Diagnostic accuracy of contrast-enhanced CT for stage III gastric cancer: A prospective Chinese multicenter study

To the Editor: China is one of the countries with the highest burdens of stomach cancer.[1] Approximately 70% of gastric cancer (GC) patients in China are diagnosed at an advanced-stage,[2] where the primary treatment strategies include proceed with surgery or initiate neoadjuvant chemotherapy (NAC). Compared with Western countries, NAC is used more conservatively for resectable GC in countries such as Japan and Korea, where D2 lymphadenectomy was routinely performed.[3] This discrepancy may result in higher overstaging rates in those countries. Contrast-enhanced computed tomography (CECT) is commonly utilized for GC diagnosis and staging. However, its concordance with pathological diagnosis based on the TNM staging system remains suboptimal.[4] We conducted a prospective, multicenter study to evaluate CECT diagnostic accuracy and the rate of misdiagnosed pathological stage I (pI) cases among patients with clinical stage III GC in China. The study involved 21 tertiary centers across China, all registered with the China Gastrointestinal Cancer Surgery Union. The objective was to provide supplementary diagnostic criteria for NAC trial design and clinical decision-making. This study was approved by the institutional review board of the medical ethics committee of Peking University Cancer Hospital (2019YJZ55), as well as by each participating center, written informed consent was provided by all participants or their legal representatives. This study was registered at https://clinicaltrials.gov/ with ID NCT04440605. From July 2020 to February 2022, outpatients with gastric adenocarcinoma were enrolled based on the following criteria: (1) histologically confirmed GC; (2) clinical TNM staging, including tumor invasion depth and lymph node metastasis, prospectively assessed by CECT; (3) no evidence of stage IV disease; (4) direct radical surgery with D2 lymphadenectomy; and (5) availability of a standardized pathological report including pT, pN, and overall pathological stage. Patients were excluded (1) surgery was canceled; (2) fewer than 15 lymph nodes were retrieved; or (3) the pathological diagnosis was not adenocarcinoma. We defined the primary endpoint as the proportion of pI tumors among patients clinically diagnosed with “cT3-4aN+” (criterion B). Secondary endpoints included the proportion of pI-stage tumors among the following clinical diagnostic criteria: (1) “cT2N3, cT3N2/N3, or cT4aN1/N2/N3” (criterion A, considered as a broader category of the cTNM stage III classification in this study); (2) “cT3-4a” (criterion C, considered as a broader category of the cTNM stage III classification in this study); (3) “cTxNy (x+y≥4)”; (4) “cN+”; (5) cT4aN+ (criterion Z). We also assessed concordance between clinical and pathological staging. Sensitivity for detecting pIII was compared across diagnostic criteria to determine the optimal criteria. Clinical TNM staging was based on the 8th edition of the American Joint Committee on Cancer (AJCC) staging manual, as detailed in Supplementary Table 1, https://links.lww.com/CM9/C673. Each patient’s clinical stage was independently assessed by at least two experts. The expected value for the primary endpoint was set at 3%; the half-width of the 95% confidence interval (CI), calculated using the Clopper-Pearson method, was set at ≤1.5%. Sample size estimation was conducted using two approaches. The first, referencing JCOG1302A and utilizing a Bayesian method, targeted an 80% probability that the half-width of the 95% CI for any point estimate would remain within 1.5%, resulting in a sample size of 968. The second method aimed to ensure that the 95% CI for the pI rate had a half-width of ≤1.5%, yielding a required sample size of 743. We determined that a minimum of 743 cases would be necessary; improved stability was expected if 968 cases were included. For continuous variables, we reported means and standard deviations or medians and interquartile ranges. Categorical variables were summarized as percentages and analyzed using the chi-squared test. Diagnostic performance was evaluated using pathological staging as the reference standard. We reported the proportion of pI-stage cases within each clinical stage III diagnostic category, along with overdiagnosis and underdiagnosis rates, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). We used the Spearman's rank correlation coefficient to investigate the monotonic relationship between the clinical staging and the pathological staging. P-values <0.05 were considered statistically significant. All statistical analyses were performed using R software (version 4.2.0; R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/). Between July 2020 and February 2022, 1915 patients were enrolled, and 1735 patients—including 808 with cIII—were eligible for the final analysis [Supplementary Figure 1, https://links.lww.com/CM9/C673]. Baseline characteristics are presented in Supplementary Table 2, https://links.lww.com/CM9/C673. The proportions of pI cases among patients clinically staged as cT1, cT2, cT3, and cT4a were 85.22% (196/230), 59.10% (211/357), 23.08% (141/611), and 7.64% (41/537), respectively [Supplementary Table 3, https://links.lww.com/CM9/C673]. The overall accuracy of clinical T stage (cT) relative to pathological T stage (pT) was 44.03% with a correlation coefficient of r = 0.621; and the diagnostic accuracy for pathological N stage (pN) was 41.50% (720/1735), with a correlation coefficient of r = 0.468 [Supplementary Tables 4 and 5, https://links.lww.com/CM9/C673]. We analyzed cN1–3 as a single group and found that the diagnostic accuracy of CECT for pathological N stage was 67.09% (1164/1735), with 69.13% (654/946) sensitivity and 66.03% (521/789) specificity for diagnosing pN1–3 [Supplementary Table 6, https://links.lww.com/CM9/C673]. Using various diagnostic criteria, we evaluated the misdiagnosis rates of pI-stage tumors: Among 1148 tumors staged as cT3/T4a (criterion C), 15.85% (182/1148) were pI and 40.68% (467/1148) were overstaged; Among 808 tumors staged as cT3/4N+ (criterion B), 10.89% (88/808) were pI and 32.30% (261/808) were overstaged; Among 570 tumors staged as “cT2N3, cT3N2/N3, or cT4aN1/N2/N3” (criterion A), 6.32% (36/570) were pI and 24.92% (142/570) were overstaged; For the criteria “cTxNy (x+y≥4)” and “cN+,” the pI rates were 11.82% of 905 and 14.43% of 922 tumors, respectively. Although, criterion Z have a narrow clinical staging scope, it only reduced the pI rate to 5.93%. None of these criteria met the target threshold (the rate of pI is less than 4.5%); all except criterion A and Z had a pI rates exceeding 10% [Supplementary Table 7, https://links.lww.com/CM9/C673]. The diagnostic performances of the six criteria were evaluated using accuracy and receiver operating characteristic curves [Supplementary Figure 2, https://links.lww.com/CM9/C673]. Sensitivity, specificity, and accuracy values for each criterion are provided in Supplementary Table 7, https://links.lww.com/CM9/C673. Among these, criterion A demonstrated the highest accuracy for pIII at 71.99%. For pIII/IV, accuracies were identical under both criterion A and criterion B (73.31%), but criterion A yielded higher specificity (85.60% vs. 73.53%) and lower sensitivity (57.14% vs. 73.03%). These findings align with the previously noted difference in pI rates, which were substantially higher under criterion B than under criterion A (10.89% vs. 6.32%). After using the more stringent criterion Z, pI rate was slightly reduced (5.93% vs. 6.32%) and the specificity was enhanced (88.95% vs. 85.60%) compared with criterion A; however, diagnostic accuracy for stage III significantly declined, as indicated by the area under the curve (0.68 vs. 0.71, P <0.001). Thus, we propose using criterion A—the original cIII criteria recommended by the 7th edition of the AJCC staging manual—to identify patients for future NAC studies. We analyzed potential factors associated with overdiagnosis of pI based on criterion A [Supplementary Table 8, https://links.lww.com/CM9/C673]. Among seven centers that diagnosed more than 50 cases, the pI misdiagnosis rate ranged from 1.45% to 11.11% and four of these centers (57.14%) reported pI rates >5%. Tumor location showed a significant association with pI misdiagnosis rate (P = 0.018) and a marginal association was observed between tumor differentiation and pI misdiagnosis rate. Other clinicopathological features, including reporters identity, pathological subtype, Lauren classification, and sex, were not significantly associated with variation in pI rates. This study represents the prospective, multi-institutional validation of the diagnostic accuracy of CECT in patients with GC in China. Reported sensitivity and specificity values for T1–T4 staging are 13–94% and 80–100%, 29–68% and 78–92%, 37–91% and 70–95%, and 29–86% and 92–99%, respectively. For N0 staging, sensitivity and specificity are 63–92% and 42–88%, respectively.[5] Based on prospective data from 21 tertiary centers, our study provides a reference benchmark for diagnostic accuracy in this context. According to the results of JCOG1302A, we adopted criterion B as our primary endpoint. Our preliminary retrospective single-center analysis showed a pI rate of only 1.80% (4/222) under criterion B. Therefore, the observed 10.89% pI rate under criterion B in our study was unexpected. We conducted several secondary analyses using data from both JCOG1302A and our study. As we excluded patients with cT1 disease to meet the same inclusion criteria in JCOG1302A, the sensitivities for detecting pIII increased to 91.94%, 73.28%, and 57.01% using criteria C, B, and A, respectively—each higher than corresponding values reported in JCOG1302A [Supplementary Table 9, https://links.lww.com/CM9/C673]. Overall staging accuracy was comparable between the two studies: after exclusion of cT1 patients in our study and cT4b in JCOG1302A, accuracy for cT staging in our study was 38.01% (572/1505) compared with 38.96% (487/1250) in JCOG1302A. Similarly, for cN0/N+ staging, accuracy in our study was 69.24% (1042/1505), compared with 63.61% (783/1231) in JCOG1302A. Besides, we noted a contrasting diagnostic preference in JCOG1302A: 1) the rate of pT3–T4 disease among cT2 patients was 24.65% (88/357) in our study, lower than the 36.17% (119/329) in JCOG1302A; and 2) the rate of pN+ disease among cN0 patients was 37.27% (303/813) lower than the 52.15% (303/581) in JCOG1302A. Considering that the overall accuracy rate of cT and cN staging were comparable between two studies. These findings suggest that tumors tend to be overstaged in Chinese centers and understaged in Japanese centers.[6,7] In conclusion, our findings indicate that the “cT3-4aN+” criterion proposed by JCOG1302A for NAC indications results in a substantial proportion of overstaged pI cases in China. We recommend adopting the more stringent criterion “cT2N3, cT3N2/N3, or cT4aN1/N2/N3” to reduce the rate of pI contamination. Funding This work was funded by Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0520600), Beijing Municipal Public Welfare Development and Reform Pilot Project for Medical Research Institutes (JYY2023-3), Capital’s Funds for Health Improvement and Research (2024-1-2152), ZLRK202525, National Key R&D Program of China (2023YFC2415200), and Beijing Nova Program (No. 20220484111). Conflicts of interest None.