Nomographic predictive models for complications after minimally invasive esophagectomy: Current status and future perspectives

Abstract

Perioperative morbidity of esophagectomy significantly affects the surgical outcome, like any major gastrointestinal procedure. Despite introduction of minimally invasive esophagectomy, the morbidity is still close to 30%-40%. The common complications following esophagectomy are pulmonary infections, cardiac events, anastomotic leakage, bleeding, chylous leak, and recurrent laryngeal nerve palsy which in turn lead to longer hospital stay, increased treatment cost and poor quality of life. A nomographic model comprising preoperative (patient, disease and treatment related) and intraoperative factors in combination with Artificial Intelligence may accurately identify the patients at higher risk of morbidity. This will aid in optimizing the modifiable risk factors preoperatively, and closely monitor these patients post operatively for early identification of complications and to initiate early corrective measures to improve the surgical outcome.

Key Words: Minimally invasive esophagectomy; Complications; Risk factors; Nomogram; Post operative outcome; Early detection of complications; Prognostication

Core Tip: Minimally invasive esophagectomy is associated with reduced pulmonary morbidity, less blood loss, faster recovery and equivalent oncological outcomes compared to open esophagectomy, however the overall procedure related morbidity continues to be high. This review highlights the key procedure related morbidity and associated risk factors for the same in patients undergoing minimally invasive esophagectomy. Various risk prediction nomograms have been proposed for anticipating and reducing these complications to improve treatment outcomes, but they have certain limitations which hinder their generalized utility. The growing use of artificial intelligence and machine learning promises to create more sophisticated models that enhance the risk predictive accuracy and help personalize treatment plans to minimize complications and achieve better treatment outcomes.

INTRODUCTION

Esophagectomy is one of the cornerstones for the management of malignant and some benign disorders of the esophagus. In the current era, a subtotal or total esophagectomy is largely performed for malignant disease and surgical resection in combination with chemotherapy or chemo-radiation still remains the standard of care for esophageal and gastroesophageal junction cancer[1]. Esophagectomy is also being performed in few common premalignant and benign indications like long segment Barrett’s esophagus, end stage achalasia cardia with sigmoid esophagus and long segment or indeterminate stricture[2]. The procedure involves esophagectomy with lymph-node dissection and esophageal reconstruction which makes it technically challenging. It is associated with significant procedure related morbidity and mortality during dissection in two (thorax and abdomen) or three (thorax, abdomen and neck) different anatomical fields[1-3]. The reported postoperative morbidity and mortality rates for traditional open esophagectomy are 30% to 50% and 1% to 6% respectively[4]. Minimally invasive approaches to perform this procedure have been introduced with the aim to reduce surgical stress and trauma which in turn reduces the postoperative morbidity. These techniques include hybrid or total minimally invasive esophagectomy (MIE) using thoracoscopy and laparoscopy or robotic platforms. MIE has been associated with fewer pulmonary complications, less blood loss and better pain scores than open esophagectomy and has gained popularity over the last two decades. In spite of these benefits, the overall post esophagectomy morbidity and mortality rates still remain high which in turn impacts the long-term quality of life of these patients[1,5-7]. Multiple factors (patient, disease and treatment related) in combination play a pivotal role in deciding the post procedure recovery and overall outcomes of esophagectomy. Prior knowledge of these factors may help identify individuals who are at an increased risk of developing complications, may allow optimization of some correctable risk factors and help select alternative treatment options in patients at very high risk of morbidity[8]. Nomograms are simple tools developed using clinical data to predict complications in clinical practice. They graphically compare known risk factors and make individualized risk prediction concise, convenient and intuitive. In surgical practice, predictive nomograms based on risk factors are commonly used to anticipate and minimize procedure related complications in susceptible individuals by taking pre-emptive corrective measures and prognosticate patients regarding the procedural outcomes[4,9]. Despite improvement in technology and surgical refinements, esophagectomy whether open or minimally invasive is not free from complications. The existing risk predictive nomographic models for MIE have several deficiencies and their clinical utility is hindered by methodological heterogeneity, poor reproducibility, lack of external validation and their focus on a specific complication. Rapid ongoing advancements in artificial intelligence (AI) and machine learning (ML) presents a significant opportunity for development of robust nomograms to predict post esophagectomy complications with great accuracy and precision. This review explores the clinical utility of currently available risk prediction nomograms for major morbidity following esophagectomy and the future potential in predicting these complications by incorporating new tools and technology.

EVOLUTION OF MIE

Esophagectomy is a procedure requiring access to the mediastinum and the abdomen with or without neck dissection depending on the resection and level of the anastomosis[1]. For this purpose, various types of minimally invasive approaches have been reported in literature. Hybrid MIE (HMIE) using thoracoscopy and laparotomy was first performed and reported by Cuschieri et al[10] in 1992 following which laparoscopic transhiatal esophagectomy was reported in 1995 by DePaula et al[11]. In 2003, Luketich reported his large series on the technique and outcomes of thoraco-laparoscopic esophagectomy[12]. In the same year, after adoption of the robotic platform, robot assisted MIE (RAMIE) was first reported in by Horgan et al[13] (transhiatal) and Kernstine et al[14] (2 stage). The current terminologies used for the various MIE techniques are not standardized and used interchangeably in literature. These terminologies should be well defined for making accurate comparison in future research. These techniques broadly include “total MIE” where the chest and abdominal phases are performed by thoracoscopy and laparoscopy respectively or using the robotic platform and HMIE where either of the two phases is performed by open technique and the other component is performed using laparoscopy, thoracoscopy or robotic approach. A transhiatal esophagectomy performed using laparoscopy for the abdominal phase is also considered as a type of total MIE. A number of studies have reported the benefits of total and HMIE over conventional open esophagectomy in terms of lower incidence of pulmonary morbidity, less blood loss, shorter hospital-stay, lower in-hospital mortality rate and better post operative quality of life with comparable oncological outcomes in terms of lymph-node yield, recurrence free and overall survival[5,15,16]. The thoracic phase of MIE can be performed in semi-prone or complete prone position and the abdominal phase is performed in supine position. Traditionally, MIE was described in semi-prone position or left lateral decubitus which demanded single lung ventilation and total lung collapse on the ipsilateral side. MIE in prone position was initially described by Cuschieri et al[10] and popularized by Palanivelu et al[17] in a large series where they highlighted better ergonomics, shorter operative time and a lower pulmonary morbidity rate associated with their technique. However, conversion to open thoracotomy in emergent instances is difficult and cumbersome which still makes the semi-prone position popular among foregut surgeons who are proponents of the same[17,18]. The reported difference in outcomes of MIE and conventional open esophagectomy have been compared in Table 1.

Table 1 Difference in post operative outcomes between minimally invasive and open esophagectomy.

Ref.	Sample size	Blood loss (mL)	Operative time (minutes)	Length of stay (days)	Respiratory morbidity	Anastomotic leak rate	In-hospital mortality	Lymph node retrieval	3-year overall survival
Smithers et al[61]	446	Lower in MIE (300 vs 600), P = 0.017	Shorter in OE (330 vs 300), P = 0.01	Shorter in MIE (11 vs 14), P = 0.03	NSD	NSD	NSD	NSD	NSD
Nagpal et al[62]	1284	Lower in MIE, P < 0.001	Shorter in OE, P < 0.001	Shorter in MIE, P = 0.004	Lower in MIE, P = 0.04	Lower in MIE, P = 0.02	NSD	NSD	NR
Yibulayin et al[6]	15790	Lower in MIE, P = 0.05	Shorter in OE, P < 0.05	Shorter in MIE, P < 0.05	Lower in MIE, P < 0.05	NSD	Lower in MIE, P < 0.05	NS	NR
Sihag et al[63]	3780	NR	Shorter in OE (443 vs 312), P < 0.001	Shorter in MIE (9 vs 10), P < 0.001	NSD	NSD	NS	NR	NR
Xiong et al[64]	488	Lower in MIE, P = 0.001	Shorter in MIE, P < 0.001	Shorter in MIE, P < 0.001	Lower in MIE, P < 0.001	NSD	NSD	NSD	NR
Yoshida et al[65]	24233	Lower in MIE, P < 0.001	Shorter in MIE, P < 0.001	NR	Lower in MIE (16% vs 18%), P = 0.001	NSD	Lower in MIE (1.7 vs 2.4), P < 0.001	NR	NR
Akhtar et al[66]	13269	Lower in MIE, P < 0.001	Shorter in MIE, P < 0.001	Shorter in MIE, P < 0.001	Lower in MIE, P < 0.001	NR	NR	NR	NR
Ofuchi et al[67]	459	NR	NR	NR	Lower with MIE (28% vs 40%)	NR	NSD	NSD	NSD
Randomized control trials
Biere et al[5]	115	Lower in MIE (200 vs 475), P < 0.001	Shorter in OE (329 vs 299), P = 0.002	Shorter in MIE (11 vs 14), P = 0.044	Lower in MIE (12% vs 34%), P = 0.005	NSD	NSD	NSD	NR
Mariette et al[15]	207	NR	NSD	NSD (14 days vs 14 days)	Lower with MIE (18% vs 30%)	More in MIE (11% vs 7%)	NSD	NSD	NSD
van der Sluis et al[16]	112	Lower with MIE (120 mL vs 200 mL)	Longer in RAMIE (60-70 minutes more)	NSD (14 days vs 16 days)	Lower with MIE (32% vs 58%)	NSD (22% vs 20%)	NSD	NSD (27 vs 25)	NSD
ROMIO study group[68]	533	NR	NR	NSD (10 days vs 11 days)	NSD	NSD (8% vs 8%)	NSD	NR	NR

MIE: Minimally invasive esophagectomy; OE: Open esophagectomy; NSD: No significant difference; NS: No significant; NR: Not reported; RAMIE: Robot assisted minimally invasive esophagectomy.

MIE RELATED MORBIDITY

Although the short-term morbidity and mortality associated with esophagectomy have remarkably reduced after the introduction and adoption of MIE and refinement in surgical technique, the overall procedure associated morbidity rate is still close to 30%-40%, which is relatively high compared to most other gastrointestinal surgical procedures[5,15,16]. Thoracoscopic esophagectomy using laparoscopy or robotics has a long learning curve and mandates high skill level due to its vicinity to vital mediastinal structures and moving target anatomy with respiration[19,20]. The various types of MIE depending on the varying level of complexity have different learning curves and gaining proficiency requires anywhere between 50 procedures to 119 procedures. The learning curve for RAMIE is 20-80 cases which is slightly less than that of video-assisted thoracoscopic esophagectomy. This learning curve is associated with a learning curve morbidity which is typically higher than that seen at the hands of experienced and proficient surgeons[21-23]. MIE being a complex surgery, some learning curve morbidity is inevitable and shortening the learning curve by the safe implementation of a standardized MIE training program for the fellows and trainees may reduce the morbidity rate and improve patient outcomes. The overall rate of individual complications, operative time and hospital stay have significantly reduced after the widespread adoption of MIE. Procedural morbidity of MIE can broadly be divided into two categories: Intraoperative and postoperative complications as mentioned in Table 2.

Table 2 List of complications post minimally invasive esophagectomy.

Timing	Category	Complications
Intraoperative		Bleeding
		Anesthesia related complications
		Adjacent visceral injury
Post-operative	Medical	Pulmonary morbidity
		Atelectasis
		Aspiration
		Bronchopneumonia
		Respiratory failure
		ARDS cardiac morbidity
		Cardiac arrest
		MI
		Angina
		Thromboembolism
		Arrhythmias
	Major surgical immediate	Anastomotic leak
		Conduit necrosis
		Chyle leak
		Recurrent laryngeal nerve palsy
	Surgical long term	Anastomotic stricture
		Delayed gastric emptying
		Gastro-esophageal reflux
		Malabsorption and nutritional deficiencies

MI: Myocardial infarction; ARDS: Acute respiratory distress syndrome.

INTRAOPERATIVE COMPLICATIONS

The reported rate of intraoperative complications during MIE ranges between 4%-10% in literature depending on the centers experience and around 60% of these are related to bleeding. Intraoperative complications may occur during the laparoscopic phase or the thoracoscopic phase while performing MIE and timely recognition and rectification of these complications on table may help mitigate the post-operative morbidity[24-26].

Laparoscopic phase

During the laparoscopic phase, the commonly reported complications include injury to named major vessels like gastroepiploic vessels, left gastric and splenic vessels during conduit preparation or performing lymphadenectomy which may be controlled surgically using sutures or energy sources and in extreme situations may demand additional resection or use of an alternate conduit. Inadvertent injury to the adjacent viscera like spleen, liver, colon, small bowel and pancreas has also been reported in literature which can be repaired surgically or may warrant resection of the injured organ.

Thoracoscopic phase

Common vessels injured during this phase include the aorta, azygous vein and bronchial artery during esophageal mobilization followed by inferior pulmonary vein and superior vena cava which are relatively uncommon. Adjacent viscera like the lung parenchyma and tracheo-bronchial tree, thoracic duct and recurrent laryngeal nerve (RLN) have also been reported. Tracheo-bronchial and lung parenchymal injuries may be life threatening and warrant prompt management whereas thoracic duct and RLN injuries if missed intraoperatively may present in postoperative period with chyle leak and vocal cord palsy respectively[27].

Post-operative complications

These are further broadly classified into medical and surgical subcategories as shown in Table 2.

Medical complications

This category mainly includes respiratory morbidity (aspiration pneumonia, broncho pneumonia, respiratory failure, acute respiratory distress syndrome), cardiovascular morbidity (atrial fibrillation, arrythmias, acute myocardial infarction, angina), thromboembolic episodes, liver or kidney dysfunction/failure and septic complications. Post operative pulmonary and cardiac complications constitute the major proportion of the medical morbidity related to esophagectomy[28].

Pulmonary morbidity

Post esophagectomy respiratory complications include atelectasis, pneumonia, aspiration pneumonitis, ARDS and pulmonary thromboembolism. Being one of the most frequent complications encountered after radical esophagectomy, its reported incidence ranges from 17% to 50%. Elderly age, diabetes mellitus, pre-existing pulmonary disease, history of smoking or alcohol use, tumor location, gross residual tumor and surgical tissue trauma are few common risk factors that can lead to the increased incidence of pulmonary complications. Post-operative risk factors that contribute to pulmonary complications include vocal cord paralysis, RLN palsy, swallowing problems, poor pulmonary hygiene, poor pain control, and post-operative respiratory muscle dysfunction. Careful selection of patients, pulmonary pre-habilitation (physiotherapy, incentive spirometry) and intra-operative precautions (bronchial artery preservation, handling lung with care) help to curtail the risk for pulmonary complications after radical esophagectom[29-32].

Cardiac morbidity

Cardiac morbidity broadly includes intra and post-operative events ranging from arrythmias, angina, myocardial infarction, thromboembolism and cardiac arrest. Such events are a usual cause of post procedural mortality after non-cardiac surgeries. Improved operative techniques and intensive monitoring of the patient’s cardiovascular status can help to minimize the incidence of such adverse events. Among the various non cardiac surgeries, esophagectomy is one of the procedures with greater odds of cardiac complications compared to other gastrointestinal surgeries. Prolonged hypotension, as a result of rapid intercompartmental fluid shift seen intraoperatively has also been reported in some cases. MIE has been reported to be superior to open esophagectomy with regards to adverse cardiovascular events. Preoperative cardiovascular function evaluation and optimization by using calcium channel blockers, angiotensin converting enzyme inhibitors and angiotensin receptor blocking agents can reduce the risk of atrial fibrillation and arrythmias which in turn decrease the overall survival rate and subsequent post procedure mortality[33-35].

Major surgical complications

The most frequent major surgical complications in the immediate post-operative period include anastomotic leak, conduit necrosis, chyle leak and RLN injury. Long-term morbidity includes anastomotic stricture, delayed gastric emptying, gastro-esophageal reflux, malabsorption and nutritional deficiencies[28]. The reported incidence of commonly encountered major surgical complications post MIE in current literature has been summarized in Table 3.

Table 3 Published series reporting the incidence of major complications associated with minimally invasive esophagectomy.

Ref.	Technique	n (cases)	Pulmonary complication (%)	Cardiac arrhythmia (%)	Anastomotic leak (%)	Conduit necrosis (%)	Chylothorax (%)	RLNP (%)
Xie et al[69]	IL	106	4.7	2.8	4.7	3.8	3.8	-
Sihag et al[63]	IL	600	12.8	-	13.8	-	-	4.3
van Workum et al[70]	IL	561	27.8	17.1	14.4	-	8.7	0.5
Naffouje et al[71]	IL	161	11.8	-	13	-	-	-
Lorimer et al[72]	IL	200	17	23	8.5	-	-	-
van der Sluis et al[73]	RM	108	33.3	8.3	18.5	-	17.6	9.3
Park et al[23]	RM	140	8.8	-	9.3	0	-	25
van der Sluis et al[16]	RM	54	27.8	22.2	24.1	-	31.5	9.3
Zhang et al[74]	RIL	76	6.6	-	9.2	1.3	1.9	6.6
Meredith et al[75]	RIL	147	6.8	11.6	2.7	-	-	3.4
Wang et al[76]	TME	194	6.2	3.6	4.6	-	-	4.6
Fujiwara et al[77]	TME	60	6.7	-	15	-	0	33.3
Kinjo et al[78]	M	106	16	9.4	10.4	0.9	-	19.8
Chen et al[79]	M	142	9.2	2.8	6.3	0.7	2.8	5.6
Li et al[80]	M	89	7.9	7.9	21.3	1.1	-	20.2
Tanaka et al[81]	M	59	13.6	11.9	6.8	0	-	22
Seesing et al[18]	M	121	36.4	19.8	26.4	33.9	-	7.4
Koterazawa et al[82]	M	162	14.8	-	17.9	-	-	19.8
van Workum et al[70]	M	226	24.8	26.1	26.5	-	13.3	9

RLNP: Recurrent laryngeal nerve palsy; IL: Ivor Lewis; RM: Robotic-assisted McKeown; RIL: Robotic-assisted Ivor Lewis; TME: Transmediastinal esophagectomy; M: McKeown.

Anastomotic leak

Anastomotic leak is the one of the most dreaded complications encountered post esophagectomy. Though less than the respiratory morbidity, the reported incidence of the same post esophagectomy ranges between 8% to 15% across all series. Anastomotic leaks can be minor or major (based on output and clinical status), radiological (without clinical symptoms) or clinically apparent with increased discharge containing saliva in neck drain or neck wound infection. Major leaks usually manifest within the first five days of the procedure with symptoms of severe sepsis. Patients usually have tachycardia, arrythmias, flushed face, effusion or hydropneumothorax with respiratory distress and may require vasopressor support and mechanical ventilation. Minor leaks are usually apparent after 5-6 days and present with neck wound infection, discharge and pleural effusion. Careful clinical suspicion with timely diagnosis and intervention is important for managing leaks successfully. Minor leaks in patients with good performance status can be managed conservatively. Major leaks usually warrant immediate re-exploration with cervical esophagectomy and gastrostomy after dismantling the anastomosis completely followed by colonic pull up or jejunal interposition during the second stage[36-38].

Conduit necrosis

Conduit necrosis is clinically characterized by acute alteration in hemodynamic parameters, new onset respiratory distress along with clinical sepsis and the patient may progress to develop septic shock if not managed properly. The diagnosis is made by bedside exploration of the neck wound, cross sectional imaging or with the help of endoscopy for an intrathoracic anastomosis. Once conduit necrosis is confirmed, conduit take down must be performed immediately along with cervical esophagostomy and feeding jejunostomy. Reconstruction is performed as a second stage by means of a colonic or jejunal interposition[39-42].

Chyle leak

Post esophagectomy chyle leak can be minor or major. The incidence of post esophagectomy chyle leak ranges from 2% to 10% in literature. The diagnostic criteria are variable and most often milky appearance of the drain fluid that increases on initiation of enteral feed is the initial suggesting event. Drain fluid triglyceride levels (> 110 mg/dL), drain fluid specific gravity, pH, differential leucocyte count and presence of chylomicrons are the usual confirmatory test used for the diagnosis. A bedside methylene blue test or ether solubility test may also aid in the confirmation. Most of the leaks are managed conservatively by using fat free or medium chain fatty acid diet, electrolyte and fluid replacement and occasional use of total parenteral nutrition. Major persistent leaks (> 1 L/day for 5 days or persistent leak > 2 weeks) usually warrant radiological intervention for embolization or surgical ligation of the thoracic duct at the level of the diaphragmatic hiatus[43,44].

RLN injury

Surgical trauma like stretch, compression or thermal damage to laryngeal nerve can lead to paresis or palsy in up to 5% to 40% cases. This variation in the incidence of RLN palsy depends on the tumor size, location and the surgical technique used for resection. Left RLN injury is more frequent than right and the reported incidence of injury is more in patients undergoing upper mediastinal lymphadenectomy and McKeown’s esophagectomy compared to Ivor Lewis esophagectomy. RLN injury can be temporary or permanent. Majority of the injuries are temporary that usually recover spontaneously within 2-3 months with compensation of opposite vocal cord. Permanent RLN injury may require medialization of cord by Teflon injection or thyroplasty[45,46].

RISK FACTORS FOR COMPLICATIONS

Various factors have been implicated and reported to affect the complications with MIE which include patient characteristics, disease and treatment related factors, surgical techniques, and the experience. Individual patient characteristics like age, gender, body habitus, performance status, nutrition status, genetic predispositions, addictions and pre-existing co-morbidities like diabetes mellitus, respiratory, cardiac or other systemic conditions play a pivotal role in determining operative outcomes and predispose patients to post operative complications. Early recognition and optimization of the correctable factors is essential for tailoring overall management strategy and obtaining favorable outcomes. Tumor location, stage and size and preoperative neoadjuvant chemotherapy or chemoradiotherapy, type of surgery and reconstruction, operative time, blood loss are some of the important disease and treatment related factors that have an association with post operative morbidity. High volume centers that practice a standardized and reproducible surgical technique report a relatively low overall and procedure related morbidity rate. The various risk factors associated with post esophagectomy morbidity have been listed in Table 4. A multidisciplinary approach with a personalized treatment strategy tailored for an individual patient that ensures comprehensive risk management should be implemented in order to mitigate post operative morbidity and specifically address the requirements of the high-risk patient groups in order to improve safety and treatment outcomes[47,48].

Table 4 Risk factors found associated with morbidity in minimally invasive esophagectomy.

Category	Risk factor	Association with higher morbidity
Patient related	Age	> 65 years[83]
	Gender	Male gender associated with more pulmonary and overall morbidity[48]
	Body habitus and BMI	Underweight/sarcopenic or morbid obesity[84]
	Performance status	ASA and ECOG score more than 1
	Nutritional status	Malnutrition (low albumin, hemoglobin level), weight loss, sarcopenia[48,84]
	Addictions	Smoking has higher risk[85-87]
	Pre-existing co-morbidities	Diabetes, COPD, cardiac disease and higher Charlson index[83,88]
Disease related	Tumor size	Tumour size > 4 cm[4]
	Tumor stage	T3 and higher tumours[4]
	Tumor location	Cervical and upper thoracic tumours[4]
Treatment related	Neo-adjuvant CRT	Standard of care but mixed evidence regarding slightly higher morbidity with CRT[89]
	Surgical approach (MIE/hybrid/open)	Open approach[5,15,16]
	Patient position (prone/semi-prone)	Semi-prone is associated with inferior outcomes in terms of pulmonary morbidity[17,81]
	Lymphadenectomy (standard or total two field/three field)	Three field and total two field have higher morbidity rates[82]
	Conduit type	Non gastric > gastric conduit[90]
	Anastomotic level	Cervical > intrathoracic anastomosis[91]
	Anastomotic technique	Overall similar for stapled and hand sewn technique. Semi-mechanical has better reported outcomes

BMI: Body mass index; ASA: American Society of Anesthesia; ECOG: Eastern Cooperative Oncology Group; COPD: Chronic obstructive pulmonary disease; CRT: Chemoradiotherapy; MIE: Minimally invasive esophagectomy.

ROLE OF RISK PREDICTIVE NOMOGRAMS

Postoperative complications are critical undesirable events for the patient and the treating surgeon. With a reported rate ranging from 30% to 50%, they adversely affect patients undergoing esophagectomy for a variety of reasons that include delay in initiation of adjuvant therapy which indirectly impacts survival, prolonged hospital stays, psychological stress and escalating treatment costs. Thus, identification of potentially high-risk patients is critical to make individualized therapeutic decisions and tailor postoperative care strategies accordingly.

Fundamentals of a nomogram

A risk predictive or risk assessment nomogram is a tool used to compute and estimate the probability of an outcome of interest based on multiple factors taken into consideration while developing it. It combines several variables depending on their degree of association with the into a simplified single interpretable result that helps in individualized risk assessment and stratification. Nomograms have been developed and used by surgeons for assessing a variety of outcome parameters such as cancer prognostication, disease risk prediction and complication or morbidity prediction. The broad steps of nomogram development include data collection, statistical analysis and modeling (univariate and multivariate analysis), nomogram construction and internal validation followed by external validation in population[4,9,49]. A well-constructed nomogram offers many advantages: Visual representation of correlation between multiple variables and the outcome of interest, facilitates comprehension and interpretation, allows personalized risk assessment to provide a customized probability estimate for each specific patient and enables the user to make more informed decisions regarding treatment options, outcomes and prognosis and wide applicability across various related surgical procedures.

Predictive nomogram for post MIE complications

Several prediction models for postoperative morbidity have been introduced across various surgical disciplines. However, the aforementioned models could not be reliably used in esophageal cancer due to the wide spectrum of procedural morbidity and the variety of predisposing risk factors as discussed above. Numerous attempts have been made to develop nomograms specifically to predict post-esophagectomy morbidity and mortality. Most of them have primarily focused either on pre-operative indicators or postoperative clinical factors and some nomograms are too complication specific (example: Developed to predict chances of pulmonary morbidity or anastomotic leakage) that limits their utility in routine practice. External validation of many of the proposed nomograms have been questioned limiting their generalized utility. In the past decade, many high-volume centers performing MIE and open esophagectomy have developed and validated different nomographic models predicting specific complications or overall risk of adverse events after surgery[4,8,9,49-51]. The currently available nomograms developed for risk assessment of patients undergoing esophagectomy have been summarized in Tables 5 and 6.

Table 5 List of risk predictive nomograms developed for prediction of post esophagectomy morbidity.

Ref.	Study size and period	Procedure	Morbidity predicted using nomogram	Risk factors influencing morbidity	Statistical significance
Yu et al[92], 2021	604 (2018-2020)	Open esophagectomy (53%) + MIE (47%)	Anastomotic leak	Smoking index, ASA score, anastomosis level (chest/neck, anastomotic technique, prognostic nutrition index	AUC = 0.766
Tong et al[93], 2022	969 (NS)	MIE	Pulmonary complications	Age, high BMI, smoking, FEV1/FVC, chemoradiotherapy, blood loss, general anesthesia, operative time, conversion to thoracotomy	C index = 0.654
Jin et al[94], 2022	609 (2015-2019)	MIE	Post operative pneumonia	Age, gender, abdominal phase approach, thoracic operative time, duration of chest tube placement, anastomotic leakage, RLN palsy	C index = 0.77
					AUC = 0.66
Su et al[95], 2023	308 (2018-2022)	MIE	Anastomotic leak	Age, post operative delirium, pleural adhesions, post operative pulmonary complication, raised TLC, low lymphocyte	C index = 0.96
					AUC = 0.96
Chen et al[9], 2022	285 (NS)	MIE (McKeowns esophagectomy)	Anastomotic leak	Aortic calcification, preexisting heart disease, obesity, low FEV1	C index = 0.67
Xi et al[96], 2020	119 (2019-2020)	Open esophagectomy (15%) + MIE (85%)	Post operative morbidity	BMI, ASA score, diabetes, eSAS (blood loss, lowest MAP, lowest HR)	C index = 0.903 (internal validation) 0.967 (external validation)
Zhong et al[50], 2025	747 (2019-2024)	MIE	Serious adverse events	Age, alcohol consumption, COPD, neoadjuvant radiotherapy, tumour size, FEV1, calcium, hemoglobin, albumin	AUC = 0.793

A C index (Concordance index) more than 0.7 suggests good discriminative capacity of the predictive model. A value less than 0.5 indicates discrimination by random chance and a value of 1.0 represents perfect discrimination. MIE: Minimally invasive esophagectomy; ASA: American Society of Anesthesia; AUC: Area under the curve for receiver operating characteristic curve; BMI: Body mass index; FEV1: Forced expiratory volume in 1 second; FVC: Forced vital capacity; RLN: Recurrent laryngeal nerve; TLC: Total leucocyte count; eSAS: Esophageal surgical Apgar score; MAP: Mean arterial pressure; COPD: Chronic obstructive pulmonary disease; HR: Hazard ratio.

Table 6 Currently available risk prediction models developed using artificial intelligence and machine learning for predicting esophagectomy related complications.

Ref.	Study size	Study design	Objective	Results
Bolourani et al[97], 2021	2037	Retrospective	ML-based prediction model for early readmissions within 30 days following esophagectomy	ML model for clinical decision AUC: 0.72
				ML model for quality review of esophagectomy AUC: 0.74
van Kooten et al[98], 2022	4288	Retrospective	ML methods for predicting postoperative complications following esophagectomy and development of a predictive model for anastomotic leakage and cardiopulmonary complications	The AUC of 0.619 for anastomotic leakage and 0.644 for pulmonary complications
Jung et al[99], 2023	604	Retrospective	ML-based methods for predicting Clavien–Dindo grade IIIa or greater complications following esophagectomy	The AUC of neural network was 0.672 for overall Clavien-Dindo grade IIIa or higher morbidity, 0.695 for medical complications, and 0.653 for surgical complications
Klontzas et al[55], 2024	471	Retrospective	ML model combining CT and clinical variables for predicting anastomotic leakage following esophagectomy	The model achieved an AUC of 0.792

ML: Machine learning; AUC: Area under the curve for receiver operating characteristic curve; CT: Computed tomography.

Limitations of currently available models

Most of the models are user-friendly and can be easily incorporated in routine clinical practice making them good tools for quick risk stratification and prognosticating patients pre-operatively. However, the currently available risk predictive nomograms for MIE are too heterogenous in terms of the key predictors used. Different models use different predictive parameters (clinical, biochemical, preoperative or post operative parameters) and most of them have been developed on a mixed cohort of patients undergoing minimally invasive, hybrid or open esophagectomy. All but one of the models have been developed and validated on a retrospective data set and few of them focus on the overall procedure related morbidity. Majority of the models are complication specific and they include other complications as their key predictive factors which limits their utility in the preoperative period for risk stratification. Based on literature review, the authors propose a nomogram model, integrating various factors to predict post MIE complications as shown in Table 7. This proposed model awaits validation in a prospective multicentric study to confirm its reproducibility across diverse clinical settings.

Table 7 A hypothetical nomogram model for prediction of post minimally invasive esophagectomy complications proposed by the authors (needs predictive performance validation).

Risk factor	Criteria	Score
Age	< 65 years	0
	> 65 years	1
Gender	Female	0
	Male	1
ASA score	≤ Grade 2	0
	> Grade 2	1
ECOG status	≤ 1	0
	1-2	1
	> 2	3
Diabetes or cardiac co-morbidities	No	0
	Yes	1
BMI	18.5-25	0
	> 25	1
Smoking/tobacco use	No	0
	Yes	1
Hemoglobin levels	> 8 g/dL	0
	< 8 g/dL	1
Serum albumin	> 3.5	0
	< 3.5	1
Tumour size	< 4 cm	0
	> 4 cm	1
Tumour location	Middle or lower thoracic esophagus	0
	Cervical or upper thoracic esophagus	1
Tumour T stage	≤ 2	0
	> 2	1
Neoadjuvant chemoradiotherapy	No	0
	Yes	1
Surgical approach	Minimally invasive	0
	Open	1
Lymphadenectomy type	Standard or extended 2 field	0
	Total 2 field/3 field	1
Conduit type	Gastric	0
	Non-gastric	1
Anastomosis level	Intrathoracic	0
	Cervical	1

ASA: American Society of Anesthesia; ECOG: Eastern Cooperative Oncology Group; BMI: Body mass index.

Future potential of nomographic models

Models developed using AI and ML: AI and ML models compile extensive volumes of clinical and imaging data to predict operative risk with high accuracy. They use convoluted neural networks and support vector machines to analyze large datasets, recognize complex patterns and make correlations that is usually overlooked by traditional statistical methods. The implementation of these techniques aids in reducing human error. ML is the domain of AI that trains models through automated learning using large datasets for recognizing specified patterns whereas deep learning (DL), a subset of ML, utilizes complex convoluted neural networks and larger datasets to simulate a more complex form of learning in order to enhance accuracy[52]. The recent surge in the use of AI and ML in healthcare for diagnostic, therapeutic, and prognostic purposes has opened up new gateways to improve the management and outcomes of patients undergoing esophagectomy. In the past decade, AI has been widely utilized for the management of esophageal cancer particularly for facilitating early diagnosis. AI-based diagnostic imaging techniques like hyperspectral and multispectral imaging, have been employed to enhance malignancy detection rates with encouraging outcomes[53]. Recent advancement and progress have been made in establishing the utility of AI and ML for surgical risk prediction, treatment planning and survival prediction for optimizing patient care[54].

Over the past five years, various AI based models have been developed for risk prediction in patients undergoing MIE leveraging increasingly larger datasets with external validation as summarized in Table 6. The predictive performance of these early ML models is comparable to the models developed using traditional statistics. Despite this, the clinical adoption of ML models has been criticized and challenged because the quality and consistency of input data are critical for model accuracy, and discrepancies in data collection across various centers limit the generalizability of these models. Integration of such complex models into clinical workflow demands simple and user-friendly interfaces for seamless interoperability with existing electronic health record systems, which often mandates significant technological investment. The “black box” nature of certain ML algorithms is a key barrier to their widespread adoption in routine practice questioning their reliability, interpretability and reproducibility. AI and ML based models possess the capacity to analyze and process large volumes of complex data which makes them suitable for personalized risk stratification which in turn facilitates the development of individualized surgical and postoperative strategies optimized for individual patient profiles, thereby advancing the paradigm of precision medicine[55,56].

Real time intraoperative assistance tools

Computer vision is a science that deals with the ability of computers to develop a high-level understanding of digital images or videos and perform specific automated tasks such as object identification, tracking and scene recognition, which aid in intraoperative video assessment. DL-mediated video data analysis is a promising approach in offering intraoperative cognitive assistance in minimally invasive surgery. Although still in its infancy, intraoperative video data analysis using DL has yielded positive preliminary results in several tasks, including instrument recognition, phase recognition, surgical performance assessment and real time intraoperative guidance for detection of key anatomical structures. Deep labV3 and Temporal Convolutional Networks for the Operating room are two such models developed for intraoperative recognition of RLN and surgical phase recognition respectively during RAMIE[57]. Substantial research is being carried out in this field globally and their results are expected in the following decade. The use of robotic assistance, combined with these advanced technologies, is a key area of innovation in MIE[58-60].

CONCLUSION

Esophagectomy remains a complex procedure with high morbidity and MIE has demonstrated reduced complication rates, faster recovery, and oncological outcomes comparable to open surgery. Despite its advantages MIE demands skilled teams, advanced technology, and meticulous preoperative planning. Integration of emerging technology like AI and ML based predictive models and real time intraoperative assistance software coupled with standardized recovery pathways and multidisciplinary care hold promise for improving risk prediction, stratification, prognostication and overall outcomes by mitigating complications. In the future, research should focus on developing robust risk prediction models specific to MIE. These models must be developed using high-quality, prospective, multi-centric data to ensure generalizability and clinical applicability. Leveraging the full potential of AI and ML in this context can significantly enhance the accuracy of prognostication and personalized perioperative management. Surgical training programs should emphasize on the use and adoption of laparoscopic and robotic platforms integrated with the newer real time safety assistance software. Curtailing the MIE associated morbidity seems possible with the ongoing advances in minimally invasive platforms integrated with newer safety assistance tools and risk predictive models.