3  Results

3.1  Patient characteristics
The patient characteristics and operative data are shown in Table 1. The mean age of all 110 patients was 59.3 ± 11.9 years. The mean distance between the tumor and the dentate line was 5.6 ± 2.8 cm. Five of the 110 patients had a very low rectal tumor, which involved the dentate line and was classified as cT2N+ on preoperative evaluation. These patients received preoperative CRT to increase the possibility of a sphincter-preserving surgery. The mean time interval to surgery was 5.6 ± 2.8 weeks between completion of CRT and primary surgery. A sphincter-preserving surgery was performed in 102 (92.7%) patients. The remaining 8 patients underwent abdominoperineal resection for a very low tethered rectal tumor, which did not respond sufficiently to preoperative CRT. Ninety-nine (90%) patients underwent minimally invasive surgery (MIS), including laparoscopic and robotic surgery. There was no conversion to open surgery in patients who underwent MIS.
Table 1  Patients characteristics and operative data.

3.2  Postoperative pathologic outcomes
At pathologic examination, ypT0, ypT0-1, and ypT0-T2 were observed in 30 (27.3%), 39 (35.5%), and 65 (59.1%) cases, respectively. Lymph node metastasis was observed in 34 cases (30.9%). The proportion of patients with a positive CRM was 2.7% in all patients. Of patients who underwent preoperative CRT followed by TME, 29 patients (26.4%) achieved a pCR (Table 2). T downstage was defined as the downstage of the tumor from cT2 to ypT0-1 or cT3 to ypT0-2 or from cT4 to ypT0-3. In all, 68 patients (61.8%, 68/110) were found to have a T downstage after preoperative CRT, and 57 patients (57.1%, 60/105) had less than ypT2 stage in patients with cT3-4 stage.
Table 2  Postoperative pathologic outcomes.

3.3  Endoscopy, mrTRG, and combination modality for predicting the good-response group using ROC curves
When the good response group was defined as ypT0 (Fig. 3A), the values of AUC showed significant differences between the 3 modalities (P < .001). Endoscopy showed significantly higher AUC (0.818) than did mrTRG and combination modality in differentiating good response from poor response (P < .001). Among the 3 modalities, endoscopy showed the highest value and mrTRG showed the lowest value. When the good response group was defined as ypT0-1 (Fig. 3B), the 3 modalities showed no significant differences in differentiating good tumor response from poor response (P = .117). However, the AUC (0.717) of the combination modality was significantly higher than that of mrTRG (P = .011). When the good response group was defined as ypT0-2 (Fig. 3C), the 3 modalities showed significant differences in differentiating good tumor response from poor response (P = .011). The AUC (0.697) of the combination modality was the highest among the methods, and it was significantly higher than that of mrTRG (P = .026).
Figure 3  (A) Comparison of receiver operating characteristic (ROC) curves according to endoscopy, magnetic resonance tumor regression grade (mrTRG), and the combination modality (endoscopy added to mrTRG) for predicting ypT0. (B) Comparison of ROC curves predicting ypT0-1. (C) Comparison of ROC curves predicting ypT0-2.

3.4  Diagnostic performance of each modality in the evaluation of tumor response to chemoradiation using GEE
Tables 3 to 5 show the diagnostic performance of between the 3 modalities and ypT stages of 110 tumors. Among a total of 110 patients, 34 patients showed good tumor response by both the endoscopy and combination modality (Fig. 4). Of these 34 patients, 23 were confirmed as true absence of tumor by histopathologic analysis, which meant that the PPV of endoscopy for ypT0 was 67.65% (95% confidence interval [CI]: 51.92–83.37) by using the GEE. On the contrary, 11 of 30 patients with good tumor response by MR tumor response were confirmed as ypT0 by histopathologic analysis (PPV 36.67%, 95% CI: 19.42–53.91). Both endoscopy and combination modality showed the highest values in sensitivity (79.31%), PPV (67.65%), NPV (92.11%), and accuracy (84.55%) and theses values showed significantly higher diagnostic performances than those of mrTRG for the prediction of ypT0 (Table 3). When evaluating ypT0-1 by using 3 modalities, there was no significant difference at distinguishing ypT0–1 from ypT2–4 (PPV of endoscopy: 52.94% (95% CI: 26.16–69.72), PPV of mrTRG: 43.33% (95% CI: 25.60–61.07), PPV of combination modality: 46.67% (95% CI: 32.09–61.24), respectively, P = .330). However, combination modality showed significantly higher sensitivity than that of mrTRG (70.00% vs 43.33%, P = .001) (Table 4). When comparing 3 modalities for the prediction of ypT0-2, all modalities showed relatively high diagnostic performance in differentiating ypT0-2 from ypT3-4 (PPV of endoscopy: 85.29% [95% CI: 73.39–97.20], PPV of mrTRG: 80.00% [95% CI: 65.69–94.31], PPV of combination modality: 82.22% [95% CI: 71.05–93.39], respectively, P = .688). In addition, sensitivity, NPV, and accuracy showed significantly higher values in combination modality than those of mrTRG (sensitivity: 56.92% vs 36.92%, P ≤ .001, NPV: 56.92% vs 48.75%, P = .003, accuracy: 67.27% vs 57.27%, P = .003, respectively) (Table 5).
Figure 4  Response assessment with endoscopy; red scar with telangiectasia (black arrow) and white scar (white arrow) shows typical sign of clinical complete response after chemoradiation therapy.
Table 3  Diagnostic performance power predicting ypT0.
Table 4  Diagnostic performance power predicting ypT0-1.
Table 5  Diagnostic performance power predicting ypT0-2.