Development and validation of a preoperative nomogram for predicting the surgical difficulty of laparoscopic colectomy for right colon cancer: a retrospective analysis

Background: In laparoscopic right hemicolectomy for right colon cancer, complete mesocolic excision is a standard procedure that involves extended lymphadenectomy and blood vessel ligation. This study aimed to establish a nomogram to facilitate evaluation of the surgical difficulty of laparoscopic right hemicolectomy based on preoperative parameters. Materials and methods: The preoperative clinical and computed tomography-related parameters, operative details, and postoperative outcomes were analyzed. The difficulty of laparoscopic colectomy was defined using the scoring grade reported by Escal et al. with modifications. Multivariable logistic analysis was performed to identify parameters that increased the surgical difficulty. A preoperative nomogram to predict the surgical difficulty was established and validated. Results: A total of 418 consecutive patients with right colon cancer who underwent laparoscopic radical resection at a single tertiary medical center between January 2016 and May 2022 were retrospectively enrolled. The patients were randomly assigned to a training data set (n=300, 71.8%) and an internal validation data set (n=118, 28.2%). Meanwhile, an external validation data set with 150 consecutive eligible patients from another tertiary medical center was collected. In the training data set, 222 patients (74.0%) comprised the non-difficulty group and 78 (26.0%) comprised the difficulty group. Multivariable analysis demonstrated that adipose thickness at the ileocolic vessel drainage area, adipose area at the ileocolic vessel drainage area, adipose density at the ileocolic vessel drainage area, presence of the right colonic artery, presence of type III Henle’s trunk, intra-abdominal adipose area, plasma triglyceride concentration, and tumor diameter at least 5 cm were independent risk factors for surgical difficulty; these factors were included in the nomogram. The nomogram incorporating seven independent predictors showed a high C-index of 0.922 and considerable reliability, accuracy, and net clinical benefit. Conclusions: The study established and validated a reliable nomogram for predicting the surgical difficulty of laparoscopic colectomy for right colon cancer. The nomogram may assist surgeons in preoperatively evaluating risk and selecting appropriate patients.


Introduction
Since it was first described in 2009 [1] , complete mesocolic excision (CME) has become a standard procedure for right colon cancer and is associated with a high lymph node yield and disease-free survival [2] . Compared with the open procedure, laparoscopic right hemicolectomy reportedly has the advantage of faster recovery and comparable long-term results [3] . Thus, laparoscopic right hemicolectomy with CME is widely accepted for colon cancer. However, owing to the complexity of laparoscopic right hemicolectomy, difficult cases have a notable incidence of intraoperative complications (18.0%) and of conversion from laparoscopy to the open approach (6%) [4] . Although conversion and intraoperative complications should not be considered surgical failures, these factors influence the postoperative recovery. Thus, it is extremely important to preoperatively HIGHLIGHTS • Surgical difficulty correlated positively with perioperative mortality and negatively with long-term survival. • Prediction model of laparoscopic right hemicolectomy to evaluate the surgical difficulty. • Computed tomography parameters were introduced to establish the prediction model. • The nomogram incorporating seven independent predictors showed a high C-index.
evaluate the surgical difficulty and select appropriate patients in accordance with the surgeons' experience.
Previous studies have mainly focused on the establishment and validation of nomograms for predicting surgical difficulty in laparoscopic splenectomy and laparoscopic surgery for rectal cancer and sigmoid cancer [5][6][7][8] . However, the systematic evaluation of the surgical difficulty of laparoscopic right hemicolectomy is very rare.
Preoperative computed tomography (CT) scanning is recommended to determine the optimal therapeutic strategy for patients with colorectal cancer. CT is applied to detect distant metastasis and is also used as a local staging tool that can measure the tumor size and invasion depth with acceptable accuracies and sensitivities [9] . Moreover, the arteries, veins, and blood supplies of the tumor and right colon are visible under CT angiography. As right hemicolectomy with CME involves the identification and ligation of several blood vessels, such as the ileocolic artery, ileocolic vein, right colonic artery, right colonic vein, and Henle's trunk [10] , the distribution and variations of the blood vessels seen on CT angiography may influence the surgical difficulty.
The aim of the present study was to define preoperative predictors (including CT parameters) that affect the surgical difficulty and to establish a preoperative nomogram for predicting the surgical difficulty of minimally invasive right hemicolectomy, thus facilitating the appropriate selection of patients.

Patients
Data of consecutive patients with right colon cancer who were scheduled for elective radical resection in two tertiary medical centers between January 2016 and May 2022 were retrospectively retrieved from our prospectively collected database and screened for inclusion. The inclusion criteria were: age 18-80 years, with no sex limitation; complete clinical data and follow-up. The exclusion criteria were: preoperative CT image data were incomplete and could not be followed up; T4b cancer or bulky tumors; open surgery; synchronous or double primary cancer. A final total of 568 patients were included. This work has been reported in line with the STROCSS (Strengthening the Reporting of Cohort Studies in Surgery) criteria [11] , Supplemental Digital Content 1, http://links.lww.com/JS9/A198.

Data collection
For each patient, the following data were extracted from the database: (1) Patient characteristics, namely sex, age, BMI, American Society of Anesthesiologists (ASA) score, smoking status, alcohol consumption status, and comorbidities. (2) Preoperative laboratory results (plasma triglyceride and serum glucose concentrations). (3) Radiological data. Preoperative contrast-enhanced CT scans were conducted using a CT scanner (GE Lightspeed VSX, 64slice), and the digital data were collected and analyzed by two dependent experienced radiologists who were blinded to the clinical data. The radiologists measured and defined the image parameters, and reanalysis was performed when the differences in results between the two radiologists exceeded 5%. The CT parameters were the adipose thickness at the drainage area of the ileocolic vessels, adipose area at the drainage area of the ileocolic vessels, adipose density at the drainage area of the ileocolic vessels, type of Henle's trunk, presence of the right colonic artery, abdominal wall adipose area, and intraabdominal adipose area. There are six types of Henle's trunk, each of which may affect the difficulty of surgery [12] . Type I: right gastroepiploic vein (RGEV) + anterosuperior pancreaticoduodenal vein (ASPDV) + superior right colic vein (SRCV); type II: RGEV + ASPDV; type III: RGEV + ASPDV + right colic vein (RCV) + superior right colic vein (SRCV); type IV: RGEV + ASPDV + RCV; type V: RGEV + SRCV; type VI: others. (4) Intraoperative data, namely operative time, intraoperative blood loss, and need for conversion. (5) Postoperative data, namely length of hospitalization, morbidities graded using the Clavien-Dindo classification, and perioperative mortality. The 3-year survival rate was analyzed by follow-up.

Surgical procedure
All procedures were performed by surgeons with experience in more than 100 cases of laparoscopic right hemicolectomy. The whole procedure was performed as previously described [13] . Briefly, after general anesthesia, the patient was placed in the supine position with their legs spread. The operator stood on the patient's left side, the first assistant stood on the patient's right side, and the camera assistant stood between the patient's legs. Pneumoperitoneum was created with CO 2 maintained at 12 mmHg, and a 10-mm trocar for laparoscopy was inserted 3 cm below the umbilicus. After meticulous abdominal exploration, four trocars were placed at the left and right sides of the abdomen. The medial approach was commonly applied. The last ileal loop was identified, and the dissection was started behind the pedicle of the ileocolic vessels. Once the correct plane was identified, we switched to dissecting along the superior mesenteric vein (SMV). The ileocolic vessels were clipped at the root and transected with lymph node dissection. The surgical trunk of the SMV was exposed to the level of the inferior edge of the pancreas while being extended in the cephalad direction up to the origin of the right colic artery, if present, and toward the middle colic artery. The Henle's trunk was dissected and the accessory right colic vein was exposed. The accessory right colic vein and right branches of the middle colic vessels were clipped and divided. After finishing the ligation of the vessels and dissection of the lymph nodes, the proximal transverse and hepatic flexure of the right colon was easily freed and the lateral attachments were freed up, thus completing the right colon mobilization. The anastomosis was accomplished extracorporeally with an end-to-side anastomosis.

Surgical difficulty criteria and grade
The surgical difficulty criteria were defined as previously described by Escal et al. [14] with modifications. Briefly, five criteria were chosen to evaluate the surgical difficulty and each criterion was assigned a score weighted with coefficients. The total score was 0-10 and the surgical difficulty grade was classified into two levels; a score of less than 3 was taken to indicate a non-difficult procedure, while a score of at least 3 indicated a difficult procedure (Table 1).

Statistical analysis
The SPSS 25.0 (IBM SPSS Statistics) and R software programs were used for the statistical analysis. The Shapiro-Wilk test and Q-Q plot were used to determine the normality of continuous variables. The t test was used to compare continuous variables, while the χ 2 test was used to compare categorical variables. Univariable and multivariable logistic analyses were performed to identify preoperative variables that may affect the surgical difficulty and the perioperative mortality. Kaplan-Meier models and log-rank tests were used for the univariable analysis of survival. A machine learning algorithm was applied to determine the best prediction model. A predictive nomogram was developed and validated internally and externally. The Spearman correlation coefficient was used to evaluate the relationship between the surgical difficulty grading score and perioperative outcomes. Receiver operating characteristic curves were used to evaluate the discriminatory power of this scoring system. Statistical significance was defined as P < 0.05.

Patient characteristics
A total of 568 patients were included in the final analysis. Among the 418 patients from the first medical center, there were 300 and 118 patients assigned to the training data set and internal validation data set, respectively. As shown in Table 2, in the training data set, the median age was 62 years and the median BMI was 24.5 kg/m². The comorbidities were diabetes (43 patients), chronic heart disease (29 patients), and hypertension (101 patients). There were 93 patients with a history of smoking and 62 with a history of alcohol consumption. Thirty-six patients had a high plasma triglyceride concentration (88.0%) and 43 had a high serum glucose concentration (85.7%).
An overview of the radiological results is shown in Table 3. The median adipose thickness, area, and density values in the drainage area of the ileocolic vessels were 4.06 mm, 1.73 mm 2 , and − 60.30 HU, respectively. The right colonic artery was present in 113 patients. One hundred ten patients (36.7%) had a tumor with a diameter at least 5 cm.
The relevant baseline data and an overview of the radiological results in the internal validation data set and external validation data set are presented in Tables 2 and 3. There were no significant differences in baseline data between the training and internal validation data sets (P > 0.05).
In the internal validation data set, 82 (69.5%) patients were classified as the non-difficulty group and 36 (30.5%) were classified as the difficulty group. Compared with the non-difficulty group, the difficulty group had a significantly longer operative  time (P < 0.01), and longer duration of hospitalization (P < 0.01) ( Table 4).
In the external validation data set, 104 patients and 46 patients were assigned into non-difficulty and difficulty group, respectively. As shown in Table 4, a significantly longer operative time, higher conversion rate, and longer duration of postoperative hospitalization were observed in the difficulty group.

Results of the univariable and multivariable logistic regression analyses
All continuous data sets in the analysis were normally distributed or approximately normally distributed (Tables 2 and 3). Univariable and multivariable analyses showed that the independent risk factors for difficult laparoscopic right hemicolectomy were the adipose thickness at the drainage area of the ileocolic vessels [odds ratio (  5 and 6).
Use of a machine learning algorithm for the selection of the optimal prediction model Four prediction models [logistic regression, support vector machine (SVM), boosting, and random forest] were established by machine learning. The SVM model had the lowest root mean square error (RMSE) (0.675), while the boosting and random forest models had RMSEs of 0.737 and 0.762, respectively; however, the logistic regression model was found to be unsuitable for RMSE evaluation. The area under the receiver operating characteristic curve (AUC) values were 0.823 for the logistic regression model, 0.841 for the SVM model, 0.827 for the boosting model, and 0.832 for the random forest model; the other predictors (accuracy, precision, and recall) are shown in Table 7. There were no significant differences between models in the results of these evaluation methods, which implies that the four models all represent good predictors. Because prediction models can be explained more easily using a nomogram, we chose to establish a logistic regression model in the present study.

Development and validation of a predictive nomogram for surgical difficulty
Based on the independent risk factors, a predictive model was established to assess the surgical difficulty of laparoscopic right hemicolectomy (Fig. 1). The calibration curves of the training data set (Fig. 2) and validation data sets (Fig. 3) were highly consistent with the fitting line, which reflects the high accuracy of the prediction model. The C-index of the prediction model was 0.922 (95% CI: 0.882-0.961) for the training data set, 0.992 (95% CI 0.981-1.000) for the internal validation data set, and  The C-indexes were all greater than 0.7, indicating that the prediction ability of the model had high credibility (Fig. 4).

Discussion
D3 surgery is widely recognized as the standard procedure in laparoscopic radical right hemicolectomy for right colon cancer.
Despite multiple extensive discussions about the surgical access for this procedure involving completely medial access by the page-turning approach [13] and the cephalic approach [15] , these surgical approaches inevitably require total exposure of the SMV. The relatively thin venous wall makes the exposure of the SMV and the management of its branches the most complex part of the D3 procedure. Inexperienced operators are very prone to cause bleeding during the search for the left wall of the SMV or ligation of the vessels, leading to failure of the lumpectomy and conversion to intermediate open surgery. Therefore, the preoperative assessment of the difficulty of SMV exposure and vascular staging are important to appropriately guide the operation. In addition, the size of the tumor and the degree of obesity of the patient alter the complexity of the operation [16][17][18] . Thus, a comprehensive assessment is valuable as a priori knowledge for precise surgery and selection of appropriate patients for laparoscopic surgery. Considering few studies have assessed the surgical difficulty of the right hemicolectomy, we established a surgical difficulty score system based on the study of Escal et al. [14] . In their study, they evaluated the surgical difficulty of rectal cancer based on seven criteria, namely duration of surgery, conversion to open surgery, blood loss, length of hospital stay, morbidity, margin status of the surgical specimen, and the addition of transanal total mesorectal excision to transabdominal surgery. We adopted five of these criteria that suited the surgical characteristics of right hemicolectomy and validated the criteria in our department. Consistent with previous research [19] , we found that increased surgical difficulty was positively correlated with perioperative mortality and negatively correlated with long-term survival, confirming that the criteria selected for the evaluation of surgical difficulty in our study were relatively objective and strongly correlated with the perioperative and long-term survival outcomes.
In the present study, retrospective analyses of 300 patients' data using a multifactorial approach revealed that the factors influencing the occurrence of surgical difficulty were the adipose thickness, area, and density at the drainage area of the ileocolic vessels, presence of the right colonic artery, presence of type III Henle's trunk, abdominal adipose area, plasma triglyceride concentration, and tumor diameter at least 5 cm. The BMI was not an influencing factor, which is in contrast to a previous study that reported that BMI influences the difficulty of surgery for colon cancer [20] . Considering the inconsistency of the relationship between the BMI and abdominal visceral fat [21] , the concepts of the adipose CT value, adipose density, adipose volume, and adipose thickness were introduced to more accurately reflect the adipose volume and adipose density in the free region of the right hemicoelomic vessels [22,23] . We found no effect of the adipose density of the abdominal wall and adipose volume of the    [24,25] . Of these, the difficulty of surgery was significantly increased in patients with type III Henle's trunk owing to the extended time required to identify and protect its branches, as the presence of more branches increases the risk of iatrogenic injury and bleeding.
Right colonic angiogenesis reportedly occurs in~12-45% of patients [26] and was found to be present in 36% of patients in our case series. The right colonic vessel is frequently overlooked due to its relatively low prevalence. Careless dissection may result in unexpected bleeding from the right colonic vessel, making the surgical procedure more difficult. In addition, the blood lipid concentration and blood glucose concentration may influence the adipose density [27] . The blood lipid concentration was revealed as a risk factor for surgical difficulty in the univariable analysis of the present study, while both the blood lipid and glucose concentrations had no effect on the surgical difficulty in the multivariable analysis.
Based on the multivariable analysis results, the optimal model (logistic regression model) was selected by the machine learning algorithm, and the prediction model of surgical difficulty was established. The calibration curve was highly consistent with the fitting line, reflecting the high accuracy of the prediction model. The C-indexes of the prediction model were 0.922 for the training data set, 0.992 for the internal validation data set, and 0.994 for the external validation data set, indicating the high credibility of the prediction capability of the model.
The present study has certain limitations. First, this is a retrospective study with small sample sizes, and larger sample sizes from more centers are warranted for further validation. Second, few studies have evaluated predictive models of the difficulty of colon surgery based on CT values, and many young surgeons lack experience in assessing CT values, which may limit the wide application of our study. Despite these limitations, to our knowledge, the present study is the first to construct a nomogram based on CT values to assess the difficulty of right hemicolectomy, which may contribute to the stratification of patients before D3 radical surgery for colon cancer. Internal validation, external validation, and bootstrap validation in the cohort indicated the good predictive power of the model, which is conducive to making surgical decisions and reducing the incidence of postoperative complications after D3 surgery.

Ethics statement
The study was approved by the Ethics Committee of Qilu Hospital of Shandong University and conducted in compliance with the Helsinki Declaration (KYLL-202111-087-1). It was registered with the China Clinical Trial Registration Center (No. ChiCTR2200060774). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

Data availability statement
The original contributions presented in the study are included in the article. Further inquiries can be directed to the corresponding author.

Provenance and peer review
Not commissioned, externally peer-reviewed.