Endoscopic assessment of rectal cancer response after neoadjuvant chemoradiotherapy: A narrative literature review

Abstract

BACKGROUND

Rectal cancer management is currently evolving with the advent of different neoadjuvant treatment strategies and organ preservation strategies. A significant proportion of patients could achieve complete clinical response after neoadjuvant treatment, which often translates to pathologic complete response (pCR) as assessed on surgical specimens after curative intent surgery. Endoscopy plays a significant role in assessing treatment response to neoadjuvant therapies.

AIM

To explore the role of endoscopy in predicting subsequent pCR after neoadjuvant treatment in rectal cancer patients.

METHODS

An extensive literature review was undertaken to identify the criteria used for assessment of endoscopic response and their ability to predict pCR.

RESULTS

Fifteen studies were identified through literature review. The most commonly used endoscopic criteria for evaluation included the presence of a flat white scar and the absence of nodularity or telangiectasia. Information on the timing of endoscopic assessment in relation to neoadjuvant treatment protocols were also extracted from the studies. In most studies, the diagnostic accuracy for predicting pCR exceeded 0.8. The main limitations identified were the retrospective design of included studies included and a moderate risk of bias.

CONCLUSION

Endoscopy can be a key prognostic factor in predicting pCR to neoadjuvant treatment in rectal cancer despite significant limitations in currently available data.

Key Words: Endoscopic assessment; Clinical response; Rectal cancer; Neoadjuvant; Total neoadjuvant; Predictors of response; Pathologic complete response

Core Tip: Endoscopic assessment of rectal cancer response after neoadjuvant treatment can accurately predict pathologic complete response when specific criteria are applied. The combination of endoscopic features, including the absence of nodularity and ulceration, and the presence of a flat white scar with telangiectasia, achieves a diagnostic accuracy exceeding 80% for predicting pathologic complete response before curative intent surgery. These findings support the use of endoscopic evaluation as a reliable tool to guide safe omission of surgery in organ-preservation strategies.

INTRODUCTION

Rectal cancer presents a common and increasingly complex problem. A new treatment paradigm, total neoadjuvant treatment (TNT), has emerged. It consists of administering all planned chemotherapy and radiotherapy before surgery, either as induction chemotherapy followed by chemoradiotherapy, or as chemoradiotherapy followed by consolidation chemotherapy. This has allowed a significant number of patients to achieve clinical response, as assessed by endoscopy and magnetic resonance imaging (MRI), thus enabling an organ preservation (OP) strategy while avoiding surgery. At the same time, the complexity of rectal cancer treatment algorithms continues to increase. Unfortunately, a simplistic one-size-fits-all approach, still applied in many centers worldwide[1], has been associated with inferior outcomes.

Changes to the rectal cancer treatment paradigm have largely been shaped by the OPRA trial, which randomized patients with stage II/III rectal cancer into either chemoradiotherapy with consolidation chemotherapy or induction chemotherapy with chemoradiotherapy in the neoadjuvant setting. Patients with near-complete or complete responses were offered a watch-and-wait policy, while those who had an incomplete response at restaging or those who had local regrowth while on watchful wait, had surgery with curative intent. Both groups had similar 5-year disease-free-survival, highlighting the safety of the watch-and-wait approach for those patients with complete or near-complete clinical responses (CCRs)[2]. Despite the increased pathologic complete response (pCR) rates achieved with TNT vs traditional chemoradiotherapy regimens, the differences in recurrence or long-term survival gain have not been established[3], suggesting that more accurate tools for tumor response prediction are needed. Notably, pCR achieved through TNT has been associated with inferior survival compared to pCR achieved with conventional regimens of neoadjuvant chemoradiotherapy followed by adjuvant chemotherapy, according to a recent trial[4].

Endoscopy represents a cornerstone of tumor assessment after TNT, classifying patients into those with a CCR, near-complete response (NCR) and incomplete clinical response. This appears to have important prognostic implications for the probability of long-term OP according to results from a subgroup analysis of the OPRA trial[5]. More specifically, CCR was associated with 77% probability at 3 years to achieve OP, vs 40% for NCR. Certain endoscopic features have been described to have predictive value for rectal cancer management after neoadjuvant treatment[6], with variable rates of positive and negative predictive value. A recent consensus[7] tried to further elucidate the definition of NCR, stating that this term should only be used for the first 6 months from the end of neoadjuvant chemoradiotherapy, after which the tumor response should be classified as complete or incomplete. Among the modalities used to assess tumor response, the authors discuss all three [digital rectal examination, endoscopy and MRI, including T2 weighted (commonly known as T2W)-MRI and diffusion-weight imaging (commonly known as DWI) MRI]. However, modalities that evaluate the endoluminal component of the tumor, namely endoscopy and digital rectal examination, appear to be the most informative. This consensus paper proposed a three-tier system for classification of tumor response into a low likelihood, intermediate likelihood, or high likelihood of achieving a complete response, with tumor assessment at standardized time intervals. Endoscopic features such as flat-white scar or flat-red scar, regenerative tissue, small ulcer < 1 cm, or ulcer > 1 cm, along with favorable appearance on T2W and/or DWI MRI, hold significant positive prognostic value. Additionally important is the fact that biopsies are not needed to assess response, as results seem to create more confusion and diagnostic uncertainty due to high false-positive and false-negative rates.

The purpose of this review is to critically evaluate the existing literature in the context of the aforementioned consensus paper, with the aim of developing a unifying framework for current data and identifying key knowledge gaps. We considered it essential to extract from the literature detailed information on complete treatment plans, including multimodal regimens, and to integrate the timing and role of endoscopic assessments within these strategies to inform clinical decision-making. Our goal was to examine endoscopic response assessment after neoadjuvant therapy holistically, as part of an overall treatment roadmap, rather than as a stand-alone diagnostic tool.

MATERIALS AND METHODS

A literature review was conducted in PubMed-indexed journals with search terms “endoscopy AND rectal cancer AND neoadjuvant”, “endoscopic assessment AND rectal cancer AND response”, up to May 15, 2024. Two individual researchers (Seretis F and Panagaki A) performed the literature search. Potential conflicts for study inclusion were resolved after discussion. References from the extracted studies were manually searched for relevant studies. Inclusion criteria were: Studies written in English, case series of patients with rectal cancer undergoing neoadjuvant treatment with chemoradiation for which response was assessed with endoscopy. Studies from 2010 onwards were included to reflect current practice. Single case reports were excluded from the analysis.

RESULTS

An initial search with the aforementioned search terms yielded 1189 articles for review. After exclusion of duplicates (25 articles), 1164 articles were reviewed. A total of 1120 articles were excluded as non-relevant after review of abstracts. Forty-four articles were reviewed as full texts for eligibility. Two studies were excluded because they were single case reports, and eight studies were excluded because they reported on combined outcomes with other modalities. Another 19 of the 44 studies were excluded from final analysis as non-relevant. A manual search of references did not retrieve any additional relevant studies for review. Finally, 15 studies were included for final analysis in the systematic review. Search strategy and results are presented according to Preferred Reporting Items for Systematic Reviews and Meta-analyses (commonly known as PRISMA) 2020 guidelines for systematic reviews[8] in a flowchart, as supplementary data.

Studies were conducted from 2010 onwards to reflect current practice. The majority of studies included locally advanced rectal cancers, namely T3 or T4 tumors or node-positive disease receiving neoadjuvant chemotherapy. The majority of endoscopic assessments were performed at 6-8 weeks after the completion of neoadjuvant treatment. Five studies used deep learning algorithms for assessing endoscopic images. One study used magnifying endoscopy and one study used chromoendoscopy with narrow-band imaging, while the remaining studies used white-light endoscopy. The characteristics of the included studies are presented in Table 1[9-23]. All studies reported the type of neoadjuvant chemoradiotherapy administered to patients.

Table 1 Characteristics of the included studies.

Ref.	Year	Number of patients	Stage	Type of neoadjuvant treatment	Time of endoscopic assessment	Endoscopic modality used	Time of surgery	Definition of ECR	pCR	ECR	Accuracy of pCR prediction
Lim et al[15]	2010-2015	87	T3/T4 and/or N+	5 × 5 1.8 Gy + 5.4 Gy boost + ChT (5FU, capecitabine or FOLFOX)	1-2 weeks prior to surgery	N/A	6-8 weeks after CRT	A flat whitish or reddish scar ulcer, or a flat active/healing stage ulcer with regular edges surrounded by normal mucosa; disappearance of the neoplastic pit pattern without magnification; and disappearance of the neoplastic nodule or stenosis	21.7%	17.9%	88.7%
Chino et al[12]	2013-2015	79	T3/T4 and/or N+	Neoadjuvant chemoradiotherapy (oral fluoracil + 50.4 Gy), long course 59%, induction chemo (mFOLFOX + bevacizumab) + nCRT for cT4, cN2 or MRF (+) or positive lateral lymph nodes (41%)	48 days after CRT	WL-C, white light conventional endoscopy; ME, magnifying endoscopy under crystal violet staining	1-3 days after endoscopy	WL-C: Linear scar/flat scar, completely closed ulcer, no white moss, no protruded nodule. ME: Regenerated pits uniformly arranged, hyper-cellular pits. Wall extension by insufflation: Normal extension	21.5%	12.6%	85% for ME
Bengulescu et al[16]	2015-2017	43	Stage II/III rectal adenocarcinoma	Long course chemoradiotherapy (5FU + RT 504 Gy in 28 fractions)	N/A	N/A	6-8 weeks after nCRT	Endoscopy good grade (complete response or good response): Erythema; telangectasia; clean ulcer at the base; no elevation	N/A	67%	N/A
Wang et al[17]	2014-2021	214	LARC	Neoadjuvant chemoradiotherapy not otherwise specified	6–16 weeks after the completion of NT	Classic endoscopy (Olympus, Tokyo, Japan). ResNet-18 and DenseNet-121 deep learning algorithms		Good response; scarring (flat and white fibrosis with vasculopathy); erythema (erosion with peripheral erythematous mucosal changes)	23.8%	N/A	Manual method 75%, DenseNet-121: 72.6%, ResNet-18: 71.6%
Ishioka et al[14]	2012-2017	61	Stage II/II rectal cancer with poor features: MRF < 1 mm, cT4, positive lateral lymph nodes, mesorectal N2 disease and/or tumor requiring abdominoperineal resection	Neoadjuvant chemotherapy (folinic acid, fluorouracil, oxaliplatin + bevacizumab) followed by 50.4 Gy radiotherapy with concurrent S-1	Median 43 days after end of radiotherapy	White-light endoscopy and chromoendoscopy with indigo carmine dye, followed by magnifying NBI	Median 47 days after end of radiotherapy	Complete response (all 5): White light endoscopy: Ulcer completely closed and linear or flat shape of scar; no residual protruded nodules; preservation of rectal wall distention by insufflation. NBI: Regular or regenerated surface pattern; invisible or isolated vessel pattern (orderly network of micro vessels comprised of thin-caliber vessels)	31.1%	11.4%	70.5% with conventional white-light endoscopy, 75.4% white-light endoscopy + NBI
Kawai et al[18]	2007-2015	198 (186 radical surgery)	Low rectal cancer or T3/4 or N (+)	50.4 Gy with concurrent fluorouracil	3-8 weeks after end of CRT	Colonoscopy	6-8 weeks after end of CRT	Changes in the marginal tumor swelling, classified as almost no change, less than half according to pre-CRT measures, and almost flattened; changes in central ulceration, assessed according to whether reepithelialization of the ulcer was present; a cancer-negative biopsy result	12.8% in surgery group, 16.7% estimation in the watch-and-wait group	22.3% in radical surgery group, 50% in watchful waiting	Flattened marginal swelling 69.7%. Ulcer reepithelization 81.2%. Cancer negative biopsy 66.4%
Van Der Sande et al[19]	2012-2015	161 (87 surgery, 74 watch and wait)	cT1-2: 21.7%, cT3: 73.3%, cT4: 5%, cNo: 25.5%, cN1: 28.6%, cN2: 46%	Long course CRT or short course RT + interval	9 weeks after end of radiotherapy median (8-12 weeks)	White light imaging flexible sigmoidoscopy, 3 readers (2 surgeons, 1 gastroenterologist)	18 weeks interval between endoscopy and surgery	A flat scar; a small flat ulcer (< 1 cm); a large flat ulcer; ulcer with an irregular border; an adenomatous mass; tumorous mass	16%	42.8% (20.3% in surgery, 79.7% in watch and wait)	AUC 0.84, 0.80, 0.84 for complete response prediction for each of the 3 readers
Haak et al[13]	2012-2015	226	cT1-2: 22%, cT3: 71%, cT4: 7%, cNo: 24%, cN1: 28%, cN2: 48%	Long course CRT or short course RT + interval	10 weeks, median: 8-15 weeks	White light imaging flexible endoscopy (EPK-I video processor, Pentax Medical Netherlands, Uithoorn, the Netherlands) + deep learning	5 weeks, median: 2-10 weeks, between endoscopy and surgery	N/A	10.6%	48%	67%-75% for different convolutional neural network models including endoscopic images and clinical variables
Felder et al[20]	N/A survey	41+17 endoscopic pictures	N/A	Neoadjuvant treatment not otherwise specified	N/A	Two cross-sectional surveys, each containing endoscopic photos of rectal cancers treated with NT. The first survey assessed reproducibility of eight endoscopic criteria using 41 unique endoscopic photos. The second survey included endoscopic pairs of pre-neoadjuvant and post-neoadjuvant treatment photos of 17 patients	N/A	Flat and white scar; telangiectasias; absence of ulceration; absence of nodularity; small mucosal nodules or minor mucosal abnormality; superficial ulceration; mild persisting erythema of the scar; and visible tumor	N/A	N/A	89%
Williams et al[9]	2014-2020	263	cT1-2: 12.6%, cT3: 75.6%, cT4: 11.8%, cN (+): 63.7%	TNT (INCT-CRT or CRT-CNCT)	8 ± 4 weeks after TNT completion	Flexible sigmoidoscopy	N/A	Flat scar; telangiectasia; ulcer; nodularity; mucosal irregularity; mild erythema of the scar; visible tumor	N/A	42.5%	N/A
Williams et al[10]	2012-2020	288	Stage II/III	TNT (INCT-CRT or CRT-CNCT)	47 days median time from end of TNT	White-light flexible endoscopy with an Olympus scope (model CF-Q160S). Convolutional Neural Network using ResNet-50	N/A	Flat white scar; telangiectasias; no ulceration; no nodularity	N/A	N/A	AUC 0.99 for training set, AUC 0.98 for main test, AUC 0.92 for local regrowth
Chen et al[21]	2013-2021	1000	T3-4 and / or N (+)	Long course radiotherapy 50 Gy or short course radiotherapy 25 Gy. Concurrent chemotherapy oral/intravenous 5FU or combined with oxaliplatin/irinotecan	6-8 weeks after NT	White-light endoscopy with endoscopy-based Deep Convolutional Neural Network with a ResNeSt-50 variant	6-12 weeks after NT	N/A	21.9%	N/A	94.21% for training set, 92.13% for validation set, 87.14% for independent set
Sohn et al[22]	2004-2013	425	cT3-4 and/or N (+)	nCRT (45 Gy + 5.4 Gy boost + 5FU/Leucovorin or capecitabine or tegafur/uracil)	Immediately before surgery	White-light endoscopy	6-8 weeks after nCRT	E-GR: Scarring (the flat and white mucosa with fibrotic changes); telangiectasia (scarring surrounded by small blood vessels); erythema (scarring or erosion with peripheral erythematous mucosal changes)	10.8%	11.1%	N/A
Han et al[23]	2004-2013	481	cT3-4	nCRT (45 Gy + 5.4 Gy boost + 5FU/Leucovorin or capecitabine or capecitabine + irinotecan or tegafur/uracil)	6-8 weeks after nCRT	White-light endoscopy	6-8 weeks after nCRT	E-GR: Scarring (the flat and white mucosa with fibrotic changes); telangiectasia (scarring surrounded by small blood vessels); erythema (scarring or erosion with peripheral erythematous mucosal changes)	11%	N/A	Pathologic good response: ≤ ypT1, positive predictive value of 0.65, negative predictive value of 0.885
Thompson et al[11]	2012-2017	109	LARC	TNT [ChT (FOLFOX or CAPEOX) + long-course chemoradiotherapy with concurrent 5FU or capecitabine]	6 weeks after TNT and subsequent follow up visits (3 total selected including first visit)	White-light endoscopy with convolutional neural network VGG-19	Watch and wait	N/A	N/A	N/A	Training set AUROC 0.83, testing set AUROC 0.83

AUC: Area under the curve; AUROC: Area under the receiver operator curve; CRT: Chemoradiotherapy; ECR: Endoscopic complete response; E-GR: Endoscopic good response; LARC: Locally advanced rectal cancer; ME: Magnifying endoscopy; MRF: Magnetic resonance fingerprinting; NBI: Narrow-band imaging; nCRT: Neoadjuvant chemoradiotherapy; NT: Neoadjuvant therapy; N/A: Not applicable; RT: Radiotherapy; TNT: Total neoadjuvant treatment; 5FU: 5-fluorouracil.

When assessing response to neoadjuvant treatment, pCR as assessed on pathologic examination of specimens after curative intent surgery was set as the gold standard for complete response to neoadjuvant treatment. The goal of the current study was to determine the diagnostic accuracy of endoscopic complete response (ECR) to predict the subsequent “gold standard” of pCR.

In three studies, a total neoadjuvant approach was used for rectal cancer management. In studies reported by Williams et al[9,10], patients underwent either chemotherapy with consolidation chemoradiotherapy or long-course chemoradiotherapy followed by chemotherapy as neoadjuvant treatment. In the first report[9], endoscopic criteria used for assessment were reported and the time of endoscopic assessment after TNT was 8 ± 4 weeks. However, the authors did not report any diagnostic accuracy data. In the second report[10], the authors applied convolutional neural network-based algorithms for assessing response at a median time of 47 days after TNT completion. In a study by Thompson et al[11], patients underwent chemotherapy followed by long-course chemoradiotherapy as neoadjuvant treatment. The endoscopic response to treatment was assessed at 6 weeks after completion of TNT and then at regular hospital visits as part of a watch-and-wait protocol. The authors in this study used artificial intelligence to create a deep learning-based model to analyze endoscopic images and reported an area under the receiver operator curve (commonly referred to as AUROC) values of 0.83.

The other 12 studies included here reported on patients receiving traditional neoadjuvant chemoradiotherapy regimens. We noticed a significant variation between studies in the time frames for endoscopic assessment. Two studies[12,13] reported on the performance of endoscopic assessment in relation to the timing of surgery, while two studies[14,15] did not report on the specific timing of endoscopic assessment. The remaining eight studies reported on the timing of endoscopic assessment in relation to the completion of neoadjuvant treatment. Significant variations between time frames and regimens were also noted.

Endoscopic assessment of rectal cancer response after neoadjuvant treatment appears to have a high diagnostic accuracy in predicting pCR. Eleven studies[9,10,12,14,16-21,23] reported on the criteria used to classify endoscopic response of patients. Telangiectasia, flat/white scar, and ulcer with clean base were features consistently used for assessment of complete endoscopic response. The pCR rates ranged from 10.8% to 31.1%. The value of endoscopic assessment in prediction of pCR was expressed as either diagnostic accuracy or area under the curve with different values reported. When extracting the endoscopic criteria used to describe endoscopic CCR from relevant studies, the following criteria were most frequently used: (1) Flat, white scar; (2) Absence of nodularity; (3) Absence of ulceration; and (4) Telangectasia.

We also reviewed endoscopic modalities used to investigate treatment response. Bengulescu et al[16] compared chromoendoscopy to classic white-light imaging and reported better outcomes with chromoendoscopy (75.4% accuracy for pCR prediction for chromoendoscopy vs 70.5% for classic white-light endoscopy). Three studies reported diagnostic accuracy in pCR prediction above the 80% threshold, which is widely considered to be the cutoff value for clinically significant diagnostic accuracy. The diagnostic discriminatory power of some studies was reported as AUROC, with the value consistently above 0.8.

Due to the heterogeneity across studies and data, including the different protocols for endoscopic assessment, it was not feasible to perform meta-analysis. Moreover, due to the variability of endoscopic criteria used to assess CCR among the different studies, it was inappropriate to cumulatively estimate the number of subjects from different studies that meet the definition of complete response. The rates of pCR (if reported) as well as the “ECR” (if reported) are included in the descriptive table. Some studies reported their outcomes using different metrics for the accuracy of endoscopic criteria to predict pCR on subsequent surgical resection, such as area under the curve. Because direct comparisons could not be made, a statistical analysis was not feasible.

Risk of bias assessment

Regarding risk of bias, we assessed the methodological quality studies using Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 guidelines[24]. Most studies were deemed as high risk of bias for patient selection. Studies that applied deep learning algorithms or chromoendoscopy were deemed as unclear risk for bias due to applicability concerns for the index test. Failure to report accurate endoscopic criteria for ECR assessment or failure to report pCR rates (considered as reference standard), also significantly downgraded the quality of studies for the purposes of our research question, namely whether ECR after neoadjuvant treatment can predict pCR in rectal cancer patients. Results are presented in Table 2[9-23]. All studies identified in our search were retrospective in nature, and thus suffer from potential selection bias. Another form of bias identified in some studies was the lack of comparison to a reference standard, namely pCR for our study. Five studies[10,11,15,18,22] used deep learning algorithms based on convolutional neural networks applied on white-light endoscopy images to enhance diagnostic accuracy. Although all five reported excellent outcomes in terms of diagnostic accuracy, they suffered from applicability concerns and lack of reference standard bias.

Table 2 Quality assessment of studies using Quality Assessment of Diagnostic Accuracy Studies-2.

Ref.	Risk of bias				Applicability concerns
	Patient selection	Index test	Reference standard	Flow and timing	Patient selection	Index test	Reference standard
Lim et al[15]	High risk	Low risk	Low risk	Low risk	Low risk	Low risk	Low risk
Chino et al[12]	High risk	Unclear risk	Unclear risk	Low risk	High risk	High risk	Unclear risk
Bengulescu et al[16]	High risk	Low risk	Low risk	Low risk	Low risk	Low risk	Low risk
Wang et al[17]	High risk	Low risk	Unclear risk	Low risk	Low risk	Low risk	Unclear risk
Ishioka et al[14]	High risk	Low risk	Low risk	Low risk	Low risk	Low risk	Low risk
Kawai et al[18]	High risk	Unclear risk	Low risk	Low risk	Low risk	High risk	Low risk
Van Der Sande et al[19]	High risk	Low risk	Low risk	Low risk	Low risk	Low risk	Low risk
Haak et al[13]	High risk	Low risk	Low risk	Low risk	Low risk	Low risk	Low risk
Felder et al[20]	High risk	Unclear risk	Low risk	High risk	Low risk	Unclear risk	Low risk
Williams et al[9]	High risk	Low risk	Unclear risk	High risk	Low risk	Unclear risk	Unclear risk
Williams et al[10]	High risk	Unclear risk	Unclear risk	High risk	Low risk	Unclear risk	Unclear risk
Chen et al[21]	High risk	Unclear risk	Unclear risk	Low risk	Low risk	High risk	Unclear risk
Sohn et al[22]	High risk	Unclear risk	Low risk	Low risk	Low risk	Low risk	Low risk
Han et al[23]	High risk	Low risk	Unclear risk	Low risk	Low risk	Low risk	Unclear risk
Thompson et al[11]	High risk	High risk	High risk	High risk	Low risk	High risk	High risk

DISCUSSION

The evolving landscape of treatment for locally advanced rectal cancer presents unique challenges in patient management. Trials such as RAPIDO and PRODIGE 23[25] pioneered the concept of TNT, demonstrating that it can achieve favorable oncologic outcomes along with high rates of pCR rates and opportunities for OP. The higher rates of pCRs achieved with TNT have enabled the development of nonoperative watch-and-wait strategies. These involve close clinical, endoscopic and radiographic surveillance at predefined intervals to allow timely intervention in cases of incomplete tumor response or local regrowth.

A recently published secondary analysis of the OPRA trial[26] suggested that CCR is associated with improved rates of OP, disease-free survival, distant-metastasis free survival and overall survival. The landmark report from Habr-Gama et al[27] from 2010 defined clinical and endoscopic findings of CCR. Findings from this study are widely used as the roadmap when reassessing patients after neoadjuvant treatment, with residual superficial ulceration, irregularity or nodule regarded as equivalent to residual disease. Importantly, the authors concluded that diagnostic biopsies to stage tumor response after treatment were not valuable. In contrast, achieving pCR after neoadjuvant treatment is associated with improved long-term oncologic outcomes, whereas the presence of any residual cancer, even limited to the mucosal (in situ) level, is linked to poorer outcomes[28]. A recent pathological analysis of non-pCR tumors sought to elucidate why accurate staging of tumor response requires diagnostic modalities with submillimeter-level resolution[29]. Assessment of the region adjacent to the tumor, the so-called invasive front, has been reported as a key finding in distinguishing responders from non-responders[30]. Therefore, it is essential to identify clinical findings as surrogates of pCR. Clinical restaging has largely relied on digital rectal examination, MRI and endoscopy, but each of these modalities is limited in its accuracy[31].

Endoscopy demonstrates a greater ability to accurately identify complete response compared to MRI[32]. Ko et al[33] reported that combining the two modalities increases diagnostic accuracy, despite the fact that they utilized endoscopy with biopsies for their endoscopic assessment of clinical response. In another report from Cho et al[34], neither MRI nor endoscopy had adequate diagnostic accuracy for accurate pCR prediction; thus, the authors suggested a combined approach. We conducted this systematic literature review to estimate the diagnostic accuracy of endoscopic assessment, with an emphasis on specific diagnostic criteria used to define a CCR, as well as the exact time frame in which to perform the endoscopic assessment after neoadjuvant treatment. An increasing body of literature on the combined use of diagnostic modalities has introduced confusion regarding the true diagnostic yield of each modality when applied independently. Combining diagnostic modalities without standardization of technique and criteria can confound results, thus preventing the dissemination of good practices across the medical community. Artificial intelligence and agnostic interpretation of imaging modalities are appealing in terms of diagnostic accuracy, but this effect may be overestimated because the results were not compared with results from studies in which strict criteria were used.

The existing studies suffer from different biases, as demonstrated in our analysis of study quality using the QUADAS-2 guidelines. A recent systematic review[35] on diagnostic performance of MRI and endoscopy for assessing complete response in rectal cancer after neoadjuvant chemoradiotherapy also concluded that the heterogeneity of studies prevented them from having great clinical value. However, with strategies such as TNT, as well as the growing need to counsel patients on watch-and-wait and OP strategies now being implemented worldwide, it is essential to examine both the strengths and limitations of our current evidence base. Concerns for proper patient selection for OP safely are further supported by recent long-term follow up results from the RAPIDO trial, suggesting that TNT is associated with an increased risk of local recurrence compared to classic treatment regimen[36].

This review revealed heterogeneity in the endoscopic criteria used for prediction and in the specific time frame of assessment. Considering the new definitions[7] and 6 months as the limit to characterize tumor response as either complete or incomplete, such variations in time of assessment become critically important. Another limitation of endoscopic assessment of tumor response after neoadjuvant therapy is its dependence on operator experience, as evaluation, even when specific endoscopic features are systematically assessed, is subject to interobserver variability. No studies have reported on this matter. Furthermore, none have specified whether endoscopic assessments were performed by gastroenterologists or surgeons. Because colorectal surgeons are often responsible for follow-up after neoadjuvant therapy, particularly in the TNT era of watch-and-wait strategies, yet gastroenterologists perform the majority of endoscopic procedures in most centers worldwide, clarifying this distinction is of paramount importance. Therefore, future research should focus on standardization of endoscopic processes, enhancement of training and compliance with time frames dictated by multidisciplinary rectal cancer management guidelines. Artificial intelligence represents a powerful tool when integrated within an appropriate clinical context and applied without deviation from standardized assessment protocols. In our separate review of studies utilizing artificial intelligence, we found that the application of deep learning algorithms to endoscopic images failed to consistently predict pCR.

We have provided exemplary endoscopic images from patients undergoing watchful waiting after neoadjuvant chemoradiation for rectal cancer (Figure 1). We have also systematically used the absence of nodularity, telangiectasia, and a completely flat white scar with no irregularities as endoscopic predictors of complete response.

Figure 1 Endoscopic image. A: Initial endoscopic image before treatments; B: Absence of central ulceration; C: Telangiectasia, absence of ulceration; D: Telangiectasia (scarring surrounded by small blood vessels); E: Flat white scar, absence of nodularity, absence of ulceration.

CONCLUSION

Endoscopic assessment of rectal cancer response after neoadjuvant (chemo)radiotherapy, when performed using standardized criteria for defining CCR, can predict pCR with good diagnostic accuracy. The endoscopic features most consistently associated with tumor response include a flat scar, absence of nodularity or ulceration, and the presence of telangiectasias. The use of deep learning algorithms applied to endoscopic images may further enhance diagnostic performance. However, existing studies are limited by various forms of bias and exhibit significant heterogeneity in study protocols, criteria, and response definitions.