Development of a predictive model for upper gastrointestinal major pathology

Aims Upper gastrointestinal endoscopy (UGIE) is the mainstay for diagnosing major organic diseases of the upper GI tract including esophageal and gastric cancer, as well as peptic ulcer and Barrett’s esophagus (BE). However, its efficacy and availability are hampered by inappropriate and over-prescriptions.

Aim was to derive a predictive model for major organic diseases of the upper GI tract to risk-stratify patients before prescribing UGIE.

Methods Retrospective analysis of prospectively collected data through endoscopy reporting system (Tesi EndoxWeb) at our center, including adult patients undergoing UGIE for clinical indication. A predictive model for major organic diseases of the upper GI tract was derived based on indication to UGIE and patients’ demographics through logistic regression, computing odds ratio (OR) with 95% confidence interval (95%CI) and its discriminant power assessed through area under ROC curve (AUROC), whereas calibration was assessed through calibration plot. Two risk categories were defined, i.e. low-risk (< 9%), and high-risk (≥ 9%). The number needed to scope (NNS) was computed.

Results Overall, 3254 patients who underwent UGIE were included. Mean patient age was 64 + 16 years, and 56% were female. Major organic disease was found in 330 (10%) patients, including 23 esophageal or gastric cancer, 255 peptic ulcers, and 65 BE cases. The best performing predictive model included age≥70 years (OR 1.65, 95%CI 1.15-2.38), male sex (OR 1.94, 95%CI 1.53-2.45) and the presence of at least one alarm feature including dysphagia, vomiting, anemia, or weight loss (OR 2.74, 95%CI 2.14-3.49). The model discriminative performance was fair (AUROC 0.68, 95%CI 0.65-0.71), whereas calibration was good according to visual inspection of calibration plot. The NNS was 19 in the low-risk group and 7 in the high-risk group.

Conclusions We derived a simple prediction tool to risk-stratify patients before prescribing UGIE, with two-fold implication, i.e., to help clinicians in appropriate UGIE prescription, and to help healthcare providers defining the priority of UGIE and optimizing resource allocation.

Publication History

Article published online:
27 March 2025