The results of this retrospective study comprehensively showed the high clinical value of PDC grade when selecting resection strategies for T1 CRC patients through multiple rounds of validation. PDC grade is an independent predictor for LNM and has the best predictive value. It was also included in the predictive model, providing a foundation for the use of PDC grade as a reference for surgery, simplifying risk stratification models, and refining prior treatment strategies. PDC grade could also be regarded as a predictor of oncological outcomes, and it was found to be an independent risk factor for tumor recurrence, distant metastasis, and patient death. We also considered its correlation with other indexes and high interobserver agreement. In view of its high clinical value and availability, we recommend highlighting its applicability in the clinic and adding it to routine histopathological reports, especially for CRC patients. This work also broadened the methodological approach for similar research; for the chosen index, the fitness of the variables (AIC value) was explored, and backward regression was properly used. The results of this work also indicate that multidisciplinary team (MDT) discussion should be recommended.
Our results revealed a correlation between a higher PDC grade and deeper submucosal invasion. Combined with the findings reported in previous studies, ECRCs with deeper submucosal invasion tend to be incompletely resected under endoscopy and even require adjuvant therapy after rection [32]. Our assumption that CRCs with a high PDC grade need to be surgically resected was indirectly confirmed. An association between PDC grade and TB grade was also observed. Previous studies all showed evidence of epithelial-mesenchymal transition (EMT) [33], which was even speculated to represent different stages of tumor growth [34]. However, their fundamental differences and similarities require further studied. The definitions of the number of cells are worth reconsidering, perhaps even trying to combine them to simplify the risk stratification model. We also found that PDC grade was related to PNI, LVI and mucinous composition. PNI, LVI and mucinous composition might promote the formation of PDCs, which provides a treatment opportunity. Moreover, Barresi V et, al. pointed out that PDCs might be relate to the biomolecular profiles of CRCs, and gene mutations, such as KRAS mutations, were more common in CRCs with high PDC grades [35]. Therefore, the PDC grade might serve as a reference for clinicians to administer targeted therapies; however, the mechanisms need to be further clarified in all CRC patients.
The depth of submucosal invasion is the basis of T1 CRC treatments in the Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines [36, 37]. Nevertheless, because of the additional time needed to measure depth and the absence/ambiguity of the muscularis mucosae in endoscopic specimens, this indicator has not been fully implemented in Europe. However, the stratification of submucosal invasion is more difficult than the stratification of the depth of submucosal invasion. There are also difficulties in promoting TBs; specific staining can be useful to help distinguish TBs, but excess staining might cause confusion with other tissue cells. The TB grade showed predictive paradoxes in our study and previous studies [19]. We found that TBs were associated with LNM in the univariate analysis (crude OR Bd 2 = 0.088, crude OR Bd 3 = 0.013); however, an association was not found in the multivariate analysis. Ueno et, al. [19] previously reported a similar phenomenon in their study on stage II-III CRC patients. We therefore believe that the difficulty in assessing TBs is the main reason, although the bias caused by our small sample size cannot be ignored. Identification of PNI and LVI in H&E-stained specimens is also difficult [38]. As a considerable prognostic predictor of CRCs [39, 40], high interobserver variability is the main limitation of tumor grade [28, 41, 42].
The PDC grade performed well in terms of predictive power. PDCs are cell clusters/nests that are larger and recognizable without specific staining [29], which saves time in identification, increases the accuracy of the report, and improves the reproducibility of related laboratory studies. Furthermore, high interobserver agreement was found for PDCs [19, 26–28], making them easier to use in the clinic and reducing training costs. In addition, PDCs was found to be an independent predictive index for LNM after excluding the confounder of the depth of submucosal invasion, which suggests that innovative techniques and devices for endoscopic full-thickness resection may be a promising alternative to major surgery when the integrity of endoscopic resection is questionable and the risk of LNM is low [43, 44]. A high PDC grade has also been previously found to be related to occult LNM [27]. Moreover, PDC grade fit the LNM best both alone and in the model. Therefore, selecting treatment methods mainly depending on the PDC grade might be feasible. Furthermore, although PDC is a pathological indicator, when referencing it to guide diagnosis and treatment strategies for T1 CRC patients, a multidisciplinary team (MDT) discussion, including gastrointestinal surgeons, oncologists, radiologists, and so on, is recommended to make comprehensive decisions. Patient age, tumor location (colon or rectum), and common comorbidities all need to be seriously considered.
Obviously, to obtain firm conclusion that PDC grade is a good reference for selecting management strategies, investigating its predictive value relative to LNM is not enough. Some research has found that local LNM is a precursor for distant metastases [15] and that 0.3%-4.5% of patients develop metastases after lymph node dissection [14, 45]. Performing lymph node dissection did not improve the oncological outcomes in these cases. Therefore, we further investigated the prognostic value of the PDC grade in detail. Our results showed that PDC grade is an independent risk factor for a poor prognosis in T1 CRC patients; moreover, patients with different PDC grades had different prognoses. Considering the mean survival time (OS, 206 months; DFS, 212 months), the average age when initially diagnosed (70 years old), and the cutoff time commonly used for oncological outcomes in the clinic (5 years), we initially speculated that G1 CRC patients do not require adjuvant therapy after tumor resections. For patients with G2 CRC, the mean survival time dropped sharply at approximately 5 years; thus, clinicians must be vigilant and follow them closely, providing adjuvant radiochemotherapy when necessary. G3 CRC patients had the shortest survival time; consequently, they needed the most extensive treatment and the closest postoperative monitoring, even though some studies found no significant relationship between monitoring intensity and postoperative recurrence [46]. Further research is needed to clarify the proper posttreatment monitoring frequency.
The procedure used to explore predictive indicators in our work is well established and provides a reference for developing new indexes. We applied the AIC value to compare fit and prove its clinical utility, expanding the use of the fit comparative method of indicators and models. Backward regression was also properly and innovatively used. We await the exploration and validation of image recognition tools based on deep learning to further avoid biases caused by histological evaluation.
This work has unique clinical value. However, we acknowledge that, as a single-center retrospective study, the limitations of the study type cannot be ignored. Moreover, the small sample size and the low absolute number of patients with oncological outcomes (LNM and oncological endpoints) could have caused bias. Nevertheless, we faithfully recorded the patients’ information in detail. Moreover, all slides were reevaluated according to the definitions in the available consensus to improve the quality of the pathological records. In addition, we used archived slides, and tissue processing was not standardized. However, this defect exactly reflects real-time practice and increases the generalizability of our results to a certain extent. Furthermore, our participants were all from a registered population in Qingdao and the surrounding area; therefore, extrapolating our results requires caution. The interstitium of participants might promote standardized sample collection and minimize possible residual confounding caused by differences in subjects’ genetic backgrounds.
To further confirm our conclusions, multicenter, prospective, and large sample studies are needed. Moreover, the predictive value of PDC grade in biopsy samples also requires further investigation. This might represent a clinical window that allows clinicians to stratify patients by their LNM risk and prognosis at initial biopsy. Then, the subsequent management of patients may be improved. Furthermore, more complete stratification models based on methods such as machine learning are also expected in the future.