Predicting anastomotic leak in patients with esophageal squamous cell cancer treated with neoadjuvant chemoradiotherapy using a nomogram based on CT radiomic and clinicopathologic factors

doi:10.21203/rs.3.rs-4070389/v1

Background

Anastomotic leak (AL) is a common complication in patients with operable esophageal squamous cell carcinoma (ESCC) treated with neoadjuvant chemoradiotherapy (NCRT) and radical esophagectomy. Therefore, this study aimed to establish and validate a nomogram to predict the occurrence of AL.

Methods

Between March 2016 and December 2022, 231 eligible patients with ESCC who underwent NCRT and radical esophagectomy were divided into training (n = 159) and validation cohorts (n = 72). Clinicopathologic and radiomics characteristics were included in the univariate logistic regression analysis, and statistically significant factors were enrolled to develop the nomogram, which was evaluated by the area under the curve (AUC) of the receiver operating characteristic curve, calibration curve, and decision curve analysis (DCA).

Results

Univariate and multivariate analyses revealed that dose at the anastomosis ≥ 24 Gy, gross tumor volume ≥ 60 cm3, postoperative albumin < 35 g/L, comorbidities, duration of surgery ≥ 270 mins, and computed tomography-based radiomics characteristics were independent predictors of AL. The nomogram AUC in the training and validation cohorts was 0.845 (95% confidence interval [CI]: 0.770–0.920) and 0.839 (95% CI: 0.718–0.960), respectively, indicating good discriminatory ability. The calibration curves showed good agreement between the predicted and actual AL occurrence and the DCA demonstrated favorable clinical outcomes.

Conclusions

We developed and validated a nomogram based on radiomics and clinicopathologic characteristics. This predictive model could be a powerful tool to predict AL occurrence in patients with ESCC treated with NCRT.

Esophageal squamous cell carcinoma

anastomotic leak

neoadjuvant

chemoradiotherapy

nomogram

Esophageal cancer is a common malignancy of the digestive tract. According to the current global cancer statistics, it has the seventh and sixth highest incidence and mortality rates, respectively [1]. Based on the CROSS and NEOCRTEC5010 studies, neoadjuvant chemoradiotherapy (NCRT) and radical esophagectomy is the current standard treatment for operable locally advanced esophageal squamous cell carcinoma (ESCC) [2, 3]. However, ESCC has a high risk of anastomotic leakage (AL) after NCRT and radical esophagectomy [4]. The incidence of AL after NCRT and surgery was as high as 22.4% in the CROSS study, and 22% in the Koëter et al. study [2, 5], representing one of the major limitations of this treatment modality. AL is mainly affected by factors such as poor anastomosis and local blood supply, resulting in anastomotic tissue tearing, which induces infection, prolongs hospitalization, decreases the quality of life, and increases mortality [6–10]. Furthermore, AL after radical esophagectomy seriously affects patient safety. Therefore, promptly identifying high-risk patients for AL and actively implementing preventive measures is crucial. Radiomics is a tool for converting medical images into mineable data through high-throughput extraction of quantitative features and holds great promise for cancer diagnosis, prognosis, and treatment response predictions [11–13]. Several studies have analyzed the risk factors for AL which include chronic wasting diseases (such as diabetes), malnutrition manifestations (such as hypoalbuminemia), prolonged surgery, and postoperative pulmonary infections [14, 15]. However, most of these studies focused on high-risk factors for the development of AL only after radical esophagectomy and did not analyze patients that were treated with NCRT and surgery. Consequently, for the first time, we aimed to establish and validate, a non-invasive and personalized predictive tool for the occurrence of AL after NCRT and esophagectomy using basic clinical, computed tomography (CT)-based radiomics, and radiotherapy data from patients with ESCC.

Patients

We retrospectively analyzed 231 patients with locally advanced ESCC who underwent NCRT and esophagectomy at the Cancer Hospital of Shandong First Medical University between March 2016 and December 2022. All methods adhered to the relevant guidelines and regulations. Patient staging was based on the eighth edition of the American Joint Committee on Cancer (AJCC) TNM classification. The inclusion criteria were pathologically confirmed ESCC; receiving NCRT and surgery; and CT scanning before the operation. The exclusion criteria were pathologic types other than esophageal squamous cell carcinoma; administration of other neoadjuvant therapy (including targeted therapy and immunotherapy); absence of preoperative CT scanning, and the dose of radiotherapy was not in the 40–50.4 Gy range (Fig. 1). AL was defined by the Esophageal Complications Consensus Group as “a total gastrointestinal defect involving the esophagus, anastomosis, sutures, or catheter regardless of the presentation or method of recognition” [16]. A minimum of two experienced radiation oncologists and imaging physicians diagnosed AL based on clinical symptoms and imaging examinations. This study was approved by the Ethics Committee of Cancer Hospital Affiliated to Shandong First Medical University, which waived the need for informed consent because of the retrospective nature of the study. We declare that patients information was kept confidential and that we adhered to the principles of the Declaration of Helsinki.

Neoadjuvant treatment regimens

All patients received NCRT before radical esophagectomy with a total dose of 40–50.4 Gy, at 1.8–2 Gy per fraction, 5 days per week. Intensity-modulated radiotherapy techniques were employed. Large-aperture CT scans were performed and CT localization images were imported into the Varian Eclipse program system. The target area was dominated by the area involved in the lesion. The gross tumor volume (GTV) included the primary tumor volume and prechemotherapy positive lymph node regions. Positive lymph nodes were defined as lymph nodes with a positive pathological biopsy, with a short axis ≥ 1 cm in the mediastinum, or clusters of small lymph nodes with a short axis < 1 cm within a region. Clinical target volume (CTV) was defined using the CTV of the primary tumor obtained from 3.0 cm of upward and downward GTV of the primary tumor expansion, 0.5 cm of margin expansion, and the CTV of the pathological lymph nodes obtained by 0.5 cm of margin expansion of the GTV from the pathological lymph nodes and edited according to anatomy as involved field irradiation. However, in this study, 53 patients with elective nodal irradiation remained, which included high-risk lymph nodal regions. The planning target volume (PTV) was defined by further expanding the CTV with a 0.5 cm margin. The preoperative chemotherapy regimen mainly comprised platinum-based drugs (nedaplatin at 75 mg/m2, carboplatin at an area under the curve (AUC) of 5, or cisplatin at 25 mg/m2 on days 1–3) along with fluorouracil at 50 mg/m2 on days 1–5 (PF regimen), or platinum-based drugs and paclitaxel at 135–175 mg/m2 and albumin-paclitaxel at 260 mg/m2 (TP regimen) via intravenous injection. Patients received 1–3 preoperative chemotherapies (PF or TP regimen) every 3 weeks, with an average of 2 treatment cycles.

Surgical treatment

Patients enrolled in this study were clinically evaluated for ESCC suitability for radical esophagectomy. Most patients underwent esophagectomy and mediastinal lymph node dissection under general anesthesia 4–8 weeks after the last neoadjuvant therapy. However, 34 patients were > 8 weeks while 11 patients were < 4 weeks. Surgery was performed using open or thoracic laparoscopy, with two- and three-incision surgical incisions and dual-field lymph node dissection as standard treatments. Standard three-field lymph node dissection was performed in patients with suspected cervical lymph node enlargement.

Radiomics

The CT scans obtained before surgical treatment were analyzed. All tumor target areas were manually depicted layer-by-layer using the medical image processing and navigation software 3D Slicer (version 5.2.1) by an author specializing in radiation oncology. Subsequently, the region of interest was confirmed by another radiologist with 10 years of experience in chest CT analysis. Z-score normalization was performed on all data as a preprocessing step to reduce the variation between different patient images. Moreover, feature selection was performed in two steps to reduce any bias or overfitting caused by several features. First, the intra-class correlation coefficient (ICC) index was calculated to filter out features with high repeatability and stability. In this study, ICC > 0.9 was used as the criterion to filter features. Features with high repeatability were initially filtered, downscaled, and filtered for key features using Least Absolute Shrinkage and Selection Operator regression [17, 18]. Finally, the radiomics score (Rad score) of each patient was determined by a linear combination of the selected characteristics, with their respective coefficients being used as weights [19].

Model construction and evaluation

Based on previous AL studies, we retrospectively analyzed basic information (such as age), radiotherapy information (such as GTV and dose at the anastomosis), and radiomics information. First, univariate logistic regression models were used in the training cohort to evaluate the predictive ability of the clinical and radiomics characteristics for AL. Second, multivariate logistic regression analysis was performed to assess factors with P < 0.05 in the univariate analysis. Third, a nomogram was constructed using factors with significant predictive values from the multivariate analysis. The discriminatory ability of the nomogram was then assessed using receiver operating characteristic (ROC) curves in the training and validation cohorts, calibration curves to assess the agreement between predicted AL and actual observations, and decision curve analysis (DCA) to validate the nomogram by quantifying the net benefit of the nomogram at different threshold probabilities to illustrate its clinical utility.

Statistical analysis

All data analyses and graphing were performed using R software version 4.0.3. The independent sample t-test and rank-sum test were employed to contrast continuous variables. The chi-square and Fisher’s exact tests were used to compare subtype variables. All tests were two-sided, and P < 0.05 indicated statistical significance.

Patients’ baseline characteristics

Ultimately, 231 patients with locally advanced ESCC who underwent NCRT and esophagectomy were included in this study, 48 (20.8%) of whom were diagnosed with AL (baseline characteristics are shown in Table 1). The patients were randomly assigned to the training (n = 159) and validation (n = 72) cohorts in a 7:3 ratio (Fig. 1). In the training and validation cohorts, AL occurred in 36 (22.6%) and 12 (16.7%) patients, respectively, and 119 (74.8%) and 59 (82.0%) patients received radiotherapy with involved-field irradiation, respectively.

Table 1

Clinical characteristics of patients in the training and validation cohorts
Characteristic	Training (n = 159)	Validation (n = 72)	P-value
Sex
Female	23 (14.5%)	6 (8.3%)	0.276
Male	136 (85.5%)	66 (91.7%)
Age (years)
≤ 60	68 (42.8%)	37 (51.4%)	0.224
> 60	91 (57.2%)	35 (48.6%)
Height¹	1.67 ± 0.07	1.68 ± 0.07	0.287
Weight¹	63.15 ± 10.41	64.46 ± 10.36	0.378
Smoking
No	64 (40.3%)	29 (40.3%)	1.000
Yes	95 (59.7%)	43 (59.7%)
Drinking
No	69 (43.4%)	35 (48.6%)	0.552
Yes	90 (56.6%)	37 (51.4%)
Comorbidities
No	95 (59.8%)	38 (52.8%)	0.321
Yes	64 (40.2%)	34 (47.2%)
Tumor location
Lower-thoracic	83 (52.2%)	34 (47.2%)	0.782
Middle-thoracic	64 (40.3%)	32 (44.4%)
Upper-thoracic	12 (7.5%)	6 (8.3%)
GTV (cm³)
< 60	112 (70.4%)	49 (68.1%)	0.715
≥ 60	47 (29.6%)	23 (31.9%)
Dose at the anastomosis (Gy)
< 24	137 (86.2%)	55 (76.4%)	0.066
≥ 24	22 (13.8%)	17 (23.6%)
Chemotherapy regimen
PF	21 (13.2%)	8 (11.1%)	0.817
TP	138 (86.8%)	64 (88.9%)
Total protein (g/L)
≤ 60	13 (8.2%)	4 (5.6%)	0.483
>60	146 (91.8%)	68 (94.4%)
Postoperative albumin (g/L)
< 35	22 (13.8%)	18 (25.0%)	0.038
≥ 35	137 (86.2%)	54 (75.0%)
Triglycerides (mmol/L)
≤ 1.7	117 (73.6%)	56 (77.8%)	0.497
>1.7	42 (26.4%)	16 (22.2%)
Red cells (10¹²/L)
≤ 4.0	80 (50.3%)	30 (41.7%)	0.224
>4.0	79 (49.7%)	42 (58.3%)
Lymphocytes (10⁹/L)
≤ 0.8	53 (33.3%)	20 (27.8%)	0.401
>0.8	106 (66.7%)	52 (72.2%)
Neutrophils (10⁹/L)
≤ 2.0	17 (10.7%)	7 (9.7%)	0.823
>2.0	142 (89.3%)	65 (90.3%)
Hemoglobin (g/L)
≤ 110	22 (13.8%)	9 (12.5%)	0.783
>110	137 (86.2%)	63 (87.5%)
Target Area
IFI	119 (74.8%)	59 (82.0%)	0.327
ENI	40 (25.2%)	13 (18.0%)
Duration of surgery (mins)
<270	117 (73.6%)	44 (61.1%)	0.056
≥ 270	42 (26.4%)	28 (38.9%)
Abbreviations: GTV, gross tumor volume; 1, presented as mean ± standard deviation; TP, platinum-based drugs and paclitaxel/albumin paclitaxel; PF, platinum-based drugs and fluorouracil; IFI, involved field irradiation; ENI, elective nodal irradiation.

Univariate and multivariate analysis

Regarding the univariate analysis in the training cohort, we observed five high-risk factors for AL as follows: dose at the anastomosis ≥ 24 Gy (25.0% vs. 10.6%, P = 0.027), GTV ≥ 60 cm3 (44.4% vs. 25.2%, P = 0.026), postoperative albumin < 35 g/L (27.8% vs. 9.8%, P = 0.006), comorbidities (58.3% vs. 20.3%, P < 0.001), and duration of surgery ≥ 270 mins (41.7% vs. 22.0%, P = 0.018) (Table 2). The eight radiomics characteristics that were most valuable in predicting AL were selected (Fig. 2), and the calculated Rad scores were (1.76 ± 1.41 vs. 0.57 ± 1.58, P < 0.001) in patients who developed AL and in those who did not.

Table 2

Univariate and multivariate analysis of clinical and radiomics characteristics for AL prediction
Univariate analysis
Characteristics	AL (n = 36)		No AL (n = 123)		P-value		Regression coefficient		OR (95% CI)		P-value
Sex
Female		8 (22.2%)		15 (12.2%)		0.217
Male		28 (77.8%)		108 (87.8%)
Age (years)
≤ 60		16 (44.4%)		58 (47.2%)		0.774
> 60		20 (55.6%)		65 (52.8%)
Smoking
No		14 (38.9%)		50 (40.7%)		1.000
Yes		22 (61.1%)		73 (59.3%)
Dose at the anastomosis (Gy)
≥ 24		9 (25.0%)		13 (10.6%)		0.027		1.520		4.572 (1.329,16.689)		0.017
< 24		27 (75.0%)		110 (89.4%)				Ref		Ref
GTV (cm³)
≥ 60		16 (44.4%)		31 (25.2%)		0.026		1.427		4.165 (1.555, 11.903)		0.006
< 60		20 (55.6%)		92 (74.8%)				Ref		Ref
Postoperative albumin (g/L)
< 35		10 (27.8%)		12 (9.8%)		0.006		1.798		6.035 (1.790, 21.769)		0.004
≥ 35		26 (72.2%)		111 (90.2%)				Ref		Ref
Tumor location
Lower-thoracic		19 (52.8%)		64 (52.0%)		0.611
Middle-thoracic		13 (36.1%)		51 (41.5%)
Upper-thoracic		4 (11.1%)		8 (6.5%)
Comorbidities
Yes		24 (58.3%)		40 (20.3%)		< 0.001		1.476		4.377 (1.729, 11.912)		0.002
No		12 (41.7%)		83 (79.7%)				Ref		Ref
Chemotherapy regimen
PF		4 (11.1%)		17 (13.8%)		0.887
TP		32 (88.9%)		106 (86.2%)
Total protein (g/L)
≤ 60		2 (5.6%)		11 (8.9%)		0.518
> 60		34 (94.4%)		112 (91.1%)
Red cells (10¹²/L)
≤ 4.0		21 (58.3%)		59 (48.0%)		0.912
> 4.0		15 (41.7%)		64 (52.0%)
Hemoglobin (g/L)
≤ 110		7 (19.4%)		15 (12.2%)		0.272
> 110		29 (80.6%)		108 (87.8%)
Rad score¹		1.76 ± 1.41		0.57 ± 1.58		<0.001		0.513		1.671 (1.229,2.353)		0.002
Duration of surgery (mins)
≥ 270		15 (41.7%)		27 (22.0%)		0.018		1.047		2.850 (1.053,7.888)		0.040
<270		21 (58.3%)		96 (78.0%)				Ref		Ref
Surgical approach
Thoracotomy		19 (52.8%)		52 (42.3%)		0.267
MIE		17 (47.2%)		71 (57.7%)
Intraoperative bleeding (ml)
≤ 100		25 (69.4%)		102 (82.9%)		0.080
> 100		11 (30.6%)		21 (17.1%)
Target Area
IFI		26 (72.2%)		93 (75.6%)		0.501
ENI		10 (27.8%)		30 (24.4%)
Abbreviations: AL, anastomotic leak; OR, odds ratio; 95% CI, 95% confidence interval; GTV, Gross Tumor Volume; TP, platinum-based drugs and paclitaxel/albumin paclitaxel; PF, platinum-based drugs and fluorouracil; Rad score, radiomics score; 1, Presented as mean ± standard deviation; MIE, minimally invasive esophagectomy; IFI, involved field irradiation; ENI, elective nodal irradiation.

Multivariate logistic regression analysis was performed on the six factors with P < 0.05 in the univariate analysis, including dose at the anastomosis ≥ 24 Gy (odds ratio [OR]: 4.572, 95% confidence interval [CI]: 1.329–16.689, P = 0.017), GTV ≥ 60 cm3 (OR: 4.165, 95% CI: 1.555–11.903, P = 0.006), postoperative albumin < 35 g/L (OR: 6.035, 95% CI: 1.790–21.769, P = 0.004), comorbidities (OR: 4.377, 95% CI: 1.729–11.912, P = 0.002), duration of surgery ≥ 270 mins (OR: 2.850, 95% CI: 1.053–7.888, P = 0.040), and Rad score (OR: 1.671, 95% CI: 1.229–2.353, P = 0.002). These six factors were independent predictors of AL (Table 2).

Modeling and validation

Based on the multivariate logistic regression analysis, the prediction model was visualized as a nomogram (Fig. 3). The ROC curve shows that the AUC of the model was 0.845 (95% CI: 0.770–0.920) in the training cohort, which was higher than the AUC of each factor (dose at the anastomosis ≥ 24 Gy, AUC: 0.572; GTV ≥ 60 cm3, AUC: 0.596; postoperative albumin < 35 g/L, AUC: 0.590; comorbidities, AUC: 0.671; duration of surgery ≥ 270 mins, AUC: 0.599; Rad score, AUC: 0.716) (Fig. 4a) (Table 3). Thus, the nomogram showed high predictive efficiency (Table 4).

Table 3

AUC (95% CI) for predictors of AL in the training and validation cohorts
Characteristic	Training AUC (95% CI)	Validation AUC (95% CI)
Nomogram	0.845 (0.770, 0.920)	0.839 (0.718, 0.960)
Dose at the anastomosis (≥ 24 Gy)	0.572 (0.495, 0.649)	0.508 (0.370, 0.647)
GTV (≥ 60cm³)	0.596 (0.505, 0.687)	0.558 (0.401, 0.715)
Postoperative albumin (< 35 g/L)	0.590 (0.511, 0.669)	0.700 (0.546, 0.854)
Comorbidities (YES)	0.671 (0.582, 0.759)	0.567 (0.408, 0.726)
Duration of surgery (≥ 270 mins)	0.599 (0.509, 0.688)	0.658 (0.407, 0.727)
Rad score	0.716 (0.629, 0.804)	0.725 (0.601, 0.849)
Abbreviations: AUC, area under the curve; GTV, gross tumor volume; 95% CI, 95% confidence interval; Rad score, radiomics score.

Table 4

Prediction performance of nomogram in training and validation cohorts.
Parameters	Training	Validation
AUC (95% CI)	0.845 (0.770, 0.920)	0.839 (0.718, 0.960)
ACC	0.811	0.861
Sen	0.722	0.667
Spe	0.837	0.900
PPV	0.565	0.571
NPV	0.912	0.931
Abbreviations: AUC, area under the curve; ACC, accuracy; Sen, sensitivity; Spe, specificity; PPV, positive predictive value; NPV, negative predictive value.

In the validation cohort, the AUC of the model was 0.839 (95% CI: 0.718–0.960), which was higher than the AUC of each factor (dose at the anastomosis ≥ 24 Gy AUC: 0.508; GTV ≥ 60 cm3 AUC: 0.558; postoperative albumin < 35 g/L AUC: 0.700; comorbidities AUC: 0.567; duration of surgery ≥ 270 mins AUC: 0.658; Rad score AUC: 0.725) (Fig. 4b). Therefore, the nomogram also showed high predictive efficiency (Table 4). Additionally, the calibration curves for the training and validation cohorts showed good agreement between the predicted AL and actual observations (Fig. 5a and b). Moreover, the DCA showed that the nomogram had satisfactory net benefits for most threshold probabilities, indicating good potential clinical outcomes for the model (Fig. 5c and d).

One of the most severe complications of esophagectomy is AL, affecting patient prognosis and increasing mortality [6–9]. The incidence of AL in this study was 20.8%, similar to the 22% reported by Koëter et al. [5] and 22.4% by the CROSS study [2]. Vande Walle et al. found that radiation dose to the fundus of the stomach was an independent predictor of anastomotic complications in patients who received NCRT and radical esophagectomy [20]. Similarly, Morita et al. found that the incidence of AL was significantly higher in patients treated with NCRT than in patients treated with surgery alone (28% vs. 16.5%) [21]. Although Koe¨ter et al. concluded that NCRT does not increase the incidence of AL, the 22% incidence is equally noteworthy [5]. To the best of our knowledge, this study is the first to develop and validate a predictive model for AL for patients undergoing NCRT and radical esophagectomy. Regarding this study, the multivariate analysis identified dose at the anastomosis ≥ 24 Gy, GTV ≥ 60 cm3, postoperative albumin < 35 g/L, comorbidities, duration of surgery ≥ 270 mins, and radiomics characteristics as independent risk factors for AL. The AUC values of the training and validation cohorts were 0.845 and 0.839, respectively, and the AUC of the nomogram in both cohorts was higher than that of each individual factor, proving that the nomogram had a high predictive value.

By analyzing the radiotherapy dose curves of each patient, we found that the dose received at the anastomosis after radical resection for esophageal cancer was strongly associated with AL occurrence. We analyzed the position of the anastomosis in each patient's postoperative CT and the isodose profile in the radiotherapy plan. When the dose at the anastomosis was ≥ 60% of the total dose (the lowest total dose included in this study was 40 Gy, therefore, the cutoff value for the dose at the anastomosis was 24 Gy), a significant increase in AL incidence was observed; in some patients, the anastomosis even remained within the PTV. If the anastomosis was exposed to a large radiation dose, subsequent anastomotic surgery and perioperative safety were affected. A similar conclusion was reached in the study by Klevebro et al. wherein the anastomosis radiation dose was strongly correlated with the severity of postoperative anastomosis complications [22]. Therefore, for patients receiving NCRT, the anastomosis should not be in the irradiation field, and a safe distance should be set aside for the anastomosis to limit the radiation dose received at the anastomosis; the use of intensity-modulated radiotherapy is routinely recommended. Moreover, if conditions permit, proton irradiation may be used in units where it is available. This is consistent with Wang et al.'s view that reducing the dose to normal tissues by using complex techniques such as intensity-modulated radiation therapy or proton beam therapy can reduce pulmonary and gastrointestinal complications [23]. However, Koëter et al. showed no significant correlation between the radiation dose and AL development at an average radiation dose of 24.2 Gy in the region of the future anastomosis [5]. Therefore, larger clinical studies are required to determine the maximum dose range at the anastomosis.

In addition, we found that GTV was an independent risk factor for AL. The larger the GTV, the larger the corresponding increase in irradiated normal tissues and resulting increase of the effect on the blood supply, which predisposes patients to AL. When the cutoff value of GTV was 60 cm3, differentiating between patients with and without AL was possible. Radiotherapy may lead to an increase in vascular fragility, thus making postoperative anastomosis more difficult. Additionally, radiotherapy may also damage the vascular endothelium, leading to fibrous tissue proliferation, which may cause vascular occlusion and blood and fluid leakage from the surgical area, affecting the prognosis of the anastomosis, similar to the view of a previous study [4]. Therefore, we suggest that for patients with large tumor volumes, NCRT should be carefully administered to avoid AL.

Malnutrition has been reported in approximately 40–60% of patients with ESCC at diagnosis [24]. In this study, we observed that a postoperative albumin level < 35 g/L was an independent risk factor for AL, similar to the results of Choudhuri et al. [25]. This is because low albumin levels affect postoperative wound and anastomotic healing, increasing AL risk. Fergus et al. established a scoring system wherein serum albumin levels affected the incidence of AL after esophagectomy and postoperative mortality [26]. Therefore, clinicians are advised to promptly assess patient’s nutritional status after surgical treatment and provide enteral or parenteral nutrition promptly when serum albumin levels are < 35 g/L [24, 27]. The study also found that comorbidities were independent risk factors for the development of postoperative AL. The comorbidities in this study included hypertension, diabetes, and chronic obstructive pulmonary disease (COPD), which were consistent with the results of Gooszen et al. for COPD (17.7% vs. 12.8%), hypertension (36.4% vs. 31.6%), and diabetes (20.1% vs. 14.5%). Thus, the patient's condition is an important factor in postoperative AL that should not be ignored, and a comprehensive preoperative evaluation should be performed to identify these risk factors as early as possible and actively manage them promptly [28].

Similar to previous studies that have shown intraoperative factors such as surgical methods are also high-risk factors for AL [29–33], in the present study, duration of surgery ≥ 270 mins was significantly associated with the occurrence of AL. The increased number of surgical operations such as repeated squeezing and lifting of the stomach causes lack of blood supply to the patient's postoperative esophageal and gastric tissue edema, thus affecting anastomotic healing.

Owing to the continuous development of imaging technologies and big data, radiomics has attracted considerable attention. Artificial intelligence is extensively utilized for extracting tumor information and establishing machine models for the prediction of tumor lymph node metastasis, tumor clinicopathological grading, and T-staging [10]. This study considered nine valuable imaging features in predicting AL and weighed them with their respective coefficients. Additionally, Rad scores were obtained and exhibited differences in univariate and multivariate analyses. This suggests the potential of radiomics to predict AL after esophagectomy. We internally validated the accuracy of our prediction model. The ROC, DCA, and calibration curves indicated that our model had a good potential clinical effect, implying that it could be replicated. Moreover, for patients predicted to have an elevated risk for AL by this nomogram, we suggest that clinicians perform regular esophagography scans and examinations to detect them promptly and provide symptomatic treatment, to reduce the impact of high-risk factors for AL.

Nonetheless, our study had some limitations. First, due to the retrospective nature of the study, selection bias was unavoidable; therefore, more prospective studies on AL are required. Second, this was a single-center study. Although internal validation indicated that the current prediction model has a relatively superior AUC of 0.839, the results will be more convincing with external validation. Finally, the sample size of the study was relatively small; therefore, future studies should include data from more research centers and larger samples.

In conclusion, we developed and validated a nomogram based on both radiomics and clinicopathologic characteristics. This predictive model can be used as a powerful tool to predict the occurrence of AL in patients with ESCC treated with NCRT and surgery.

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Cancer Hospital Affiliated to Shandong First Medical University, which waived the need for informed consent because of the retrospective nature of the study.

Consent for publication

Not applicable

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors have no relevant financial or non-financial interests to disclose.

Funding

This work was supported by the Clinical Medical Research Center of Shandong Province (2021LCZX04), Academic Promotion Program of Shandong First Medical University (2019LJ004), and Major Basic Research Project of Shandong Provincial Natural Science Foundation (ZR2022ZD31).

Author contributions

JZ and GY designed the study, analyzed and interpreted the data, and drafted the manuscript. YL, SL, HL, and DH analyzed and interpreted the data. QC and BL conceptualized and designed the study and revised the manuscript. The authors read and approved the final manuscript.

Acknowledgments

We thank the patients for allowing the use of their clinicopathological data for this study.

Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: Globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71:209-49.
van Hagen P, Hulshof MC, van Lanschot JJ, Steyerberg EW, van Berge Henegouwen MI, Wijnhoven BP, et al. Preoperative chemoradiotherapy for esophageal or junctional cancer. N Engl J Med. 2012;366:2074-84.
Yang H, Liu H, Chen Y, Zhu C, Fang W, Yu Z, et al. Neoadjuvant Chemoradiotherapy Followed by Surgery Versus Surgery Alone for Locally Advanced Squamous Cell Carcinoma of the Esophagus (NEOCRTEC5010): A Phase III Multicenter, Randomized, Open-Label Clinical Trial. J Clin Oncol. 2018;36:2796-803.
Goense L, van Rossum PSN, Ruurda JP, van Vulpen M, Mook S, Meijer GJ, et al. Radiation to the Gastric Fundus Increases the Risk of Anastomotic Leakage After Esophagectomy. Ann Thorac Surg. 2016;102:1798-804.
Koëter M, Kathiravetpillai N, Gooszen JA, van Berge Henegouwen MI, Gisbertz SS, van der Sangen MJ, et al. Influence of the Extent and Dose of Radiation on Complications After Neoadjuvant Chemoradiation and Subsequent Esophagectomy With Gastric Tube Reconstruction With a Cervical Anastomosis. Int J Radiat Oncol Biol Phys. 2016;97:813-21.
Kassis ES, Kosinski AS, Ross P Jr, Koppes KE, Donahue JM, Daniel VC, et al. Predictors of anastomotic leak after esophagectomy: An Analysis of the Society of Thoracic Surgeons General Thoracic Database. Ann Thorac Surg. 2013;96:1919-26.
Goense L, Meziani J, Ruurda JP, van Hillegersberg R. Impact of Postoperative Complications on outcomes after oesophagectomy for cancer. Br J Surg. 2019;106:111-9.
Fransen LFC, Berkelmans GHK, Asti E, van Berge Henegouwen MI, Berlth F, Bonavina L, et al. The effect of postoperative complications after minimally invasive esophagectomy on long-term survival: An international multicenter cohort study. Ann Surg. 2021;274:e1129-37.
Derogar M, Orsini N, Sadr-Azodi O, Lagergren P. Influence of major postoperative complications on health-related quality of life among long-term survivors of esophageal cancer surgery. J Clin Oncol. 2012;30:1615-9.
Scarpa M, Saadeh LM, Fasolo A, Alfieri R, Cagol M, Cavallin F, et al. Health-related quality of life in patients with oesophageal cancer: Analysis at different steps of the treatment pathway. J Gastrointest Surg. 2013;17:421-33.
Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278:563-77.
Verma V, Simone CB 2nd, Krishnan S, Lin SH, Yang J, Hahn SM, et al. The rise of radiomics and implications for oncologic management. J Natl Cancer Inst. 2017;109:10.
Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumor phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.
Huang C, Yao H, Huang Q, Lu H, Xu M, Wu J. A novel nomogram to predict the risk of anastomotic leakage in patients after oesophagectomy. BMC Surg. 2020;20:64.
Hagens ERC, Reijntjes MA, Anderegg MCJ, Eshuis WJ, van Berge Henegouwen MI, Gisbertz SS, et al. Risk factors and consequences of anastomotic leakage after esophagectomy for cancer. Ann Thorac Surg. 2020;112:255-63.
Low DE, Alderson D, Cecconello I, Chang AC, Darling GE, DʼJourno XB, et al. International Consensus on Standardization of Data Collection for Complications Associated With Esophagectomy: Esophagectomy Complications Consensus Group (ECCG). Ann Surg. 2015;262:286-94.
Zhao W, Wu Y, Xu Y, Sun Y, Gao P, Tan M, et al. The potential of radiomics nomogram in non-invasively prediction of epidermal growth factor receptor mutation status and subtypes in lung adenocarcinoma. Front Oncol. 2019;9:1485.
Zhao W, Xu Y, Yang Z, Sun Y, Li C, Jin L, et al. Development and validation of a radiomics nomogram for identifying invasiveness of pulmonary adenocarcinomas appearing as subcentimeter ground-glass opacity nodules. Eur J Radiol. 2019;112:161-8.
Liu Z, Zhang XY, Shi YJ, Wang L, Zhu HT, Tang Z, et al. Radiomics analysis for evaluation of pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. Clin Cancer Res 23:7253-62.
Vande Walle C, Ceelen WP, Boterberg T, Vande Putte D, Van Nieuwenhove Y, Varin O, et al. Anastomotic complications after Ivor Lewis esophagectomy in patients treated with neoadjuvant chemoradiation are related to radiation dose to the gastric fundus. Int J Radiat Oncol Biol Phys. 2012;82:e513-9.
Morita M, Masuda T, Okada S, Yoshinaga K, Saeki H, Tokunaga E, et al. Preoperative chemoradiotherapy for esophageal cancer: factors associated with clinical response and postoperative complications. Anticancer Res. 2009;29:2555-62.
Klevebro F, Friesland S, Hedman M, Tsai JA, Lindblad M, Rouvelas I, et al. Neoadjuvant chemoradiotherapy may increase the risk of severe anastomotic complications after esophagectomy with cervical anastomosis. Langenbecks Arch Surg. 2016;401:323-31.
Wang J, Wei C, Tucker SL, Myles B, Palmer M, Hofstetter WL, et al. Predictors of postoperative complications after trimodality therapy for esophageal cancer. Int J Radiat Oncol Biol Phys. 2013;86:885-91.
Mariette C, De Botton ML, Piessen G. Surgery in esophageal and gastric cancer patients: What is the role for nutrition support in your daily practice? Ann Surg Oncol. 2012;19:2128-34.
Choudhuri AH, Uppal R, Kumar M. Influence of non-surgical risk factors on anastomotic leakage after major gastrointestinal surgery: Audit from a tertiary care teaching institute. Int J Crit Illn Inj Sci. 2013;3:246-9.
Noble F, Curtis N, Harris S, Kelly JJ, Bailey IS, Byrne JP, et al. Risk assessment using a novel score to predict anastomotic leak and major complications after oesophageal resection. J Gastrointest Surg.2012;16:1083-95.
Messager M, Warlaumont M, Renaud F, Marin H, Branche J, et al. Piessen G, Recent improvements in the management of esophageal anastomotic leak after surgery for cancer. Eur J Surg Oncol. 2017;43:258-69.
Gooszen JAH, Goense L, Gisbertz SS, Ruurda JP, van Hillegersberg R, van Berge Henegouwen MI, et al. Intrathoracic versus cervical anastomosis and predictors of anastomotic leakage after oesophagectomy for cancer. Br J Surg. 2018;105:552-60.
Goense L, van Rossum PS, Tromp M, Joore HC, van Dijk D, Kroese AC, et al. Intraoperative and postoperative risk factors for anastomotic leakage and pneumonia after esophagectomy for cancer. Dis Esophagus. 2017;30:1-10.
Michelet P, D'Journo XB, Roch A, Papazian L, Ragni J, Thomas P, et al. Perioperative risk factors for anastomotic leakage after esophagectomy: influence of thoracic epidural analgesia. Chest. 2005;128:3461-66.
Haverkamp L, van der Sluis PC, Verhage RJ, Siersema PD, Ruurda JP, van Hillegersberg R, et al. End-to-end cervical esophagogastric anastomoses are associated with a higher number of strictures compared with end-to-side anastomoses. J Gastrointest Sur. 2013;17:872-6.
Markar SR, Arya S, Karthikesalingam A, Hanna GB. Technical factors that affect anastomotic integrity following esophagectomy: Systematic review and meta-analysis. Ann Surg Oncol. 2013;20:4274-81.
Markar SR, Karthikesalingam A, Vyas S, Hashemi M, Winslet M. Hand-sewn versus stapled oesophago-gastric anastomosis: Systematic review and meta-analysis. J Gastrointest Surg. 2015;15:876-84.

No competing interests reported.

Predicting anastomotic leak in patients with esophageal squamous cell cancer treated with neoadjuvant chemoradiotherapy using a nomogram based on CT radiomic and clinicopathologic factors

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Background

Methods

Patients

Neoadjuvant treatment regimens

Surgical treatment

Radiomics

Model construction and evaluation

Statistical analysis

Results

Patients’ baseline characteristics

Univariate and multivariate analysis

Modeling and validation

Discussion

Conclusions

Declarations

References

Additional Declarations

Status:

Version 1