Novel Pathologic Factors for Risk Stratification of Gastric “Indefinite for Dysplasia” Lesions

Methods In total, 123 IND cases with final diagnoses of cancer (29.3%), high-grade dysplasia (6.5%), low-grade dysplasia (11.4%), and nonneoplasm (52.8%) were randomly divided into test set (n = 27) and validation set (n = 96). By the image analysis, size, pleomorphism, hyperchromasia, irregularity of nuclei, and ratios of structural atypia area (SAA) to total IND area were measured in the test set. Using the validation set, consensus meetings were held for the evaluation of pathologic factors that predict the final diagnosis. Results By image analysis, the only ratio of SAA to total IND area was associated with the final diagnosis (p < 0.001). In the consensus meeting for validation, the nuclear factors, except loss of nuclear polarity (p = 0.004–0.026), could not predict the final diagnosis. Conversely, most structural factors could predict the final diagnosis. In particular, SAA > 25% was the most powerful predictive factor. We proposed criteria of risk stratification by using SAA > 25%, loss of surface maturation (LOSM), and loss of nuclear polarity (LONP) (Malignancy rate; Category 0: SAA ≤ 25% without LOSM and LONP; 0%, Category 1: SAA ≤ 25% with any of LOSM or LONP; 15.2%–16.7%, Category 2: SAA > 25% without LOSM and LONP; 44.4%–50.0%, Category 3: SAA > 25% with any of LOSM or LONP 54.5%–55.6%). Conclusions Structural atypia was more helpful than nuclear atypia and SAA > 25% was the most powerful predictor for the diagnosis of INDs of the stomach. We propose shortening the follow-up period to six months for Category 1, endoscopic resection for Category 2 and 3, postresection follow-up periods of one year for Category 2, and six months for Category 3.


Introduction
Pathologic evaluations are the gold standard for the diagnosis of gastric cancer; however, it is not always possible to distinguish between malignancy and benign conditions when architectural distortion or nuclear atypia is present [1][2][3]. On histomorphologic study, atypical regenerative changes are occasionally indistinguishable from malignancy [4,5]. The international consensus meeting held in Vienna suggested to designating this indefinite histol-ogy as a distinct category, i.e., indefinite for dysplasia (IND) [6,7].
With regard to the management of the IND, no definite criteria for selecting follow-up methods have been established. Although current guidelines advise follow-up biopsy for IND areas, a few endoscopists insist that endoscopic resection (ER) be used as a follow-up method because malignancy rates in IND areas are high (22.6%-75.0%) [8][9][10][11][12].
In a given stomach tissue sample, reactive changes of the stomach mucosa comprise cytomorphologic changes such as enlarged nuclei, pleomorphism, hyperchromasia, or vesicular changes of nuclei, and these reactive changes might not be associated with neoplastic progress. In contrast, architectural features such as loss of surface maturation (LOSM), margination, glandular cribriform patterns, glandular branching/budding, glandular arrangement, and glandular crowding are relatively more associated with tumorous conditions [4,13].
In addition to these histomorphologic interpretations, which are qualitative and subjective, quantification of each pathologic factor may facilitate subclassification of IND areas when the cytomorphologic factors do not clearly distinguish neoplasms from reactive changes [14].
With regard to IND follow-up methods, ER has been recently suggested [8][9][10][11][12]. Only a few studies have investigated IND to attempt establishing the criteria of IND follow-up methods and have mostly focused on the clinical factors suggesting malignancy [8][9][10][11][12] or been studied in Barrett's esophagus [1,2]. The aims of this study were to investigate and quantify pathologic factors for predicting carcinoma in IND through the use of image analysis and thereby facilitate the subclassification of IND conditions and determine the proper follow-up methods.  [15]. Patients lost to follow-up after the first biopsy were excluded (St. Mary's Hospital (n = 70) and St. Vincent's Hospital (n = 59)). A total of 123 cases (St. Mary's Hospital (n = 67) and St. Vincent's Hospital (n = 56)) were randomly divided into test (n = 27) and validation (n = 96) sets. The final diagnosis of nonneoplasm was established if three consecutive samples obtained from the same IND area demonstrated no neoplasms or if no detectable lesions were found in endoscopic or radiologic examination for more than 1 year. The study was approved by our institutional review board (XC17RAMI0005K).

Image Analysis.
We evaluated 20 representative nuclei per each sample of the test set (n = 27) in IND areas using Image J (National Institutes of Health, Bethesda, United States). The average areas of the nuclei were measured to evaluate mean nuclear areas for nuclear enlargement, standard deviation of nuclear areas for pleomorphism, brightness ratio of tumor to lymphocytes for hyperchromasia, and Feret diameters for nuclear irregularity. The area ratios of structural atypia to IND were calculated to quantify structural atypia. The structural atypia area (SAA) was defined as areas with ≥1 of the following: variable glandular size, irregular glandular arrangement, glandular branching/budding, and glandular cribriform patterns ( Figure 1) [16]. The area of IND was defined as an area with any well-known structural anomalies (LOSM, margination, glandular cribriform patterns, glandular branching/budding, glandular arrangement, and glandular crowding) or nuclear atypia (loss of nuclear polarity (LONP), nuclear pseudostratification in more than half, nuclear pleomorphism, nuclear hyperchromasia, and prominent nucleoli) [16]. These measurements were discussed among three pathologists (K Yim, JH Shin, and J Yoo). Receiver operating characteristic (ROC) curves were plotted to determine the cut-off values in the SAA to IND that could predict neoplasm, above high-grade dysplasia (HGD), or carcinoma. The sum of the sensitivity and specificity was obtained, and cut-off values were determined using the maximum sum.

Consensus Meeting to Establish Criteria and Validation.
We discussed and confirmed well-known pathologic factors [16] to differentiate between nonneoplasm and neoplasm, below LGD and above HGD, and noncarcinoma and carcinoma in the test set (n = 27) and to include a few other factors (neutrophils, ulcer, and intestinal metaplasia) that may affect interpretation. Each of these factors was considered as positive when definitely present in the cases.
These factors were subsequently analyzed and determined in the validation set (n = 96) by two pathologists (K Yim and JH Shin). Whether the SAA ratio above the cut-off value determined by image analysis could predict the final diagnosis was also evaluated.

Data and Statistical
Analysis. Detailed information, including age and sex, size, location, and shape of the lesions and other potential risk factors, was obtained retrospectively from medical records. The surface nodularity or spontaneous bleeding could not be obtained because it was not recorded in the medical records. Continuous data were compared using independent t-tests, and categorical variables were tested using Pearson's χ 2 method or Fisher's exact tests. The criterion for statistical significance was p < 0:05. ROC curves were plotted, and the maximum sum of sensitivity and specificity was defined as the cut-off point. Cohen's Kappa was calculated to compare the results of the two pathologists. For multivariate analysis, ROC curves were obtained for each pathologic parameter with crude factors, and differences in area under curve (dAUC) to crude factors were used to compare the predictive power of each pathologic factor. Three effectible factors (neutrophils, ulcer, and intestinal metaplasia) were used as crude factors. Analyses were performed using SPSS (version 18.0; SPSS, Chicago, IL, USA).

Final Diagnoses of INDs.
The enrolled patients (n = 123) underwent follow-up biopsy (n = 105, 85.4%), ER (n = 7, 5.7%), or gastrectomy (n = 11, 8.9%). The reason for gastrectomy after a diagnosis of IND in six patients was the presence of coexisting lesions in distant sites of the stomach (The final diagnoses were carcinomas in four patients and nonneoplasms in two). Radiological examination revealed overt malignant findings in the four patients with a final diagnosis of carcinoma. One of those eleven patients exhibited marked fibrosis and ulcer; therefore, obtaining an adequate sample by biopsy was difficult, and, furthermore, ER was not available.

Gastroenterology Research and Practice
The patient himself wanted surgery since he did not want to worry about cancer (The final diagnosis was a carcinoma). After complete workup, the final pathologic diagnoses included cancer (n = 36, 29.3%), HGD (n = 8, 6.5%), LGD (n = 14, 11.4%), and nonneoplasms (n = 65, 52.8%). Among the cases with the final diagnosis of carcinoma, 19 (52.8%) were determined by biopsies, 8 (22.2%) by ER, and 9 (25.0%) by gastrectomy. Although 67 cases showed nonneoplasm at second follow-up biopsies, after a series of follow-up biopsies, two cases showed persistent INDs (final diagnoses were carcinomas), one showed LGD, one showed HGD, and one showed cancer. The final diagnoses of persistent INDs were one nonneoplasm, one LGD, and three carcinomas ( Figure 2).  1). There were more persistent INDs or negative in the second biopsy at the same lesion in those with delayed diagnosis (2 weeks: p = 0:598, 2 months: p = 0:257, 6 months: p = 0:021, 1 year: p = 0:013). However, there were no significant differences in the final pathologic stage (Table 1). Furthermore, there were eight cases with repeated IND or negative in the second biopsy, one was nonneoplasm (12.5%), three were

Validation Consensus
Meeting. The interpretation results by each pathologist are summarized in Table 4. The factors that both pathologists interpreted as significant predictor for carcinoma were LOSM (p = 0:001, p = 0:012), glandular

Discussion
In the present study, 36/123 cases (29.3%) of IND lesions at initial biopsy were finally determined to be gastric cancers, which is similar to the rates observed in previous studies (22.6%-75.0%) [8][9][10][11][12]. Although currently, follow-up biopsy is usually recommended, recently, and ER has been proposed as a follow-up method because of high malignancy rates and diagnostic difficulty of IND [8][9][10][11][12]. Similarly, the diagnostic difficulty of IND in biopsy specimen was observed in the present study. The predictive power of biopsy was relatively low (cancers in biopsy: 52.8%, biopsy as follow-up methods: 85.3%), and persistent IND or false-negative cases (6.5%) were shown.
However, the use of ER as a follow-up method after IND diagnosis remains controversial because most tumors with IND are well-differentiated and slow growing [12,17,18]. Similar to previous findings, the present study showed that diagnostic delay did not affect the final pathological stage [12], and most tumors, even with falsenegative diagnoses or persistent IND, were welldifferentiated tumors that remained in the early gastric cancer stage (Table 2).
There were some attempts to establish guidelines for follow-up methods of INDs. Most previous studies about IND were focused on the clinical features suggesting malignancy [8][9][10][11][12], diagnosis agreements of pathologists [1,2,12], INDs of Barrett's esophagus [1,2], or diagnostic progress by additional immunohistochemistry [1,2]. Reportedly, ≥10 mm lesion size, depressed lesion, spontaneous bleeding, and surface nodularity on endoscopy were independent risk factors for gastric cancer [10][11][12]. However, other studies had shown that endoscopic lesion size, gross appearance, location, number of biopsies, and Helicobacter pylori infection were not predictive of gastric cancer [8,9]. In the present study, endoscopic lesion size, gross appearance, location, and number of biopsies were not associated with the final diagnosis. Predicting neoplasm Predicting above HGD Predicting carcinoma Figure 3: Sum of the sensitivity and specificity for predicting final diagnoses according to area ratio with structural atypia to total indefinite for dysplasia by plotting receiver operating characteristic curves. Cut-off values were defined as 25.3% for neoplasm, 25.3% for above highgrade dysplasia, and 26.1% for carcinoma.

Gastroenterology Research and Practice
To the best of our knowledge, pathological analysis for IND in gastric specimen had not been attempted until Kwon et al. [12] reported that distorted architecture, i.e., increased glandular crowding and irregular glandular arrangements, was strongly associated with carcinoma. The present study evaluated and quantified each pathologic factor that could further classify INDs in gastric biopsy specimen. As a result, we established a novel SAA > 25% as a quantitative pathologic factor in this study. This novel factor integrated variable glandular size, irregular glandular arrangement, glandular branching/budding, and glandular cribriform by measuring areas with ≥1 of the above, and it was the best predictor for final diagnosis (dAUC = 0:135 -0:182 for predicting carcinoma, Figure 4 and Table 5) and showed high reproducibility (Cohen ' s Kappa coefficient = 0:956). Therefore, the quantification of pathologic factors, while at a basal level, could be helpful for an accurate diagnosis. SAA > 25% could not integrate LOSM, margination, and all factors regarding nuclear atypia because these factors were difficult to quantify by area measurement.
Nuclear atypia could not predict the final diagnosis, except for LONP (p = 0:004 -0:026, Cohen ' s Kappa = 0:869). Reactive changes, including inflammation or ulcers, may lead to enlarged nuclei and pleomorphism of nuclei nuclear hyperchromasia or vesicular changes [4,13]. In contrast, most pathologic factors associated with structural atypia showed good predictability for the final diagnosis that strongly correlated between the two pathologists (Cohen ' s Kappa = 0:682 -0:855, Table 4). This confirmed that changes in the structure were relatively preserved in the reactive settings, such as inflammations or ulcers [4,13].

Conclusions
In conclusion, we propose criteria for the risk stratification of INDs, as summarized in Table 6. The criteria were based on the following: (1) Only factors with statistical significance by both pathologists were included. (2) The predictive pathologic factors could be grouped into SAA > 25%, LOSM, and LONP. (3) SAA > 25% was the best predictor for diagnosis.
(4) The ability to detect cancer was focused than premalignant lesions. Because most tumors in IND areas are well-  Table 5.