Gastric hydrodistension CT versus CT without gastric distension in preoperative TN staging of gastric carcinoma: analysis of single-center cancer registry

Accurate staging of gastric cancer is essential for the selection and optimization of therapy. Hydrodistension of the stomach is recommended to improve the accuracy of preoperative staging with contrast-enhanced multidetector computed tomography (MDCT). This study compares the performance of contrast-enhanced gastric water distension versus a nondistension MDCT protocol for T and N staging and serosal invasion in comparison to surgical histopathology. After propensity score matching, 86 patients in each group were included for analysis. The overall accuracy of distension versus nondistension group in T staging was 45% (95% CI 35–56) and 55% (95% CI 44–65), respectively (p = 0.29). There was no difference in the sensitivity and specificity in individual T staging and assessment of serosal invasion (all p > 0.41). Individual stage concordance with pathology was not significantly different (all p > 0.41). The overall accuracy of N staging was the same for distension and nondistension groups (51% [95% CI 40–62]). The majority of N0 staging (78–81%) were correctly staged, whereas N3 staging cases (63–68%) were predominantly understaged. In summary, there was no significant difference in the diagnostic performance of individual TN staging and assessment of serosal invasion using MDCT with or without gastric water distension.

Propensity score matching. Table 1 tabulates the characteristics of the cohort study at baseline and after propensity score matching. There were class imbalances in the age, sex, origin of CT, pT staging, and pN staging at baseline. The baseline cohort was mostly male (61%, 314/516), with cancers that occurred in the antrum (53.5%, 276/516) and had CT performed as inpatients (74%, 381/516). Early gastric cancer (T1 staging) accounted for 33% (168/516). In the postmatched cohort, both Distension and Nondistension groups consisted of 86 patients each, with 50% (86/172) of cases occurring in the antrum and 32% (55/172) of cases of T1 staging, similar to the baseline cohort. Table 1. Patient characteristics before and after propensity score matching. SMD standardized mean difference. SMD of greater than 0.1 is significant. Age given in median with interquartile range in parenthesis. Data presented as counts and percentages in parenthesis.  www.nature.com/scientificreports/ Diagnostic performance. Interrater reliability for the independent readers is shown in Table 2. In both Distension and Nondistension groups, the ICC was good to excellent for T and N staging, indicating that readers had a high degree of agreement. Table 3 tabulates the T and N staging performance metrics of Distension and Nondistension groups derived from consensus reading. The overall accuracy of Distension versus Nondistension groups in T staging was 45% (95% CI 35-56) and 55% (95% CI 44-65), respectively (p = 0.29), and there was no difference in the overall accuracy for N staging in either group (51% [95% CI  in both). There were no significant differences in sensitivity and specificity of individual T and N staging between Distension and Nondistension group. Figure 1 shows clinical case examples of T1 to T4b staging cancers in Distention and Nondistention groups. Table 4 compares the diagnostic performance for assessing serosal invasion in Distension and Nondistension groups. The sensitivity was greater than 90%, and the positive predictive value greater than 80% in both groups. However, no significant difference in diagnostic performance was demonstrated between Distension and Nondistension groups.
The majority of N0 staging cases were correctly staged, whereas the N3 staging cases were predominantly understaged. This is also reflected in the higher sensitivity of N0 staging in both groups (Distension 78% versus Nondistension 81%, p = 1).
The European MAGIC trial on perioperative chemotherapy found that 8.3% of patients who underwent gastrectomy alone had pT1 disease 7 . In the present study, 24% (5/21) of the Distension group and 42% (5/12) of the Nondistension group overstaged pT1 as greater than equal to T3 staging (p = 0.48). In a Korean series with modern 64-dectector CT and where 60% (76/127) of lesions were pT1 (due to an extensive country-wide screening program), the mean stage concordance for pT1, pT2, pT3 and pT4a was 97%, 66%, 70%, and 80%, respectively 26 . As in most series, the higher stage concordance for pT1 may be because the absence of lesions on MDCT in a known gastric cancer patient is often diagnostic for pT1 staging 15,27,28 .
There are a few limitations in this study. First, the water-distension protocol involved supine-only imaging irrespective of the location of the tumor. In this study, antrum and body tumors accounted for more than three www.nature.com/scientificreports/ quarters of all tumors. Second, virtual gastroscopy or tailored multiplanar reconstruction relative to the primary lesion was not performed as these techniques required additional reading and time at the workstation; this is not part of the standard workflow in our institution. Third, pT1 accounted for one-third of cases, which reflects the real-world prevalence in our population. While EUS staging may be preferable and more sensitive for T1 lesions, MDCT still forms an important part of preoperative staging. Finally, the independent readers were aware of the location of the tumor identified by endoscopy, this may have affected the diagnostic performance. However, in routine clinical practice, final clinical staging is performed by members of the gastric cancer multidisciplinary tumor board with access to endoscopic images for review.
In conclusion, the overall accuracy, individual T and N staging sensitivity and specificity of MDCT with or without gastric distension with water were not significantly different. The sensitivity and specificity in the detection of serosal invasion on MDCT were not affected with or without gastric distension with water.

Material and methods
Study design and study cohort. This was a single institution retrospective review of the gastric cancer registry from 2010 to 2015, which encompasses the 7th edition of the American Joint Committee on Cancer (AJCC)/Union for International Cancer Control (UICC) staging system. Patient demographics, origin of CT image (inpatient, outpatient, outside referral, and emergency department), site of tumor (antrum, body, fundus or diffuse [greater than two contiguous sites]), clinical (based on the initial CT stage from multidisciplinary tumor board) and pathological TNM stage, histopathological subtype based on WHO and Lauren's classification (intestinal, diffuse, and mixed), and the CT protocol were evaluated.
The registry includes all patients who were diagnosed or referred to our hospital for the treatment of gastric cancer. The following exclusion criteria were applied: clinical metastasis at presentation, pathological metastasis at surgery, no surgical histopathology, nonadenocarcinoma histopathology, CT performed at the referral hospital, and CT performed at the emergency department. CT images from referral hospitals and from the emergency department were excluded because of inconsistent image quality and/or reconstruction parameters. The study design flowchart is shown in Fig. 2.
We sought to compare the diagnostic performance of two CT protocols (described in detail below) in the evaluation of T and N staging to that of surgically-resected specimens. This study was approved by the Chang Gung Medical Foundation Institutional Review Board and a waiver of informed consent was approved as the research involves no more than minimal risk to subjects. All study methods were carried out in accordance with relevant guidelines and regulations. The STARD guidelines were used to ensure the reporting of this diagnostic accuracy study 39 . CT protocols. Owing to the reimbursement policy of the national health care system, patients whose CT was performed at the time of presentation had restricted access to secondary dedicated staging CT after the diagnosis of gastric cancer. In this group of patients, gastric distension was not performed prior to CT acquisition. For the purpose of this study, we defined two types of protocols: a water-distension dual-phase protocol for staging of newly diagnosed gastric cancer (hereafter referred to as "Distension"), and a nonwater distended ("Nondistension") protocol.
Per the Distension protocol, the patient ingest up to 1000 mL of purified water prior to CT examination to distend the stomach. Scan acquisition was supine-only covering the whole stomach during arterial phase and the abdomen to the pelvis during the venous phase. Axial, coronal and sagittal arterial phase images were routinely reconstructed into 3 mm slice thickness and interval; axial venous phase images were reconstructed into 5 mm slice thickness and interval.
Protocols designated as Nondistension included contrast-enhanced single and dual-phase CT of the abdomen and pelvis without prior distension of the stomach. For this protocol, axial and coronal images from venous phase images were reconstructed into 5 mm thickness and interval. All CT examinations were performed with 16-or 64-detector CT scanners.
Propensity score matching. The study cohort consisted of imbalanced groups (Distension versus Nondistension 402:114). Propensity-score matching was performed logistic regression model with the following covariates: age, sex, pT staging, pN staging, Lauren's classification, tumor site, CT origin, year of diagnosis. The standardized mean difference was used to evaluate matching between groups, with a value of 0.1 or higher indicating an imbalance. Propensity score matching was performed using R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria) with the MatchIt package (version 3.0.2) 40 using the nearest neighbor matching method and a caliper distance of 0.2 without replacement.

Secondary image and tumor stage evaluation by independent readers.
Over the 5-year study period, different and/or additional radiologists joined the gastric cancer multidisciplinary tumor board and rendered the clinical TNM stage. To mitigate bias and evaluate the reliability and reproducibility of image readings by different radiologists, additional image review by 4 radiologists who were members of the gastric cancer multidisciplinary tumor board (C.M.C, W.H.C, Y.S.L, Y.C.L with 11, 7, 7, and 3 years of experience, respectively) was performed. The clinical TN stage established during the tumor board conference was included as a 5th independent reader. All readers performed the image interpretation separately and were independently blinded to the pathological TN staging. Images from the same CT protocol were batched and then randomized for reading on two separate occasions. Readers noted the clinical T (1, 2, 3, 4a and 4b), and clinical N (0, 1, 2, 3) staging based on the AJCC/UICC 7th edition. In the independent reader study, interrater reliability with intraclass correlation coefficient (ICC) estimates and their 95% confident intervals were calculated using R with the irr package (version 0.84.1) based on a mean-rating (k = 5), absolute-agreement, 2-way random effects model. Intraclass correlation coefficient values less than 0.5 were indicative of poor reliability, values between 0.5 and 0.75 indicated moderate reliability, values between 0.75 and 0.9 indicated good reliability, and values greater than 0.9 indicate excellent reliability 41 . Consensus TN staging was calculated from the mode (majority consensus) of all 5 reader observations. Diagnostic performance (sensitivity, specificity, positive predictive value, negative predictive value) with 95% confidence intervals was calculated from the consensus TN staging. Clinical versus pathological TN staging concordance (under stage, correct stage, over stage) was calculated. The statistical significance was set at p < 0.05.

Data availability
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.