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

Area of Science:

• Oncology
• Medical Informatics
• Epidemiology

Background:

• Upper gastrointestinal (UGI) cancers frequently present at late stages, hindering effective treatment.
• Early detection through risk prediction models can identify high-risk individuals for timely investigation.

Purpose of the Study:

• To systematically review and synthesize evidence on diagnostic risk prediction models for UGI cancers.
• To evaluate the performance and characteristics of existing UGI cancer risk prediction models.

Main Methods:

• Systematic literature search of MEDLINE, Embase, and CENTRAL for studies on UGI cancer risk prediction models.
• Inclusion of studies reporting quantitative model performance measures.
• Analysis of model development, validation, predictors, and performance metrics.

Main Results:

• 82 studies described 162 UGI risk models for pancreatic, gastric, oesophageal, gallbladder, and biliary tract cancers.
• Logistic regression was common, with increasing use of machine learning.
• Only 33 models underwent external validation, with 15 achieving an Area Under the Curve (AUC) ≥0.80.

Conclusions:

• Several UGI cancer risk prediction models exhibit strong performance upon external validation.
• Further research is required to compare top-performing models and assess their clinical utility, cost-effectiveness, and acceptability.
• Potential exists for developing integrated UGI multi-cancer prediction models due to overlapping risk factors.