Diagnostic Risk Prediction Models for Upper Gastrointestinal Cancers: A Systematic Review.

Abstract

Upper gastrointestinal (UGI) cancers are often detected late. Risk prediction models could facilitate earlier detection by identifying patients at risk for further investigation. We systematically reviewed evidence on UGI diagnostic risk prediction models. A search of MEDLINE, Embase, and CENTRAL was conducted for studies reporting on the development and/or validation of diagnostic risk prediction models for UGI cancers (pancreatic, gastric, esophageal, gallbladder, and/or biliary tract). Studies had to report at least one quantitative measure of model performance to be eligible for inclusion. A total of 82 studies describing 162 UGI risk models were included. Models predicted gallbladder (n = 6), gastric (n = 25), esophageal (n = 34), gastroesophageal (n = 14), and pancreatic (n = 83) cancers. Most models used logistic regression, but machine learning was increasingly used from 2019. In total, 366 unique variables were incorporated across models. Only 33 models were externally validated, with 15 achieving an AUC ≥0.80. This review highlights that several models perform well in predicting UGI cancers on external validation. Future research is needed to compare the best-performing models and assess their clinical utility, acceptability, and cost-effectiveness. Given the significant overlap in at-risk populations and predictors across UGI cancers, there may also be scope to develop UGI "multicancer" models.