Researchers have developed a new framework called AWARE that improves the accuracy of clinical risk prediction in electronic health records using tabular foundation models. These models are designed to handle high-dimensional data and class imbalance issues common in medical datasets. The authors compared various models, including classical and deep learning approaches, on a multi-cohort benchmark. They found that retrieval quality and alignment between retrieval and inference stages were key factors affecting the performance of tabular in-context learning models. AWARE outperformed other methods by up to 12.2% under extreme class imbalance conditions. --- Why it matters: This research matters because it provides insights into developing more accurate clinical risk prediction models that can handle real-world data complexities, such as high dimensionality and class imbalance. These findings have implications for the development of AI-powered healthcare systems that can better support medical decision-making. Source: https://arxiv.org/abs/2604.01841