Routine Scans and Scopes Fall Short in Predicting Stomach Cancer Chemotherapy Response, Study Finds

For patients with stomach or gastroesophageal junction cancer, the question of whether preoperative chemotherapy is actually working has long been one of the most frustrating blind spots in oncology. Surgeons and oncologists routinely order endoscopies and CT scans before operating, hoping the images will reveal how much of the tumor has been destroyed by treatment. But a new retrospective study from Ankara Bilkent City Hospital in Türkiye suggests that these everyday clinical tools may be far less reliable than many clinicians assume when it comes to predicting what pathologists will find under the microscope after surgery.

The research, published in Frontiers in Oncology, examined how well routine endoscopic and computed tomography response assessments corresponded to major pathological response (MPR) in patients who received FLOT-based chemotherapy before gastrectomy. FLOT, which combines fluorouracil, leucovorin, oxaliplatin, and docetaxel, has become a standard perioperative regimen for fit patients with resectable gastric or gastroesophageal junction adenocarcinoma following the landmark FLOT4-AIO trial. Yet while the regimen can induce substantial tumor regression, the extent of that regression can only be definitively determined by examining the resected specimen, leaving clinicians to rely on imperfect imaging and visual assessments in the meantime.

The study team screened 450 institutional records and identified 47 patients who met strict criteria: they had received preoperative FLOT-based therapy, undergone gastrectomy, and had both pretreatment and post-treatment endoscopies detailed enough to classify the response of the primary tumor. Of these, 37 patients also had comparable paired CT assessments, allowing a head-to-head comparison between the two modalities. The remaining 10 patients lacked comparable CT studies, which meant all CT and combined-modality analyses were restricted to the smaller paired cohort.

Pathological response proved to be a moving target in terms of how it was documented. The researchers found that gastrectomy pathology reports used a bewildering variety of grading systems: Ryan or modified Ryan grading was explicitly reported in 10 cases, Becker grading in 2, a TNM8-based regression score in 2, an unnamed numerical score or grade in 21, and purely descriptive terminology in 12. To create a common analytical framework, the team harmonized all reports into four ordered categories: complete response, near-complete response, partial response, and poor or no response. Major pathological response was defined as complete or near-complete response, while pathological complete response (pCR) referred to the complete disappearance of viable tumor.

The results were sobering. Major pathological response occurred in 17 of 47 patients, or 36.2 percent, including 10 patients with pathological complete response, representing 21.3 percent of the cohort. These rates are broadly consistent with what has been reported in the FLOT era, but the real question the study addressed was not whether FLOT works, but whether clinicians can reliably recognize that it is working before the operation. The answer, based on this data, appears to be: not very well.

Endoscopic response, classified using a framework informed by Japanese gastric cancer conventions, showed only a weak and statistically non-significant ordinal association with pathological response. The Spearman rank correlation coefficient was just 0.189 with a p-value of 0.204. When the four-level endoscopic categories were analyzed against the binary MPR outcome using receiver operating characteristic analysis, the area under the curve (AUC) in the full cohort was 0.619, with a 95 percent confidence interval spanning from 0.466 to 0.771. An AUC of 0.5 represents random guessing, so this figure indicates only modest discriminative ability at best.

In the paired cohort of 37 patients, where both endoscopy and CT were available for the same individuals, the picture was similarly underwhelming. Endoscopy produced an AUC of 0.561, while CT yielded a numerically higher AUC of 0.675. The difference of 0.114 in favor of CT was tested using the DeLong method for comparing correlated ROC curves, but the result was not statistically significant, with a p-value of 0.35. The confidence interval for this difference was wide, ranging from approximately negative 0.13 to positive 0.35, meaning the data could not rule out either clinically meaningful superiority of CT or even a small advantage for endoscopy. An exploratory logistic regression model combining both modalities produced an apparent AUC of 0.692, an absolute increase of only 0.017 over CT alone, which the researchers emphasized was an uncorrected estimate developed and evaluated in the same small cohort.

One intriguing finding emerged from the endoscopic complete response category. Both patients who were classified as having a complete endoscopic response, meaning no visible tumor at the index site and only scar or healed mucosal changes, turned out to have pathological complete response on their resection specimens. This gave endoscopic complete response a positive predictive value of 100 percent for pCR, albeit based on just two observations. However, the sensitivity was only 20 percent, meaning this signal caught just 2 of the 10 patients who actually achieved pCR. The researchers noted that while this finding may merit prospective evaluation as a high-specificity signal, it cannot be regarded as a standalone decision rule given the fragility of the estimate.

The biological explanation for the discordance between modalities is straightforward once considered carefully. Endoscopy directly visualizes the luminal and mucosal surface of the stomach, but viable tumor cells may persist within deeper layers of the gastric wall that the scope cannot see. CT provides information about extraluminal disease and lymph nodes, but the primary gastric lesion is frequently non-measurable under RECIST v1.1 criteria, meaning the radiological response category may be driven by changes in lymph nodes or other target lesions rather than the primary tumor itself. Surgical pathology, by contrast, assesses treatment effect throughout the entire resected primary tumor bed. These three approaches are simply measuring different things, and the study demonstrates that the gaps between them are not easily bridged by routine clinical practice.

The authors are careful to frame their findings as hypothesis-generating rather than definitive. Only 47 of 450 screened records met the documentation requirements, with the vast majority excluded because no post-treatment endoscopy was performed or because the endoscopy report was insufficient to classify response. This selection process means the analytical cohort represents patients in whom clinicians chose to obtain and document restaging endoscopy, and the results may not generalize to the broader FLOT-treated population. The paired CT cohort contained only 11 MPR events, giving the comparison sufficient power to detect only large differences in discrimination. The researchers call for standardized prospective assessment with predefined intervals, uniform endoscopic criteria, primary-tumor-focused imaging measurements, blinded independent review, and central pathological grading before either modality can be used with clinical confidence to estimate pathological response. Until such evidence exists, they conclude, endoscopy and CT should be interpreted as complementary tools for staging and reassessment rather than as substitutes for the pathological truth that only surgery can reveal.

Subject of Research: Predicting pathological response to preoperative FLOT chemotherapy in gastric and gastroesophageal junction cancer using endoscopy and CT imaging

Article Title: Association of endoscopic and CT response with pathological response after preoperative FLOT-based therapy for gastric and gastroesophageal junction cancer: a retrospective cohort study

Article References: Gencer, H., Ucar, G., Algın, E., Sendur, M. A. N., & Bostanci, E. B. (2026). Association of endoscopic and CT response with pathological response after preoperative FLOT-based therapy for gastric and gastroesophageal junction cancer: a retrospective cohort study. Frontiers in Oncology, 16, Article 1960800. https://doi.org/10.3389/fonc.2026.1960800

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1960800

Keywords: gastric cancer, gastroesophageal junction cancer, FLOT chemotherapy, pathological response, endoscopy, computed tomography, tumor regression, RECIST, major pathological response, neoadjuvant chemotherapy, gastrectomy, ROC analysis