From surface to depth: Combining white-light and ultrasound endoscopy to predict complete response in rectal cancer

Abstract

BACKGROUND

For patients diagnosed with locally advanced rectal cancer (LARC), radical total mesorectal excision subsequent to neoadjuvant chemoradiotherapy (nCRT) is the standard therapeutic approach. Accurate assessment of complete response (CR) is critical to identify appropriate patients for the nonoperative “Watch and Wait” (W&W) strategy.

AIM

To evaluate white light endoscopy (WLE) and endoscopic ultrasound (EUS) diagnostic efficacy and develop an integrated nomogram to predict CR.

METHODS

We retrospectively analyzed 145 LARC patients receiving nCRT. A dual-review and consensus strategy classified WLE and EUS findings into three categories. Endoscopic diagnostic performance for CR prediction was analyzed, and the association between lesion thickness and CR status was evaluated. Based on logistic regression, a clinical nomogram was developed integrating significant predictors together with objective parameters (tumor location, lesion thickness). The combined model’s clinical utility was assessed by calibration curves, decision curve analysis, and receiver operating characteristic curves.

RESULTS

Of 145 patients, 35 (24.1%) achieved CR. CR correlated with distal tumor location, lower cT stage, and lower preCRT carcinoembryonic antigen level. Positive predictive values for WLE endoscopic CR (eCR) and EUS complete rectal wall restoration were 76.9%-78.6% and 66.7%-85.7%, respectively. EUS findings revealed thinner lesions in CR patients. Overall, WLE and EUS yielded area under the curve (AUC) of 0.741-0.785 and 0.727-0.794, respectively. Multivariate analysis identified WLE non-eCR [odds ratio (OR) = 0.140, P = 0.026] and EUS mass echo (OR = 0.093, P = 0.017) as independent negative predictors. We developed an combined nomogram with these above evaluated endoscopic features and lesion location, demonstrating excellent discrimination (AUC = 0.829, 95% confidence interval: 0.748-0.909), good calibration, and substantial clinical net benefit.

CONCLUSION

Combining WLE and EUS features provides a practical “surface-to-depth” model and a minimally invasive tool for predicting CR, helping clinicians to select suitable patients for W&W strategy.

Key Words: Locally advanced rectal cancer; Neoadjuvant chemoradiotherapy; Watch and Wait strategy; Complete response; White light endoscopy; Endoscopic ultrasound

Core Tip: Accurately identifying a complete response (CR) in patients diagnosed with locally advanced rectal cancer undergoing neoadjuvant treatment is the main challenge for adopting the “Watch and Wait” strategy. By combining white light endoscopy and endoscopic ultrasound features, we constructed a “surface-to-depth” nomogram achieving an area under the curve of 0.829 (95% confidence interval: 0.748-0.909) to predict CR. This integrated model provides clinicians with a practical, minimally invasive tool to select eligible candidates who might obtain clinic benefit from organ preservation management.

INTRODUCTION

In order to reduce the risk of local recurrence and improve sphincter conservation rate, radical total mesorectal excision subsequent to neoadjuvant chemoradiotherapy (nCRT) has been widely considered as a standard treatment for patients diagnosed with locally advanced rectal cancer (LARC)[1,2]. About 15%-20% of LARC patients could achieve the absence of viable cancer cells in the surgical resected specimen after preoperative nCRT, a condition referred as pathological complete response (pCR)[3]. In this clinical setting, the patients who achieve such remarkable response after nCRT be candidates for nonoperative management. Known as “Watch and Wait“ (W&W) approach, which was first proposed in 2004[4]. This strategy involves no immediate surgery but close surveillance, could significantly improve the quality of life of patients and potentially avoid the risks and complications of surgery[5].

One of these current challenges for the strategy lies in the accurate and effective evaluation of treatment response, especially the precise identification of patients who are truly suitable for W&W. Currently, the clinical assessment of tumor regression after nCRT mainly relies on a variety of radiological means, including magnetic resonance imaging (MRI) and positron emission tomography-computed tomography (PET-CT)[6-8]. Although MRI is the most important modality for evaluating tumor regression in LARC patients after nCRT, radiation-induced tissue edema and fibrosis often obscure visualization of residual tumors, resulting in limited specificity and a propensity to either overestimate or underestimate tumor response[9]. Likewise, PET-CT provides excellent metabolic information, but its local diagnostic performance for predicting complete response (CR) is limited by post-radiation inflammatory changes. Given its ability to display the lesion morphology directly, endoscopic evaluation has become a critical component for evaluating tumor regression after preoperative chemoradiotherapy (CRT).

Conventional white light endoscopy (WLE) provides the most direct information regarding superficial mucosal alterations and enables evaluation of tumor regression by identifying features such as flat scars, telangiectasia, disappearance of neoplastic nodules, and resolution of ulcerations[10-12]. However, WLE assessment is partially relies on the endoscopist‘s subjective experience. In recent years, to improve the accuracy of assessment, more studies have focused on the evaluative role of advanced endoscopic techniques such as chromoendoscopy, magnifying endoscopy, narrow-band imaging and quantitative fluorescence endoscopy[13-15].

Endoscopic ultrasound (EUS) is equipped with an ultrasound probe at the tip of a conventional colonoscope, which can visualize the five-layer structure of the rectal wall rather than only the mucosal changes as with WLE.

EUS is routinely used for initial staging of LARC, as it can provide accurate and reliable information regarding both the depth of tumor invasion (Tstage) and regional lymph node involvement (Nstage), whereas its diagnostic accuracy for restaging after nCRT remains controversial[16,17]. A small-sample prospective study found that EUS yielded better diagnostic performance than MRI for assessing rectal carcinoma patients after neoadjuvant therapy, presenting improved sensitivity and specificity[18].

However, few studies have analyzed how to systematically integrate WLE and EUS features to maximize diagnostic accuracy. By analyzing surface morphological features under WLE and structural changes of the rectal wall under EUS, our study aims to establish a more precise endoscopic assessment system, thereby providing robust and reliable clinical evidence for physicians in screening patients suitable for the “W&W” strategy.

MATERIALS AND METHODS

Study population

We searched the database to identify consecutive patients who underwent flexible colonoscopy between 1 January 2019 and 31 December 2022 and received nCRT in the Sun Yat-sen University Cancer Center. Their clinical data were retrospectively reviewed. Inclusion criteria were as follows: (1) Primary rectal adenocarcinoma; (2) Tumor staged T3/T4, or with radiologically confirmed clinically positive regional lymph nodes on CT or rectal MRI regardless of T stage; and (3) Post-treatment endoscopy restaging performed within 3 weeks to 12 weeks after the completion of neoadjuvant therapy. Exclusion criteria consisted of: (1) Patients with recurrent rectal cancer; (2) Underwent palliative CRT; and (3) Anal squamous cell cancer. Endoscopically, tumors located within 15 cm from the anal margin were defined as rectal cancer. Lesions located within 5 cm from the anal margin were defined as distal rectal tumor, those at 5-10 cm and 10 cm-15 cm were classified as middle rectal and proximal rectal cancer.

A total of 13 patients were assigned to the W&W group, all with clinical follow-up data of at least 24 months. The remaining patients (n = 132) in radical surgery (RS) group underwent standardized curative resections in less than 8 weeks after restaging endoscopy. All operative specimens underwent histopathological analysis, and the pathological TNM classification and stage were assigned in line with the classification established by the American Joint Committee on Cancer (AJCC). The histopathologic regression was categorized into four grades according to the criteria of Tumor Regression Grade (TRG, AJCC 8^th): Grade 0, Grade 1, Grade 2, Grade 3. TRG Grade 0 was defined as CR. In W&W group, sustained CR lasting more than 24 months with no signs of luminal tumor regrowth was defined as CR. The study was carried out in accordance with the guidelines of the Declaration of Helsinki, and obtained approval from the Institutional Review Board (Ethics Committee).

Assessment of WLE

All the WLE images were re-reviewed by two experienced endoscopist (Chi J and Bai KH). They were asked to assign the most predominant endoscopic feature of each patient’s white light endoscopic images according to MKSCC criteria[19].

Endoscopic CR (eCR) was defined as lesions that met the criteria listed below: (1) White and flat scar; (2) Telangiectasia; (3) No ulcer; and (4) No nodularity (Figure 1A).

Figure 1 Representative white light endoscopy and endoscopic ultrasound images at response assessment. A: White light endoscopy (WLE) image displaying endoscopic CR (eCR) morphology, characterized by a flat white scar with telangiectasia, no ulceration, and no nodularity; B: WLE image displaying near-eCR morphology, mild mucosal irregularity and superficial ulceration without obvious tumor mass; C: WLE image displaying non-eCR morphology, with visible exophytic or ulcerative tumor lesion at the original tumor site; D: Endoscopic ultrasound (EUS) image displaying completely restored wall layers, showing clear, distinct, and continuous anatomical layers; E: EUS image displaying partially restored layers, demonstrating some focal structural disruption or thickening but with visible layer structures; F: EUS image displaying mass echo pattern, presenting as a distinct hypoechoic lesion disrupting the normal wall layers.

Near-eCR included lesions that satisfied any of conditions as follows: (1) Irregular mucosa; (2) Small mucosal nodules or faint mucosal abnormalities; (3) Superficial ulceration; and (4) Mild persistent erythema of the scar (Figure 1B).

Non-eCR was characterized by visible tumor (Figure 1C).

The two readers were blinded to each other’s interpretation, as well as the clinical examination and surgical histopathology of the patients. They were only aware that the patients had received chemoradiotherapy but were unware of whether they underwent surgery or W&W. Assessments from two independent readers were then compared, and discrepant results were resolved by joint discussion to generate a final consensus outcome.

Endoscopic biopsies were not performed as standard clinical procedure, but were conducted according to the judgement of endoscopists. No (pre)malignant feature of the biopsy histology was defined as negative, or else was positive.

EUS examination

Radial echoendoscopes (OLYMPUS/FUJIFILM) were adopted for all EUS examinations in our study. On EUS images, the normal rectal wall showed a characteristic five-layer structure, and tumors appeared as hypoechoic lesions. The maximum thickness of the lesion was measured and recorded. Measurements obtained at baseline were defined as EUS1, and those obtained after neoadjuvant treatment as EUS2. All restaging EUS images after neoadjuvant treatment were also categorized by two independent readers (Chi J and Bai KH) into three grades based on transmural morphological features of the rectal wall:

Grade 1: The layers of the rectal wall were completely restored with clear, distinct and continuous anatomical stratification (Figure 1D).

Grade 2: The layers of the rectal wall were partially restored, demonstrating focal structural disruption or thickening, yet with identifiable layered architecture and no obvious mass echo (Figure 1E).

Grade 3: Obvious mass echo was present (Figure 1F).

Consistent with our approach for WLE assessment, a dual-review and consensus strategy was performed.

Statistical analysis

Statistical analyses were performed using SPSS V26.0 (IBM, United States) and R 4.6.0. Two-sided P value < 0.05 was defined as statistical significance.

Baseline characteristics were summarized descriptively: Non-normally distributed continuous variables were reported as median (interquartile range, IQR), and categorical variables as n (%). Cohen’s kappa (κ) was used to evaluate interobserver agreement. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and diagnostic accuracy were calculated to analyze endoscopic characteristics and biopsy performance. Receiver operating characteristic (ROC) curves and the area under the curve (AUC) with 95% confidence interval (CI) were used to assess the diagnostic performance of WLE and EUS for CR prediction. Pearson correlation analysis was applied to examine the linear correlation between EUS-measured maximum thickness and CR status. Categorical variables and non-normally distributed continuous variables were compared via the χ² test and the Mann-Whitney U test, respectively. Independent predictors were identified by univariate and multivariate logistic regression. A nomogram was constructed based on these predictors. The performance of the integrated model was evaluated from three parts: Discrimination (measured by AUC), calibration [assessed via calibration plots with 1000 bootstrap resamples and the Hosmer-Lemeshow (H-L) test], and clinical utility [evaluated by decision curve analysis (DCA)].

RESULTS

Patient characteristics

A total of 145 patients were enrolled for analysis (Figure 2). Of these patients, 100 (69.0%) were men and median age was 58 (IQR 51.5-65.5) years. The median duration between the completion of CRT and endoscopic evaluation was 9 (IQR 7-10) weeks. As Figure 2 showed, 132 (91.0%) patients subsequently underwent RS, of whom 23 (17.4%) had CR (TRG Grade 0) at pathologic assessment. 13 (9.0%) patients underwent a W&W strategy, of which one patient experienced a luminal regrowth after 40 months. The remaining 12 (92.3%) patients had sustained CR status, with a median follow-up time of 42.5 (IQR 37.25-46.75) months. Of total, 35 (24.1%) patients were considered as CR, and 110 (75.8%) were non-CR. Baseline characteristics of all included patients were summarized in Table 1.

Figure 2 Flowchart of patient selection and study protocol. CR: Complete response; pCR: Pathological complete response; LARC: Locally advanced rectal cancer; nCRT: Neoadjuvant chemoradiotherapy.

Table 1 Baseline of enrolled patients, n (%).

	All	CR	Non-CR	P value
Total	145	35	110
Sex				0.633
Male	100 (69.0)	23 (65.7)	77 (70.0)
Female	45 (31.0)	12 (34.3)	33 (30.0)
Age, median (IQR), year	58 (51.5-65.5)	56 (51-65)	58 (51.3-66)	0.65
Location				0.035
Proximal	34 (23.4)	3 (8.6)	31 (28.2)
Middle	80 (55.2)	21 (60.0)	59 (53.6)
Distal	31 (21.4)	11 (31.4)	20 (18.2)
Histology				0.801
Well/moderate	125 (85.5)	31 (88.6)	93 (84.5)
Poor	17 (11.7)	3 (8.6)	14 (12.7)
Special	4 (2.8)	1 (2.9)	3 (2.7)
cT				< 0.001
cT2	11 (7.6)	8 (22.9)	3 (2.7)
cT3	83 (57.2)	23 (65.7)	60 (54.5)
cT4	51 (35.2)	4 (11.4)	47 (42.7)
cN				0.361
cN0	22 (15.2)	7 (20.0)	15 (13.6)
cN+	123 (84.8)	28 (80.0)	95 (86.4)
cM				1
cM0	141 (97.2)	34 (97.1)	107 (97.3)
cM+	4 (2.8)	1 (2.9)	3 (2.7)
Pre-nCRT level of CEA, median (IQR), ng/mL	4.84 (2.48-12.59)	2.72 (1.74-9.15)	5.5 (2.64-15.04)	0.019
Post-nCRT level of CEA, median (IQR), ng/mL	2.42 (1.57-4.14)	2.28 (1.54-3.54)	2.49 (1.57-4.42)	0.32
Interval C/RT to endoscopy, median (IQR), weeks	9 (7-10)	9 (7-11)	8 (7-10)	0.207
Interval endoscopy to surgery, median (IQR), weeks1	3 (2-5)	5 (2-6)	3 (2-5)	0.094

¹Data in patients undergoing surgery.

CR: Complete response; nCRT: Neoadjuvant chemoradiotherapy; CEA: Carcinoembryonic antigen; IQR: Interquartile range.

Biopsies

Biopsies were taken in 9 of 145 patients (6.2%). The histology findings of biopsy showed no premalignant or malignant features in 3 (33.3%) patients, adenoma in 1 (11.1%) patients, low-grade dysplasia (LGD) in 1 (11.1%) patients, high-grade dysplasia in 2 (22.2%) patients, and adenocarcinoma in 2 (22.2%) patients. 2 out of 3 patients had no residual tumor on a biopsy were proved to be non-CR, while 1 out of 6 patients had a positive biopsy histology were CR. Endoscopic biopsy findings were consistent with the histopathologic response assessment, with an accuracy of 77.8%, sensitivity of 66.7%, specificity of 83.3%, PPV of 66.7%, NPV of 83.3%, Youden’s index of 0.5, and P value of 0.226 (Table 2).

Table 2 Correlation between endoscopic biopsy and assessment of histologic response.

Biopsy histology, n	CR	Non-CR	Total	Diagnostic performance	Value (%)
Negative	2	1	3	Accuracy	77.8
Positive	1	5	6	Sensitivity	66.7
Adenoma	0	1		Specificity	83.3
LGD	1	0		PPV	66.7
HGD	0	2		NPV	83.3
Adenocarcinoma	0	2		Youden’s index	0.5
Total	3	6	9	(P = 0.226)

LGD: Low-grade dysplasia; CR: Complete response; HGD: High-grade dysplasia; PPV: Positive predictive value; NPV: Negative predictive value.

Predictive ability of white light endoscopic features

Two independent endoscopists (R1, R2) evaluated the predictive performance of WLE features for CR, with an interobserver κ-coefficient of 0.624 (95%CI: 0.503-0.746), indicating good concordance.

Among the various WLE features (Table 3), eCR was obviously the most predictive manifestation of a CR (76.9% to 78.6%). While, near-eCR was predictive of a CR in 41.4% to 45.2%.

Table 3 Positive predictive value for a complete response of endoscopic feature and endoscopic ultrasound feature.

	R1, % (n/N)	95%CI	R2, % (n/N)	95%CI
Endoscopic features
eCR	78.6 (11/14)	52.4-94.3	76.9 (10/13)	49.8-92.5
Near-eCR	45.2 (14/31)	27.8-63.4	41.4 (12/29)	24.3-59.9
Non-eCR	10.0 (10/100)	5.3-17.5	12.6 (13/103)	7.1-20.5
Endoscopic ultrasound features
Completely restored	66.7 (10/15)	42.8-90.5	85.7 (12/14)	57.2-98.2
Partially restored	35.0 (14/40)	20.2-49.8	34.0 (16/47)	20.8-49.3
Mass echo	12.2 (11/90)	5.5-19.0	8.3 (7/84)	3.4-16.4

eCR: Endoscopic complete response; near-eCR: Near-endoscopic complete response; non-eCR: Non-endoscopic complete response; CI: Confidence interval.

Overall AUC was calculated based on the 3-point ordinal scale (non-eCR, near-eCR, and eCR) to evaluate the overall diagnostic performance (Figure 3A). The overall diagnostic performance for predicting CR was comparable between the two readers, with an AUC of 0.785 (95%CI: 0.689-0.882) for Reader1 (R1) and 0.741 (95%CI: 0.636-0.846) for R2. The AUC calculated based on consensus interpretations (R1 + R2) was 0.764 (95%CI: 0.661-0.866).

Figure 3 Receiver operating characteristic curves for predicting complete response by two independent endoscopists (R1, R2) and consensus strategy by joint discussion (R1 + R2) (n = 145). A: White light endoscopy; B: Endoscopic ultrasound. AUC: Area under the curve; CI: Confidence interval.

The stringent strategy (SS) defined CR as eCR only, while the lenient strategy (LS) considered eCR combined with near-eCR as CR. For the two strategies evaluated, all diagnostic accuracy parameters are summarized in Table 4. When using the SS (eCR only), both readers demonstrated excellent specificity (97.3%) but low sensitivity (31.4% for R1 and 28.6% for R2). Conversely, applying the LS (including near-eCR as positive) substantially improved sensitivity to 71.4% for R1 and 62.9% for R2, accompanied by a high (NPV ≥ 87.4%), though at the cost of reduced specificity (81.8%).

Table 4 Diagnostic accuracy of endoscopy using a confidence level score for the prediction of a complete response.

	Stringent strategy1 (only eCR)		Lenient strategy2 (eCR + near-eCR)
	R1	R2	R1	R2
Sensitivity, % (n/N)	31.4 (11/35)	28.6 (10/35)	71.4 (25/35)	62.9 (22/35)
95%CI	16.1-46.8	13.6-43.5	56.5-86.4	46.9-78.9
Specificity, % (n/N)	97.3 (107/110)	97.3 (107/110)	81.8 (90/110)	81.8 (90/110)
95%CI	94.2-100.0	94.2-100.0	74.6-89.0	74.6-89.0
Accuracy, % (n/N)	81.4 (118/145)	80.7 (117/145)	79.3 (115/145)	77.2 (112/145)
95%CI	75.1-87.7	74.3-87.1	72.7-85.9	70.4-84.1
PPV, % (n/N)	78.6 (11/14)	76.9 (10/13)	55.6 (25/45)	52.4 (22/42)
95%CI	57.1-100.0	54.0-99.8	41.0-70.1	37.3-67.5
NPV, % (n/N)	81.7 (107/131)	81.1 (107/132)	90.0 (90/100)	87.4 (90/103)
95%CI	75.1-88.3	74.4-87.7	84.1-95.9	81.0-93.8

¹The stringent strategy: Only “endoscopic complete response” was considered as a positive test result.

²The lenient strategy: Both “endoscopic complete response” and “near-endoscopic complete response” were considered as positive test results.

eCR: Endoscopic complete response; near-eCR: Near-endoscopic complete response; PPV: Positive predictive value; NPV: Negative predictive value; CI: Confidence interval.

Predictive ability of EUS features

Substantial interobserver agreement was observed for EUS feature evaluation, with a κ value of 0.710 (95%CI: 0.604-0.815).

As presented in Table 3, completely restored yielded the highest PPV across all EUS features: 66.7% to 85.7% of patients exhibiting a CR. By comparison, partially restored and mass echo were predictive of a CR in 34%-35% and 8.3%-12.2%, respectively.

ROC curves were further constructed to quantify and compare the predictive efficacy for CR based on independent assessment and consensus diagnosis following discrepant-case discussion (Figure 3B). The corresponding AUC values were 0.727 (95%CI: 0.624-0.830) for R1, 0.794 (95%CI: 0.702-0.886) for R2, and 0.774 (95%CI: 0.677-0.872) for consensus discussion (R1 + R2), respectively.

Among the 145 patients, 73 had available data on initial tumor thickness (EUS1). The median maximum thickness at EUS1 and EUS2 was 11.0 mm (5.0 mm-39.7 mm) and 7.3 mm (2.3 mm-29.0 mm) (Figure 4A). The correlation between maximum tumor thickness on EUS and tumor response was evaluated using Pearson correlation analysis among 73 patients with complete data, as summarized in Table 5.

Figure 4 Changes in maximum tumor thickness measured by endoscopic ultrasound. Data are presented as the median (range) in millimeters (mm) (n = 73). A: Comparison of tumor thickness at baseline (EUS1) and post-treatment (EUS2); B: Comparison of post-treatment tumor thickness (EUS2) between the complete response (CR) and non-CR groups. CR: Complete response; EUS: Endoscopic ultrasound.

Table 5 Correlation between maximum thickness by endoscopic ultrasonography and tumor response (n = 73).

	CR
	Correlation factor	P value
EUS1	-0.014	0.907
EUS2	-0.232	0.048
EUS2/EUS1	-0.226	0.055

CR: Complete response; EUS: Endoscopic ultrasound.

A negative correlation was observed between preoperative tumor thickness (EUS2) and CR status (r = -0.232, P = 0.048), with thinner tumors predictive of CR achievement. The EUS2/EUS1 ratio also showed a trend toward negative correlation with CR status (r = -0.226, P = 0.055), which, while not statistically significant, implies that a greater reduction in tumor thickness from baseline to post-treatment is associated with a higher likelihood of CR. As shown in Figure 4B, the median post-treatment tumor thickness (EUS2) was 6.5 mm (range: 2.3 mm-12.2 mm) in the CR group and 7.7 mm (range: 3.0 mm-29.0 mm) in the non-CR group. Patients with CR exhibited a significantly thinner post-treatment tumor thickness compared to those with non-CR (P = 0.011).

Combined WLE with EUS features to predict CR

Table 6 presented the results of univariate and multivariate logistic regression analyses evaluating WLE and EUS features for predicting CR. To minimize interobserver bias, the final consensus dataset (R1 + R2) was adopted for all subjective WLE and EUS evaluations. In univariate analysis, compared with proximal lesions, distal lesions were associated with a significantly higher rate of CR (OR = 5.683, P = 0.015), and middle lesions showed a similar trend (OR = 3.678, P = 0.047). For WLE features, non-eCR (OR = 0.036, P < 0.001) and near-eCR (OR = 0.218, P = 0.045) morphology were both significantly associated with a lower likelihood of CR. Regarding EUS features of bowel wall layer restoration, compared with completely restored layers, partially restored layers (OR = 0.101, P = 0.006) and mass echo pattern (OR = 0.02, P < 0.001) were both significantly correlated with reduced CR. Additionally, each 1-mm increase in maximum thickness on EUS2 was associated with a significantly lower rate of CR (OR = 0.823, P = 0.01). In the multivariate model, both non-eCR morphology on WLE (OR = 0.140, P = 0.026) and mass echo on EUS (OR = 0.093, P = 0.017) remained independent negative predictors of CR.

Table 6 Univariate and multivariate logistic regression analyses of endoscopic features to predict complete response.

Characteristics	Univariate analysis			Multivariate analysis
	OR	95%CI	P value	OR	95%CI	P value
Location
Middle vs proximal	3.678	1.017-13.301	0.047	2.505	0.589-10.655	0.214
Distal vs proximal	5.683	1.409-22.929	0.015	2.792	0.547-14.244	0.217
WLE feature
Near-eCR vs eCR	0.218	0.049-0.963	0.045	0.425	0.079-2.283	0.319
Non-eCR vs eCR	0.036	0.009-0.146	< 0.001	0.140	0.025-0.792	0.026
EUS feature
Partially vs completely restored	0.101	0.020-0.517	0.006	0.219	0.035-1.373	0.105
Mass echo vs completely restored	0.02	0.004-0.106	< 0.001	0.093	0.013-0.658	0.017
Max thickness (EUS2)	0.823	0.711-0.954	0.01	0.969	0.831-1.13	0.691

Reference groups: Proximal (location); endoscopic complete response (white light endoscopy); completely restored (endoscopic ultrasound). OR: Odds ratio; CI: Confidence interval; eCR: Endoscopic complete response; near-eCR: Near-endoscopic complete response; WLE: White light endoscopy; EUS: Endoscopic ultrasound.

A nomogram prediction model was developed based on independent predictive factors (Figure 5A). By summing the scores of each factor, clinicians can calculate the total points to estimate the individual predicted value. Calibration analysis (Figure 5B) demonstrated favorable agreement between the predicted and the actual probabilities for CR. The mean absolute error was as low as 0.032, with the H-L test showing no statistical significance (P = 0.507). As DCA shown in Figure 5C, the nomogram (blue line) provided a higher standardized net benefit than both “treat-all” (orange line) and “treat-none” (green line) strategies across a broad range of threshold probabilities from approximately 0.05 to 0.90. As depicted in Figure 5D, the AUC for the nomogram predicting CR status was 0.829 (95%CI: 0.748-0.909). At the optimal cutoff value, the nomogram yielded a sensitivity of 77.1% and a specificity of 75.4%.

Figure 5 Development and validation of the predictive nomogram for complete response. A: Nomogram incorporating clinical location, whitelight endoscopy, and endoscopic ultrasound feature to estimate the individualized probability of complete response; B: Calibration curve demonstrating excellent agreement between the nomogram-predicted and actual observed probabilities; C: Decision curve analysis showing that the nomogram provides a superior clinical net benefit across a wide range of threshold probabilities compared to default strategies; D: Receiver operating characteristic curve for the nomogram predicting complete response. CR: Complete response; WLE: White light endoscopy; EUS: Endoscopic ultrasound; H-L: Hosmer-Lemeshow; AUC: Area under the curve; CI: Confidence interval.

DISCUSSION

Needless to say, the accurate identification of patients with true CR is the cornerstone of the “W&W” strategy, in which endoscopic examination serves as a crucial component in the evaluation process. The endoscopic criteria for complete clinical responders proposed by Habr-Gama et al[20] in 2010 remain a fundamental standard in clinical practice, defined by mucosal whitening, telangiectasia with mucosal integrity, and no evidence of superficial or deep ulceration, palpable nodules, or stenosis. As mounting studies focused on post-treatment endoscopic findings, the three-tier endoscopic assessment has been proposed, which establishes the neareCR category to bridge the diagnostic gap between eCR and noneCR[11,19]. This category encompasses subtle abnormalities such as faint erythema, minimal mucosal nodules, or closed ulcer with thin white moss and covered by regenerated mucosa.

In the present study, we adopted the three-tier endoscopic assessment framework to stratify treatment response, which has demonstrated reliable diagnostic performance in predicting CR. The overall diagnostic utility was consistent (κ = 0.624) between the two readers, achieving acceptable to good areas under the curve (AUC: 0.785 for R1, 0.741 for R2 and 0.764 for R1 + R2). These consistent results show that this endoscopic classification (eCR, near-eCR, and non-eCR) is reproducible and does not rely excessively on the endoscopist’s subjective experience.

Under the SS (defining only eCR as a positive result), both readers exhibited relatively high specificity (97.27%) and PPV (≥ 76.9%). This strict threshold acts as a rule-in strategy in clinical practice, which ensures that patients diagnosed with eCR are highly likely to be true complete responders. However, the major limitation of this conservative strategy is its markedly low sensitivity (approximately 28%-31%). If clinicians adopt the rigid criterion, many true CR patients would be under-selected and subjected to unnecessary surgery. Then we proposed a LS that also included near-eCR as a positive result. As expected, this modification improved the sensitivity to over 60% (71.43% for R1 and 62.86% for R2) and yielded an excellent NPV (≥ 87.38%). This LS serves as a rule-out tool: Any patient classified as non-eCR can be excluded from the cohort with CR. The inverse relationship between sensitivity and specificity at different cut-offs highlights the dilemma in endoscopic evaluation. Rather than making a binary (yes/no) decision, the near-eCR category should be treated as a transitional grade. These patients may benefit from a prolonged observation period or multimodal restaging (such as combining with other imaging modalities) to allow further tumor regression.

Meanwhile, the utility of endoscopic biopsy to stage tumor response after treatment warrants careful consideration[21,22]. Kuo et al[23] reported that post-treatment superficial rebiopsy yielded a disappointingly low predictive value for pCR, at only 21.4%. In an exploratory analysis of a small subset of 9 patients who underwent endoscopic biopsy, we observed an accuracy of 77.8% and a sensitivity of 66.7% (P = 0.226). One-third of patients with negative biopsy results was ultimately proven to be non-CR in the final surgical pathology (false negative), probably due to insufficient biopsy tissue. Conversely, an intriguing paradoxical phenomenon was observed in present study: One of the six patients with a positive restaging endoscopic biopsy showing LGD was ultimately confirmed to have achieved pCR at resection. While rare, such false positive biopsy results have also been documented in previous studies[24,25]. This discrepancy can be largely attributed to two mechanisms: The biopsy forceps might accidentally remove microscopic residual tumor (curative biopsy); alternatively, the tumor may have completely regressed during the interval between postCRT colonoscopy and surgery, which was nearly 6 weeks in our case. However, due to the extremely limited sample size, the relevant findings should be interpreted with great caution.

Despite its established role in locoregional staging of rectal cancer, EUS performance in restaging post nCRT remains debated: Radiation-induced profound fibrosis, mural edema, and inflammatory cell infiltration typically present as hypoechoic areas that are sonographically indistinguishable from residual viable tumor tissue[16,26]. While the morphological challenges, our study suggests that quantitative measurements derived from EUS hold substantial predictive value. We observed that the post-treatment maximum tumor thickness (EUS2) was significantly thinner in the CR group compared to the non-CR group (6.5 mm vs 7.7 mm, P = 0.011), and thinner tumors were negatively correlated with non-CR status (P = 0.048). Furthermore, the tumor thickness reduction ratio (EUS2/EUS1) showed a trend toward significance (P = 0.055). These findings indicate that absolute tumor bed thickness and the relative dynamic relative changes may provide complementary information to evaluate tumor regression. The lack of strict statistical significance for this ratio may be attributed to the sample size of patients with paired data (n = 73). Consistent results were also observed in a small-scale prospective study[26]. In clinical practice, rather than being employed as a stand-alone tool to rule in or rule out CR, EUS might serve as an adjunctive parameter alongside WLE visual features, based on the quantitative measurement of post-treatment mural thickness, to optimize the multidisciplinary selection of candidates for organ preservation.

Univariate analysis revealed that several endoscopic characteristics, including tumor location, WLE morphology, restoration of bowel wall layer on EUS and maximum thickness, were significantly associated with CR status. Among all evaluated factors, a non-eCR morphology on WLE (OR = 0.140, P = 0.026) and mass echo on EUS (OR = 0.093, P = 0.017) emerged as the independent negative predictors of achieving CR in the multivariate model. Despite lacking multivariate significance, tumor location and max thickness were also incorporated into the predictive nomogram as objective anatomical and quantitative morphometric parameters. This final model integrated four routinely obtainable clinical and endoscopic variables-location, WLE feature, Max thickness, and EUS feature-to quantify the individual probability of achieving CR. To highlight the incremental value of our combined approach, models relying solely on WLE features yielded AUC values below 0.80 (0.741-0.785). By contrast, integrating EUS-derived ‘depth’ parameters elevated the nomogram’s AUC to 0.829. This combination exhibited excellent predictive performance with minimal calibration bias. DCA further comfirmed its superior clinical value over “treat-all” or “treat-none” strategies across a broad range of threshold probabilities. WLE and EUS exert their complementary strengths in the “surface-to-depth” model and provides a practical, quantitative tool for guiding the W&W strategy: WLE performs well in detecting subtle mucosal abnormalities, but it cannot evaluate deep mural involvement. While EUS effectively compensates for this limitation by assessing the recovery of intestinal wall layers and quantitatively measuring transmural thickness.

Several limitations of the present study should be addressed. First of all, this research was retrospective and limited to a single medical center. Only static endoscopic images obtained by other endoscopists were evaluated, and no video recordings were available. This study design inherently carries potential selection bias, which markedly restricts the external validity and generalizability of our conclusions to broader patient populations or other medical centers. Second, the sample size was relatively small. In this work, CR was defined to include both (pCR, confirmed by surgical specimens) and clinical CR (cCR, determined after more than 2 years’ follow-up within the W&W group). Although both categories indicate no residual viable tumor, they rely on distinct assessment criteria, introducing heterogeneity in outcome definitions. The limited sample size of the W&W subgroup further prevents robust independent subgroup analyses. Larger prospective cohorts focused on the W&W strategy are warranted to specifically validate model performance for cCR identification. Third, endoscopic evaluation was subjective, and inter-observer variability may affect the reproducibility of WLE and EUS features. Fourth, our model was only internally validated using bootstrap resampling. Without external validation as an independent cohort, its reproducibility and general clinical applicability still require additional verification. Notwithstanding these limitations, our study has important clinical implications for the management of LARC patients after nCRT.

CONCLUSION

WLE and EUS are complementary tools for evaluating cCR in LARC patients after nCRT. Notably, all variables included in our model are derived from minimally invasive, and routinely performed clinical procedures, without additional costs, thus greatly enhancing its translational value and clinical feasibility. The combined WLE-EUS model employed in our study, which shifts from a surface-only perspective to a holistic surface-to-depth assessment, has the potential to serve as a practical tool for CR prediction and provide a user-friendly, quantitative tool for guiding the W&W strategy. However, its clinical application warrants further validation in prospective, multicenter cohorts.