benign and premalignant pancreatic lesions. Its clinical indications have progressively expanded, and the most recent clinical trial have shown equivalent rates of radical resection and overall survival between MIDP and ODP in patients with resectable pancreatic cancer [23], which endorses MIDP as the standard procedure for both malignant and benign pancreatic tumors confined to the left-sided pancreas. The potential for worsened perioperative outcomes due to conversion to open surgery remains a concern in MIDP cases. In our cohort, patients who underwent conversion surgery experienced more adverse intraoperative occurrences compared to those who completed MIDP, characterized by longer operative time, increased blood loss and a heightened need for transfusions, which were also reported in previous studies [6, 12, 13]. Furthermore, the early postoperative outcomes were less favorable for patients requiring open procedures. Of note, a higher POPF rate was observed in the converted MIDP group than in the completed MIDP group (28% vs. 11.9%, P = 0.003). In the study conducted by Jiang et al., the rate of B/C grade POPF in the conversion group (43.3%) was higher than that in the MIDP group (27.2%) after propensity score matching [7], albeit this difference lacked statistical significance. Lof et al. [6] investigated the different impacts of elective and emergency conversions on both short-term and long-term outcomes. When compared to planned open resection, patients undergoing elective conversion had comparable overall morbidity rates, whereas those experiencing emergency conversion had significantly higher blood loss, transfusion rates, and overall morbidity rates. Given that emergency conversion in their study was defined as occurring due to uncontrollable bleeding, it is logical that these patients would experience more blood loss and have greater requirements for transfusion. The close association between high blood loss and the incidence of postoperative complications in pancreatic surgery has been reported [24, 25]. Therefore, high blood loss may be a contributing factor to the observed increase in overall morbidity in patients who required emergency conversion.

Conversion to open surgery in MIDP is a multifactorial event that can be caused by tumor-related, patient, surgeon, or operative risk factors, and the incorporation of multidimensional impact is required for a comprehensive assessment of the conversion risk. A full understanding of various indications for conversion provides insights into the identification of risk factors. The most frequently reported reasons for conversion during MIDP were vascular involvement (23.7%), concern for oncological margin (21.9%), and bleeding (18.9%) [2]. In our cohort, these respective reasons accounted for 30%, 28%, and 26% of the converted cases. Tumor invasion of other organs is seen as the top contraindication for MIDP [1]. Patients with suspicion of malignancy present malignant morphological features in computed tomography or magnetic resonance imaging, and intraoperative findings may reveal tumors extending beyond the pancreas and invading surrounding structures. In these cases, greater extension of dissection is required to achieve a complete resection; in some instances, this necessitates additional procedures such as gastrectomy or colectomy. The undertaking of these interventions amplifies the likelihood of converting to open surgery, as they compound surgical intricacy and hinder the surgeon's maneuverability within the constraints of minimally invasive methodologies. Likewise, tumor proximity to major vessels poses significant technical obstacles for radical resection during minimally invasive attempts, which directly contributes to an elevated risk of conversion. The two predictors mentioned above are both identifiable through careful preoperative imaging analysis, aiding in the risk assessment for conversion.

The extent of pancreatic resection also has an impact on the completion of MIDP. Subtotal pancreatectomy was found to increase the risk of conversion in MIDP compared to left pancreatectomy in the study of Casadei et al. [26]. The dissection procedure is challenging in subtotal pancreatectomy, where the pancreatic resection line is located on the right side of the portal vein; meticulous anatomical manipulation is required to ensure the safe management of surrounding major vessels. Therefore, tumors located in the neck or those situated close to the neck may encounter an increased risk of conversion, owing to the potential necessity for a subtotal pancreatectomy. Robot-assisted surgery was reported to be associated with a lower conversion rate than laparoscopic surgery in several retrospective studies [12, 13, 15]. In the present study, RDP was associated with a lower conversion rate than LDP in the univariate analysis, but surprisingly, this advantage was not significant in the multivariate analysis. We believe this may be because the studies that concluded RDP has an advantage over LDP did not take into account the impact of the MIDP learning curve when comparing the two surgical approaches, as Lof et al. pointed out [27]. Typically, pancreatic surgeons perform RDP procedures after obtaining initial MIDP experience in LDP procedures, as a consequence, the lower conversion rate observed in RDP compared to LDP can likely be a result of accumulated experience. Given that the MIDP experience could be obtained from either LDP or RDP, the model development ought to encompass both types of procedures—combining LDP and RDP cohorts—instead of focusing on either in isolation. The DIPLOMA randomized control trial [23] also demonstrated comparable conversion rates between LDP and RDP (LDP vs. RDP, 11.6% vs. 12.9%, P = 0.85). Therefore, the association between RDP and reduced conversion rates necessitates cautious interpretation, particularly when adjustments for MIDP experience are lacking.

There has not been a universally acknowledged benchmark defining the number of procedures necessary for mastering MIDP. A significant relationship between conversion rate and surgical experience was observed in a previous study (10.3% vs. 20.2%) [28] when using 15 procedures to define surgeons as experienced or inexperienced. A recent multicenter study conducted by European Consortium on Minimally Invasive Pancreatic Surgery (E-MIPS) investigated the number of procedures required to overcome the MIDP learning curves using various evaluation metrics. The turning points for curves assessed by conversion, operative time, blood loss, and optimal postoperative outcome (textbook outcome) were at the 40th, 56th, 71st, and 85th procedures, respectively [29]. Unlike the single-metric learning curves in the previous study, multidimensional CUSUM analysis was employed to depict the learning curve in the present study, as recommended by Mueller et al. [30]. The analysis of surgical performance was based on the combined evaluation of intraoperative and postoperative variables. The 58th procedure was identified as the turning point of surgical experience in the learning curve, with conversion rates of 24.1% and 14.1% before and after the 58th procedure, respectively. Sex was identified as a patient factor related to conversion in our analysis, which was consistent with the results in the studies of Stiles et al. [12] and Casadei et al. [11]. Male sex had also been reported as a risk factor for increased blood loss in pancreatic surgery [25]. However, the specific mechanisms remain to be investigated.

A nomogram for predicting the probability of conversion from laparoscopic to open distal pancreatectomy was proposed by Casadei et al. [11] in 2020. This predictive model contains 4 variables, including type of pancreatic resection (left or subtotal pancreatectomy), tumor location, BMI, and sex. It has an area under the curve (AUC) of 0.842, but its performance remains unvalidated. Furthermore, our findings indicate that only two variables among the four—sex and the type of pancreatic resection—emerged as independent risk factors. By analyzing the risk factors proposed by other studies, the current study aimed to generate and validate a prediction model with enhanced capacity for identifying patients at high risk of open conversion in MIDP. Five independent factors (sex, suspicion of malignancy, tumor close to major vessels, type of pancreatic resection, and MIDP experience) were finally combined to establish the final predictive model of our study. The model in the present study showed better performance in identifying MIDP conversions in both the training and validation cohorts than the previous nomogram.

The current study has several limitations. Firstly, the dataset for this retrospective study originated from a tertiary healthcare institution located in southeastern China; hence, selection bias was possible, and the results may be affected by regional factors. Notably, heterogeneity in patient populations, team familiarity and previous experience may lead to discrepancies in the number of procedures necessary to reach proficiency [30]. Constructing learning curves separately for surgeons with similar backgrounds may be one of the possible ways to address this issue, which is analogous to the “precision medicine” concept for patients. Secondly, while the model's efficacy was substantiated via internal cohort validation, external validations are required to comprehensively assess its stability. One of the strengths of this study is that the predictive model was based on risk factors that have been identified in other studies, thereby potentially mitigating the limitations inherent to single-center studies through an evidence-based modeling approach. Thirdly, although we endeavor to include all of the underlying clinical factors influencing conversion to construct a model with optimal discriminatory power, some risk factors may not be incorporated owing to the limited data that can be collected. Despite the limitations that may exist, the current study verified the significance of the previous model, and improved its moderate discrimination capacity by constructing a new model incorporating other independent predictors. We will focus on the prospective evaluation and external validation of the model to confirm its value in the future.

Conclusion

In the present study, we performed an adjusted analysis for conversion-related risk factors identified through a literature review. A total of five independent preoperative risk factors were ultimately incorporated to develop the multidimensional prediction model. This model facilitates personalized evaluation of the conversion risk for patients being considered for MIDP, which would have future utility when generating surgical strategies and counseling patients about the probability of open conversion.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No.: 82073139), the Natural Science Foundation of Fujian Province, China (No. 2020J02054), the Joint Funds for the Innovation of Science and Technology, Fujian Province (Grant number: 2021Y9058) and the Medical Minimally Invasive Center Program of Fujian Province and National Key Clinical Specialty Discipline Construction Program, China.

Declarations

Disclosures

Cong Chen, Xianchao Lin, Ronggui Lin, Yuanyuan Yang, Congfei Wang, Haizong Fang, Heguang Huang and Fengchun Lu have no conflict of interest or financial ties to disclose.