Year: 2026
Type: dataset
Abstract: Large observational datasets have the potential to improve predictive performance and generalizability of clinical prediction models. We test whether and to what extent clinical prediction benefits from such data by comparing multiple modeling approaches across outcomes and databases, examining model transportability and how performance scales with sample size. We show that deep learning methods do not consistently improve with more data, and conventional methods such as regularized logistic regression remain competitive. Future work should prioritize integrating routinely captured but underutilized data types and enriching existing datasets, for example, by incorporating curated clinical knowledge.
Author: John, Luis Henrik
Institution: Erasmus MC
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7163474948
