Esophageal Cancer
Conditions
Brief summary
The goal of this observational study is to learn about the clinical utility of an artificial intelligence (AI) large language model in patients undergoing screening, diagnosis, treatment, and prognosis assessment for esophageal cancer. The main question it aims to answer is: Does the AI model improve early detection rate, diagnostic accuracy, treatment personalization, and prognostic prediction for esophageal cancer compared to standard care? Participants already receiving routine esophageal cancer management (including endoscopy, imaging, pathology, and clinical follow-up) as part of their regular medical care will have their de-identified data processed by the AI model; researchers will compare model-based recommendations and outcomes with standard care benchmarks over 3 years. Last updated on Oct 31, 2027
Interventions
OTHERObservational study; no assigned intervention. Participants receive routine esophageal cancer management (endoscopy, imaging, pathology, clinical follow-up) as standard care.
Routine esophageal cancer management including endoscopy, imaging, pathology, and clinical follow-up as per standard clinical practice. No additional, experimental, or assigned intervention is administered. The AI large language model processes de-identified data from routine care for comparative analysis against standard care benchmarks over 3 years.
Sponsors
The First Affiliated Hospital of Henan University of Science and Technology
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to 90 Years
Healthy volunteers
No
Inclusion criteria
* 1\. Aged 18 years or older. 2. Individuals with normal findings or inflammatory changes: endoscopic or pathological reports indicating "no significant abnormalities detected" or changes consistent with inflammation. 3\. Individuals with benign lesions: pathological reports specifying "absence of tumor cells" or a diagnosis consistent with benign lesions. 4\. Individuals with precancerous lesions: pathological reports with a definitive diagnosis of Low-grade Intraepithelial Neoplasia (LGIN) or High-grade Intraepithelial Neoplasia (HGIN). 5\. Individuals with malignant tumors: pathological reports confirming a diagnosis of esophageal squamous cell carcinoma or esophageal adenocarcinoma.
Exclusion criteria
* 1\. Diagnostically uncertain: Lack of definitive pathological evidence, or with doubtful clinical diagnosis. 2\. Poor data quality: Low-quality key imaging data (endoscopy, CT) that is unsuitable for analysis (e.g., severe artifacts, missing images). 3\. Severe missingness of key clinical or follow-up data (missing rate \> 20%). 4. Confounding by other malignancies: Presence of other active malignant tumors other than esophageal cancer within 5 years prior to enrollment. 5\. Loss to follow-up: Failure to obtain key survival or recurrence follow-up information in the retrospective cohort.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Area under the ROC curve (AUC) of the multimodal model for diagnosing esophageal cancer, calculated by ROC analysis using pathological biopsy as the gold standard, based on 5-fold cross-validation on the internal validation set. | 1 year and 5 months | The AUC ranges from 0.5 to 1.0, with higher values indicating better diagnostic performance. |
| Overall accuracy (proportion of correct classifications) of the multimodal model for diagnosing esophageal cancer, derived from the confusion matrix of the model's predictions on the internal validation set, with pathological biopsy as the gold standard. | Up to 3 years | The accuracy ranges from 0% to 100%, with higher percentages indicating better classification performance. |
| Concordance index (C-index) of the multimodal model for predicting overall survival and progression-free survival, derived from Cox proportional hazards model on time-to-event data. | Up to 3 years | The C-index ranges from 0.5 to 1.0, where 0.5 indicates random chance and 1.0 indicates perfect prediction; higher values indicate better predictive discrimination. |
Countries
China
Outcome results
None listed