Background: Mathematical models provide a critical complement to empirical studies for evaluation of primary and secondary prevention strategies for upper gastrointestinal cancers. Existing evidence syntheses on mathematical modelling in upper gastrointestinal cancer prevention remain largely strategy- and disease-specific, rather than providing an overarching map of model-based evidence across primary and secondary prevention.
Objectives: We aimed to systematically document how mathematical models have been applied in upper gastrointestinal cancer prevention and to identify evidence and methodological gaps, thereby providing a foundation to guide future policy-oriented modelling studies for upper gastrointestinal cancer control.
Methods: We systematically searched literature up to November 4, 2025, from PubMed, Embase, Web of Science and China Biology Medicine disc, and included mathematical modelling studies evaluating primary and secondary prevention strategies for upper gastrointestinal cancers. Information on prevention strategies and modelling approaches was extracted.
Results: We screened 2709 records and included 135 studies. Most of studies were conducted in the United States of America (USA) (n=46, 34.07%), China ranking the second (n=31, 22.96%). Model calibration was performed in 40 studies (29.63%). Model validation was reported in 41 studies (30.37%), and 24 studies (17.78%) conducted both calibration and model validation. Studies from the USA (calibration: 52.17%; validation: 34.78%) more frequently reported model calibration and validation than those from China (calibration: 22.58%; validation: 32.23%). Helicobacter pylori eradication was the most frequently evaluated risk-factor interventions (n = 31; 64.58%), while no study assessed the impact of risk-factor interventions on esophageal cancer specifically, except for studies evaluating these interventions in multi-site cancers. Markov models were the most used model in our included studies (n=89, 65.92%), followed by microsimulation models (n=24, 17.78%). Among studies using precursor-state natural history models, the most frequently reported gastric cancer structure progressed stepwise from normal mucosa to atrophic gastritis, intestinal metaplasia, and dysplasia, followed by sequential transitions to local, regional, and distant gastric cancer; this structure was reported in seven studies. For esophageal cancer, the most common structure modelled stepwise progression from normal epithelium to Barrett’s esophagus, low-grade dysplasia, high-grade dysplasia, and ultimately esophageal adenocarcinoma, and was reported in 18 studies. Atrophic gastritis was the most frequently included precursor state in gastric cancer models (n=20, 62.50%), Barrett’s esophagus was the most frequently included precursor state in esophageal cancer models (n=35, 74.47%), and most of the studies (n=37, 78.72%) targeted on esophageal adenocarcinoma. Overall, assumptions, strengths, and limitations were closely tied to model purpose and structure, and heterogeneity in natural history representation, intervention implementation assumptions, and reporting practices constrained direct cross-study comparability of findings.
Conclusion: Despite rapid growth in mathematical modelling for upper gastrointestinal cancers control, there remains a need for more high-quality, transparently reported studies, particularly those focused on Chinese populations, and for increased modelling attention to esophageal squamous cell carcinoma in China and other high-burden settings.