Using artificial intelligence tools to predict recurrence risk in pancreatic cancer. A systematic literature review and meta-analysis

A systematic review and meta-analysis of 10 studies (2019–2024) evaluating the diagnostic accuracy of artificial intelligence (AI) algorithms for predicting pancreatic cancer (PCa) recurrence was conducted. The pooled sensitivity and specificity estimates were 0.77 [95 % CI: 0.58–0.95] and 0.79 [95 % CI: 0.57–1.00], respectively. Key limitations of the study included high heterogeneity (I² > 98 %), which could be related to the small number of included studies and insufficient standardization of the validation methods. Background: Artificial intelligence (AI) tools provide new possibilities in predicting the course of pancreatic cancer. Purpose: To conduct a meta-analysis of the diagnostic accuracy of AI algorithms (sensitivity and specificity) for predicting PCa recurrence and to compare the effectiveness of different types of algorithms. Methods. A systematic literature search was conducted in leading scientific databases, covering publications from 2019 to 2024. The review included studies that applied artificial intelligence tools to predict the risk of pancreatic cancer recurrence. The data search and analysis were conducted in three stages: a primary search of studies using keywords and inclusion criteria; screening of the titles and abstracts to select relevant studies; and a detailed assessment of the full texts of the selected articles. The data synthesis included an analysis of the performance of the AI models, the types of data used (clinical, genomic, radiological, etc.), and the validation and testing strategies for the proposed algorithms. A random-effects model was used for the sensitivity and specificity meta-analysis, with the calculation of pooled estimates, 95 % confidence intervals, and heterogeneity indices (I², τ²). A meta-regression was also performed to assess the impact of the algorithm type on sensitivity. Statistical analysis was carried out in R (metafor package) with forest plot visualization. Results: This systematic review included 10 studies, of which 5 were selected for the meta-analysis. The results demonstrate a pooled sensitivity of 0.77 [95 % CI: 0.58–0.95] and specificity of 0.79 [95 % CI: 0.57–1.00] for AI algorithms in predicting PCa recurrence. In the analysis of individual algorithm types, artificial neural networks (ANNs) showed a pooled sensitivity of 0.87 [0.73–1.01], while support vector machines (SVMs) had a negative impact on sensitivity (coefficient –0.45 [–0.69 to –0.21]). The meta-analysis revealed high heterogeneity of the studies (I² = 98.84 % for sensitivity and I² = 99.42 % for specificity), requiring cautious interpretation of the results. Conclusion: AI models are promising tools for predicting pancreatic cancer recurrence, but require data standardization and prospective validation in clinical practice.