Generative artificial intelligence (AI) is emerging as a transformative force in medical research, enabling clinicians to develop machine-learning models using natural-language prompts rather than conventional programming approaches. This innovation lowers technical barriers to data analysis and facilitates rapid exploratory research across a broad range of fields, including prognostic modeling, risk-factor assessment, medical imaging, and public-health studies. Prompt-based analytical workflows also contribute to reproducibility by allowing prompts to be stored and shared as a transparent record of analytical processes. However, several challenges remain, such as the limited interpretability of AI models, variability in generated outputs, concerns regarding data security and governance, and insufficient AI literacy among clinicians. Addressing these limitations requires appropriate ethical oversight and strict adherence to scientific rigor. Looking forward, advances in multimodal AI that integrate clinical, imaging, and genomic data are expected to further expand the role of generative AI, evolving it from a technical support tool into a collaborative research partner. For vascular specialists, the responsible adoption of these technologies will be increasingly important for advancing research and improving patient care.

Statistics and machine learning (ML) serve complementary roles in clinical research. Statistics focuses on explanation through causal inference, effect estimation, uncertainty quantification, and model-assumption checking, while ML emphasizes prediction and generalization using algorithmic methods and cross-validation. Using Breiman’s “Two Cultures” and Shmueli’s “Explanation vs. Prediction” as a framework, we summarize key conceptual and practical differences between data-model approaches (e.g., regression) and algorithmic approaches (e.g., decision trees, ensemble methods). We highlight differences in objective functions, evaluation metrics, and interpretability, and discuss ROC/AUC, calibration, and threshold setting in clinical decision-making. We propose a hybrid framework consisting of an explanatory layer for causal validity, a predictive layer for generalization, and a bridging layer for visualization, interpretability, and context-specific thresholds. Using our study on predicting aneurysm sac shrinkage after EVAR, we show that integrating statistical inference with ML decision trees enables both rigorous explanation and individualized prediction.