One recently published Chinese study concluded that MCCE could detect GCs in a large population, but its role as a first-line screening tool for GC remains to be further validated . Because the risk of GC is closely related to H. pylori infection status, MCCE’s diagnostic accuracy for H. pylori infection status is of critical importance in risk stratification. Moreover, the morphological features of early GC or high-grade precancerous lesions also differ according to different H. pylori infection statuses, which further established the rationale for our study.
Yoshii et al. demonstrated that the overall diagnostic accuracy of three H. pylori infection statuses was 82.9% on white light endoscopy by the Kyoto classification of gastritis . In this study, we found that most of the key findings documented in the Kyoto classification of gastritis were recognizable on MCCE, H. pylori infection status could be accurately diagnosed via MCCE, and the overall diagnostic accuracy was 80.2%, comparable with EGD. Previous studies demonstrated that MCCE could detect various types of gastric lesions, including erosions, polyps, ulcers, and even superficial early gastric cancers [5, 9, 12, 16]. In our study, we found that the Kyoto classification of gastritis generally applied well to MCCE in the diagnosis of H. pylori infection status.
In the diagnosis of current infection status, the most reliable finding was mucosal swelling (sensitivity 76.5%, specificity 88.7%, PPV 80.2%), whereas in other recently published EGD studies, that diagnosis was established mainly based on observation of diffusive redness. This difference, we speculate, might have been the reason why MCCE had a higher DOR for current infection compared with conventional EGD (77.2 vs. 21.7) [10, 16, 17].
MCCE can reliably diagnose noninfection status, with a sensitivity, specificity and PPV of 83.8%, 85.0% and 82.2%, respectively. This diagnosis is mainly based on observation of RAC; although FGP and streak redness were also of high specificity and PPV, these two findings were relatively uncommon. However, MCCE’s DOR for noninfection status in our study was much lower than that of Yoshii’s EGD study (30.7 vs. 98.6), in which the authors made the diagnosis based on the same findings. The Kyoto gastritis classification defines RAC as microvascular networks observed in the lower part of the gastric corpus, mainly the lesser curve side [10, 14]. MCCE’s diagnostic performance on past-infection status was suboptimal in our study, largely due to the lack of specific findings. In addition, interobserver variability might also have played a role in its low diagnostic performance. A new discovery in our study was that the combination of RAC and map like redness could be used as a highly specific predictor for past infection, with a specificity, PPV and DOR of 98.9%, 86.7% and 39.4, respectively. This combination of findings is especially helpful for determining past infection status when there is diagnostic ambiguity.
Our study had several strengths. First, this was a prospective study in which the reviewers were blinded to the final results, and we used UBT results as the gold standard for the diagnosis of H. pylori infection, making the results reliable and robust. Second, we have found several combinations of findings with a high diagnostic value, which is useful when the diagnosis was uncertain based on observation of a single finding. Third, we performed regression analyses in which the diagnostic performance of MCCE was assessed by combining 10 findings in the Kyoto classification of gastritis. Fourth, we had an expert and a nonexpert review of MCCE images and resolved interobserver disagreement by a referee, making our results reproducible in future studies.
Our study had several limitations. First, although all participants were prospectively recruited, approximately half of the included participants were H. pylori noninfected (44.5%), while the proportion of past-infection participants was particularly low (15.9%); thus, according to STARD (Standards for reporting of diagnostic accuracy studies), selective bias was inevitable . Second, according to the Kyoto classification of gastritis, sticky mucus and hyperplastic polyps are also key findings for H. pylori infection, but these findings were not included in our study, nor could we rate the degrees of atrophy and intestinal metaplasia on MCCE, so the scoring system of Kyoto classification of gastritis described in previous studies [19,20,21,22] was not available in this study. Therefore, the Kyoto classification of gastritis used in this study was actually a modified version [10,11,12, 14, 23]. Third, spontaneous eradication of H. pylori might have occurred in a small portion of the study participants, which could have impacted the evaluation of the diagnostic accuracy, which might have been underestimated in our study [10, 24].
In future studies, more specific findings for past infection are warranted because using map-like redness as the predictor does not appear to have sufficient diagnostic power. Additionally, in recent years, the introduction of artificial intelligence (AI) has improved the diagnostic accuracy of GI neoplasms as well as EGD’s diagnostic accuracy on H. pylori infection status [23, 25]. Hopefully, our results could help establish MCCE’s AI diagnosis of H. pylori infection status, thus improving GC’s early detection in a more reliable way [25, 26]. Moreover, efforts to establish scoring models for atrophy and intestinal metaplasia on MCCE are needed, which may help us better stratify GC risks via MCCE [19,20,21].