Accuracy of Flexible Sigmoidoscopy and MRI in Restaging Rectal Cancer after Neoadjuvant Therapy: A Secondary Analysis of the OPRA Randomized Clinical Trial

Abstract

OBJECTIVE:

To measure the accuracy of flexible sigmoidoscopy (FS) and MRI in identifying a true response (TR) to total neoadjuvant therapy (TNT) in patients with locally advanced rectal cancer.

BACKGROUND:

Tumor response on restaging FS and MRI determines patient eligibility for watch-and-wait (WW). Prospective studies have not yet evaluated the accuracy of these testing modalities.

METHODS:

This was a secondary analysis of the OPRA trial. Patients underwent restaging assessment with FS and MRI 8±4 weeks after completing TNT. Those with a clinical complete (cCR) or near-complete (nCR) response were recommended for WW. Patients with an incomplete response were recommended for total mesorectal excision. A TR, defined as a sustained cCR for ≥2 years on WW or a pathologic complete response on surgical specimen, was the reference standard for the diagnostic performance analysis.

RESULTS:

Three hundred and four patients completed TNT; 265 (82%) had sufficient follow-up for inclusion in the diagnostic performance analysis. Fifty-two percent had a TR. Accuracy, sensitivity and specificity were 60%, 95% and 23% for MRI and 65%, 96% and 33% for FS. Combining FS and MRI improved accuracy to 67%. The combined modalities reliably predicted 5-year TME-free survival (log-rank p < .001), with patients whose tumor response was classified as nCR by both assessments most likely to experience local regrowth.

CONCLUSIONS

Endoscopic and radiologic assessment of rectal tumor response to neoadjuvant therapy is accurate in two-thirds of patients. FS is more likely than MRI to detect a TR and should take precedence when exams are discordant.

Mini Abstract

Prospective studies have not yet evaluated the diagnostic performance of restaging flexible sigmoidoscopy and MRI in locally advanced rectal cancer patients considered for watch-and-wait surveillance. This secondary analysis of the OPRA trial demonstrates that the combined modalities correctly identify a true response to total neoadjuvant therapy in two-thirds of patients.

INTRODUCTION

The safety and long-term success of watch-and-wait (WW) surveillance for locally advanced rectal cancer treated with total neoadjuvant therapy (TNT) relies on the ability to distinguish between patients with a true response (TR) and those with residual tumor. However, current restaging techniques are imperfect and cannot always correctly make this distinction.^1–5 Approximately 10% of patients taken to surgery for suspected residual disease will have a pathologic complete response while 30% of those placed on WW surveillance will develop local tumor regrowth. ^6–11

Restaging after neoadjuvant therapy is based on digital rectal exam (DRE), MRI, and flexible sigmoidoscopy (FS) performed several weeks after completion of neoadjuvant treatment. Tactile information from the DRE is highly subjective, correlates poorly with pathologic staging, and provides no value for high-rectal tumors.¹² Rectal MRI provides multiplanar images of the bowel wall, mesorectum, and locoregional lymph nodes. However, T2-weighted sequences lack resolution for differentiating fibrosis from small post-treatment tumors.^13,14 Attempts to improve diagnostic accuracy by adding diffusion-weighted imaging, defining size cutoffs for suspicious lymph nodes, or quantifying tumor regression using the magnetic resonance tumor regression grade (mrTRG) have yielded variable results.^13,15–19 The endoscopic criteria for a clinical complete response (cCR) first proposed in 2010 are overly restrictive, ²⁰ as many patients without residual disease still exhibit minor mucosal abnormalities months after completing neoadjuvant treatment.^21–24 Furthermore, FS assesses only the tumor surface and cannot provide information on the deeper layers of the bowel wall or mesorectum. To date, no international consensus exists for a standardized grading system of radiologic or endoscopic tumor response.

The reported accuracy of each restaging modality (DRE, MRI, and FS) varies widely.^14,25–27 Previous studies have used pathological response as the reference standard in patients undergoing mandatory total mesorectal excision (TME) at a fixed-interval several weeks after completion of neoadjuvant therapy. This approach likely underestimates the rate of TR, as tumors did not have sufficient time to fully regress after neoadjuvant treatment. Other studies in patients offered nonoperative management have been limited by differences in the definition of a true response and inconsistent WW eligibility criteria. Finally, very few studies have investigated the accuracy of FS and MRI in the same patient cohorts.

The Organ Preservation in Rectal Adenocarcinoma (OPRA) trial prospectively applied standardized criteria to assign patients a clinical response grade for each restaging modality and used the response grades to select patients for WW or TME.¹¹ This secondary analysis of the OPRA trial aimed to evaluate (i) the accuracy of FS and MRI, both individually and in combination, in identifying a TR to neoadjuvant therapy, and (ii) the ability of the OPRA trial’s novel tumor response grading system to predict long-term oncologic18 i outcomes.

METHODS

Trial Design

The OPRA trial randomized patients with stage II or III rectal adenocarcinoma to receive TNT in the form of either systemic chemotherapy followed by chemoradiation or chemoradiation followed by systemic chemotherapy. Patients were enrolled between April 2014 and March 2020. Institutional review boards at all participating institutions approved the trial protocol (NCT02008656). Eligibility criteria, trial design, restaging, and surveillance protocols have been described previously.¹¹

At 8 ± 4 weeks after completion of TNT, patients underwent a restaging assessment consisting of DRE, FS, and MRI. The patient’s primary surgeon and radiologist independently assigned clinical response grades for the restaging clinical exam (FS and DRE) and MRI, respectively. Response was classified as clinical complete (cCR), near-complete (nCR), or incomplete (iCR) based on the endoscopic, DRE, and MRI features listed in eTable 1 in the Supplement. For tumors with features across two or three response tiers, the lowest response category was assigned. In cases where the FS and MRI clinical response grades differed, precedence was given to the FS. Patients with a cCR or a nCR proceeded to WW, while those with an iCR were recommended TME.

Of note, surgeons assigned a single response grade for the clinical assessment, using features from both the DRE and the FS. We will refer to this assessment as the endoscopic exam (FS), given that DRE data was missing for 29 (10%) of the 277 patients, and endoscopic criteria determined the clinical response grade for 255 (92%) patients.

Analysis of Diagnostic Performance

Accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were determined for restaging FS and MRI separately and in combination. The reference standard was TR, defined as either a sustained cCR on WW surveillance for ≥2 years after completion of TNT or a pathologic complete response evident in the surgical specimen. For the purposes of this study, high-grade dysplasia or carcinoma in situ in the surgical specimen was considered residual disease. WW patients with less than 2 years of follow-up were excluded, as they did not meet the criteria for a sustained cCR.

Sensitivity corresponded to the ability to detect a TR, while specificity corresponded to the ability to identify residual disease. PPV was defined as the probability that a testing modality correctly detected a TR, while NPV was the probability that a testing modality correctly identified residual tumor. P-values comparing PPV and NPV for each test were calculated using relative predictive values.²⁸ Positive and negative likelihood ratios were used to determine positive and negative post-test probabilities, respectively. Contingency tables were prepared, with either a cCR or a nCR considered indicative of TR.

Oncologic Outcomes

All patients with restaging data were included in an intention-to-treat analysis of disease-free survival, distant metastasis-free survival, local recurrence-free survival, overall survival, and TME-free survival. Patients who entered WW surveillance were assessed for probability of local regrowth. Outcomes were assessed in relation to clinical response grades based on FS and MRI at restaging. Based on the results of both tests, patients were grouped as cCR/cCR (both FS and MRI suggested cCR), cCR/nCR (FS suggested cCR and MRI nCR, or vice versa), nCR/nCR, (both FS and MRI suggested nCR), and any-iCR (if either FS or MRI suggested iCR). Outcomes of interest were measured from the date of the restaging MRI or FS, whichever occurred last. Events for these survival analyses were defined as previously described.¹¹ Patients without events were censored at the last date of follow-up.

Statistical Analysis

Patient demographic and baseline characteristics were summarized using frequencies and percentages for categorical variables and medians with interquartile ranges for continuous variables. Categorical variables were compared between patient groups using the Fisher exact test; continuous variables were compared using the Kruskal-Wallis rank sum test. Survival curves were estimated with Kaplan-Meier curves and compared across patient groups using the log-rank test. Statistical analysis was performed using R software version 4.3.3 (R Foundation for Statistical Computing). P values were two-sided and were considered statistically significant if <.05.

RESULTS

Of the 324 patients randomized in the OPRA trial, 304 (94%) underwent restaging. Twenty-seven of the 304 patients were excluded due to missing restaging MRI and/or FS data (n = 12), mucinous tumors that could not be accurately restaged by MRI (n = 8), or treatment-related toxicities (n = 7) (Figure 1). For the remaining 277 patients, median intervals from the end of TNT to restaging by MRI and FS were 6.9 and 7.7 weeks, respectively. The median interval from either restaging assessment to last follow-up was 4.0 years.

Figure 1.

Consort diagram

The distribution of clinical response grades determined by FS and MRI at restaging was as follows: cCR/cCR, 73 patients (26%); cCR/nCR, 78 patients (28%); nCR/nCR, 60 patients (22%); Any-iCR, 66 patients (24%). These groups did not differ significantly in demographics, baseline tumor characteristics, or interval from the end of TNT to either FS or MRI (Table 1). A higher proportion of patients in the cCR/cCR group had received consolidation TNT compared to induction TNT, with this difference approaching statistical significance (p = .052).

Table 1.

Patient Demographics and Treatment Characteristics

	cCR/cCR n = 73	cCR/nCR n = 78	nCR/nCR n = 60	Any iCR n = 66	p-value
Age (IQR)	56 (49,66)	61 (51, 68)	58 (49, 68)	57 (48, 65)	0.5
Female gender	29 (40)	28 (36)	16 (27)	25 (38)	0.4
Treatment arm					0.05
INCT-CRT	26 (36)	41 (53)	31 (52)	38 (58)
CRT-CNCT	47 (64)	37 (47)	29 (48)	28 (42)
Baseline tumor distance from AV (IQR) a	4.50 (3.3, 7.0)	4.30 (3.0, 6.9)	4.00 (2.8, 5.3)	4.50 (3.0, 6.0)	0.3
cT stage					0.4
1/2	12 (16)	6 (7.7)	7 (12)	4 (6.1)
3	53 (73)	65 (83)	45 (75)	52 (79)
4	8 (11)	7 (9.0)	8 (13)	10 (15)
cN positive	46 (63)	55 (71)	49 (82)	50 (76)	0.1
Days from end of TNT to restaging FS (IQR)	53 (40, 63)	54 (42, 62)	60 (45, 70)	54 (40, 68)	0.2
Days from end of TNT to restaging MRI (IQR)	46 (34, 57)	46 (39, 57)	52 (38, 63)	50 (35, 63)	0.3

Data are presented as numbers (percentages) of patients, unless otherwise specified. Percentages may not add up to 100 because of rounding. Abbreviations: cCR, clinical complete response; nCR, clinical near-complete response; iCR, clinical incomplete response; IQR, interquartile range; INCT-CRT, induction chemotherapy followed by chemoradiation; CRT-CNCT, chemoradiation followed by consolidation chemotherapy; AV, anal verge; cN = baseline node-positive disease; TNT, total neoadjuvant therapy. FS= flexible sigmoidoscopy

^aMissing data for 1 patient in the cCR/nCR group.

Diagnostic Performance of FS and MRI

Twelve (4%) of the 277 patients were excluded from the diagnostic performance analysis due to insufficient follow-up on WW surveillance (Figure 1). For the remaining 265 patients, FS and MRI clinical response grades were concordant in 144 patients (54%) and discordant in 121 patients (46%) (Table 2). A TR was subsequently identified in 137 (52%) patients (n=126 sustained cCR and n=11 pathologic complete response).

Table 2.

Distribution of Clinical Response Grades

		Flexible Sigmoidoscopy
		cCR	nCR	iCR
MRI	cCR	73	35	8
	nCR	43	60	22
	iCR	4	13	19

Abbreviations: cCR, clinical complete response; nCR, clinical near-complete response; iCR, clinical incomplete response. Color legend: Green, cCR/cCR; yellow, cCR/nCR; orange, nCR/nCR; gray, iCR by either assessment.

Diagnostic performance was analyzed using broad criteria (cCR or nCR) to define a TR (Table 3). This approach resembled the design of the OPRA trial, which allowed patients with either a cCR or a nCR to enter WW surveillance. Restaging MRI had an accuracy, sensitivity, specificity, PPV and NPV for detecting a TR of 60%, 95%, 23%, 57% and 81%, respectively. Endoscopy outperformed MRI across all parameters, with an accuracy, sensitivity, specificity, PPV and NPV of 65%, 96%, 33%, 60% and 88%, respectively. FS was more likely to correctly identify a TR than MRI (p=0.03). Combining modalities yielded a higher PPV compared to each individual test (FS; p=0.045 and MRI; p<0.001). There were no significant differences between modalities in correctly identifying residual tumor. A secondary diagnostic performance analysis using narrow criteria (cCR only) to define a TR can be found in eTable 2 in the Supplement.

Table 3.

Diagnostic performance of MRI and flexible sigmoidoscopy (FS) in identifying true response in patients with either a cCR or a nCR

	True Response (TR)			Accuracy	Sensitivity	Specificity	PPV	NPV	(+) Likelihood Ratio	(-) Likelihood Ratio	(+) post-test probability	(-) post-test probability
		Yes	No
MRI				60%	95%	23%	57%	81%	1.23	0.23	57%	20%
cCR or nCR	Yes	130	99
	No	7	29
FS				65%	96%	33%	60%	88%	1.42	0.13	60%	12%
cCR or nCR	Yes	131	86
	No	6	42
Combined				67%	91%	41%	62%	82%	1.56	0.21	62%	18%
cCR/cCR, cCR/nCR or nCR/nCR	Yes	125	75
	No	12	53

Abbreviations: cCR, clinical complete response; nCR, near-complete clinical response; FS, flexible sigmoidoscopy; PPV, positive predictive value; NPV, negative predictive value.

Local Regrowth and Rectum Preservation

WW was offered to 196 (71%) of the 277 patients. Findings concerning for local regrowth were identified in 69 (35%) of the 196 patients. All but five (7%) cases of regrowth occurred within 2 years of restaging. At five-years, patients with a nCR/nCR at restaging had the highest rate of local regrowth (59.3% [95% CI, 43.1–71%]), followed by patients with a cCR/nCR (35.3% [95% CI, 22.2–46.2%]) and patients with cCR/cCR (20.3% [95% CI, 10.2–29.2%]) (log-rank P < .001); a converse pattern was observed for rectum preservation (TME-free survival) (Figure 2). Of the patients with suspected local regrowth who had surgery, two in the nCR/nCR group (2/30; 7%) and one in the cCR/cCR group (1/14; 7%) had a pathologic complete response.

Figure 2.

Kaplan-Meier curves by combined clinical response grade for (A) TME-Free Survival and (B) Probability of Local Regrowth in WW Patients

Oncologic Outcomes

Kaplan-Meier survival estimates in relation to restaging clinical response grades are shown in Figure 3. At 5 years, 89.6% (95% CI, 82.5–97.3%) of patients with a cCR/cCR remained disease-free, followed by 72.1% (95% CI, 61.7–84.2%) with a cCR/nCR, 68.9% (95% CI, 57.8–82.2%) with a nCR/nCR, and 49.8% (95%CI, 38–65.4%) with Any-iCR (log-rank P < .001).

Figure 3:

Kaplan-Meier curves by combined clinical response grade for (A) disease-free survival (B) distant metastasis free survival (C) local recurrence free survival and (D) overall survival

DISCUSSION

In this secondary analysis of the OPRA trial, we demonstrate that post-treatment FS is more likely to detect a TR compared to MRI, suggesting that surgeons should give precedence to the endoscopic exam when patients have discordant findings. Combining FS and MRI resulted in the highest probability of detecting a TR, but did not significantly improve the likelihood of identifying residual disease compared to each individual test. The combination of clinical response grades assigned by FS and MRI predicted TME-free survival, with patients classified as a nCR/nCR at particularly high risk for local regrowth. These findings have important implications for counseling patients on the likelihood of achieving long-term rectum preservation and may help physicians select candidates for WW surveillance more effectively.

Restaging rectal adenocarcinoma after neoadjuvant therapy presents several challenges. First, tumor regression is time-dependent and continues to evolve for many months after completion of neoadjuvant therapy.^23,29 Restricting response evaluation to a single time-point makes it difficult to distinguish between ongoing tumor resolution and persistent disease. A treating physician must choose to either limit WW eligibility to patients with a cCR at restaging or allow those with minor post-treatment abnormalities consistent with a nCR to pursue nonoperative management. The first strategy risks unnecessarily removing the rectum in patients with ongoing tumor regression that may evolve into a complete response with longer follow-up, while the latter may delay a curative resection in patients with unrecognized residual disease. Additionally, no single testing modality approximates the gold standard of pathologic complete response for confirming complete eradication of tumor. The reported diagnostic performance of restaging FS and MRI varies widely in the literature, with accuracy, sensitivity, and specificity for detecting residual tumor ranging from 80 to 89%, 26 to 55%, and 80 to 97%, respectively, for endoscopy^4,22,26,30 and from 62 to 79%, 22 to 65%, and 63 to 91%, respectively, for MRI. ^4,13,27,31 Most studies evaluating diagnostic performance rely on retrospective data and use pathologic complete response as the reference standard. They are often weakened by the use of nonstandard clinical response definitions, inconsistent WW eligibility criteria and an inability to account for the time-dependent nature of tumor regression.

This secondary analysis of the OPRA trial is the first study to use prospective data to evaluate the diagnostic performance of restaging FS and MRI in a large cohort of patients offered WW. The OPRA trial introduced a novel three-tiered grading schema for restaging MRI, which achieved an accuracy of 60%. While these results are consistent with the literature, the clinical response grading system did not improve radiologist ability to detect a TR over previously reported methods such as the mrTRG score.^27,31 These findings underscore that current multimodal imaging techniques lack sufficient resolution to reliably detect malignant lymph nodes or to distinguish post-treatment fibrosis from residual tumor.^13,14,32 FS outperformed MRI in identifying a TR, with an accuracy of 65%. With nearly half of patients assigned discordant clinical response grades by FS and MRI, our analysis suggests that the endoscopic exam should take precedence given its superior performance. In practice, these modalities are complementary, and treatment decisions based on discrepant clinical response grades must be individualized. While FS may better identify endoluminal disease, it cannot replace MRI in detecting worrisome features in the deeper layers of the bowel wall, mesorectum or regional lymph nodes.

Although surgeons integrate data from restaging FS and MRI to determine patient eligibility for WW, very little information exists regarding their combined diagnostic accuracy. Maas et al conducted the only prospective study on this topic in a cohort of 50 patients, 9 of whom pursued nonoperative management.³ The authors reported that the combined exams correctly predicted TR in 98% of cases when both identified a complete response.³ In contrast, the combination of FS and MRI in our study yielded a positive-posttest probability of 62%. Although Maas et al reported a better diagnostic accuracy for FS and MRI, the results may be difficult to replicate in real-world clinical practice. In addition to using narrow criteria to define a complete response, the study had a small sample size with limited follow-up and relied on four expert clinicians from the same institution to classify response. The strengths of the OPRA trial include its large sample size, pre-defined parameters for assigning clinical response grades, and a standardized interval between end of TNT and the restaging assessment. Clinical response grades were determined by physicians across 18 institutions with various levels of experience, which provides a more realistic approximation of diagnostic accuracy than relying on expert opinion alone. Novel approaches to image interpretation, such as machine learning and radiomics, may offer opportunities to further improve the diagnostic performance of the restaging assessment.^33–35

The combined clinical response grades were strongly associated with TME-free survival and long-term oncologic outcomes. These findings build on previous work by our group demonstrating the predictive value of the three-tiered grading schema when applied individually to either FS or MRI.^36,37 Patients who entered WW with a nCR/nCR at restaging developed local regrowth at a rate almost twice as high as those with a cCR/nCR and almost three times as high as patients with a cCR/cCR. Despite the higher rate of local regrowth in patients with a nCR/nCR, our data show that nearly 40% had a sustained cCR and ultimately preserved the rectum. It remains unclear whether the remaining 60% of patients with a nCR/nCR who ultimately require surgery are subjected to an unnecessary risk of tumor progression while on WW. Recent work by Fernandez et al has suggested that local regrowth may be associated with the development of distant metastatic disease.³⁸ While these findings raise concerns about the oncologic safety of a WW strategy in patients at high-risk for local regrowth, this study has several shortcomings. It is retrospective, compares two dissimilar groups, uses patient data that pre-dates the widespread use of TNT, and employs statistical methods that cannot clearly delineate the timing of metastatic disease relative to appearance of local regrowth.³⁹ In the absence of prospective data from randomized control trials, we continue to advocate for the inclusion of patients with a nCR in WW protocols. Surgeons should counsel patients with a cCR/nCR or nCR/nCR at restaging about their elevated risk of local regrowth. Patients who make an informed decision to pursue nonoperative management should be followed at shorter, 6- to 8-week intervals with both FS and MRI. As recommended in the OPRA trial, patients with ongoing evidence of tumor regression may continue surveillance while those without improvement in response and those with local regrowth should proceed immediately to surgery without delay.¹¹

This study has several limitations. First, tumor assessment by DRE was incorporated into the FS clinical response grade and was not evaluated separately. Second, the order in which FS and MRI were performed at restaging was not standardized. Radiologists and surgeons were not blinded to the results of the other restaging exam, and this may have introduced bias into the grading system. Finally, since tumor response grades for each patient were assigned by a single surgeon and a single radiologist at one of 18 participating institutions, we could not determine whether accuracy was associated with clinician experience level. Notwithstanding these limitations, the data reflect diagnostic performance in real-world clinical practice.

CONCLUSIONS

Restaging FS outperforms MRI in identifying a TR after neoadjuvant therapy. Combining tests narrowly improves detection of a TR compared to FS alone. However, overall accuracy remains lower than expected, with the combined tests incorrectly classifying tumor response in approximately one-third of patients. These findings highlight the limitations inherent to current restaging modalities. While FS better identifies residual tumor from subtle mucosal changes, only MRI can detect disease restricted to the deeper layers of the bowel wall or to the locoregional lymph nodes. However, the overall accuracy of MRI is constrained by the inability of current multimodal imaging techniques to reliably differentiate post-treatment fibrosis from residual tumor.

The combined three-tiered clinical response grading system introduced by the OPRA trial has important implications for rectum preservation and survival. The concordance between FS and MRI also has prognostic value, with patients classified as a nCR by both modalities at higher risk for local regrowth compared to other groups. While the results of this study provide important information for risk-stratifying patients interested in nonoperative management, physicians must also consider the limitations of restaging FS- and MRI-based tumor assessments when counselling patients about potential treatment options.