Search papers, labs, and topics across Lattice.
The authors adapt Foresight Learning to the clinical domain, automatically generating training data from time-ordered clinical notes in MIMIC-III to predict future clinical events. This approach converts notes into question-answer pairs about future events, using later documentation to resolve the labels. Training a LoRA adapter on this automatically generated dataset significantly improves prediction accuracy and calibration compared to the base model and even slightly outperforms GPT-4 on held-out questions.
Forget hand-engineered features: this work shows how to train LLMs to predict clinical events directly from unstructured notes, achieving surprisingly strong results.
Longitudinal clinical notes contain rich evidence of how patients evolve over time, but converting this signal into training supervision for clinical prediction remains challenging. We extend Foresight Learning to clinical prediction by converting time-ordered MIMIC-III notes into examples consisting of past patient context, a natural-language question about a possible future event, and a label resolved from later documentation. This process yields 6,900 prediction examples from 702 admissions across medications, procedures, organ support, microbiology, and mortality. A small LoRA adapter trained on these examples improves over the prompted base model, reducing expected calibration error from 0.1269 to 0.0398 and Brier score from 0.199 to 0.145, while slightly outperforming GPT-5 point estimates on held-out questions. The approach enables reusable clinical prediction supervision from longitudinal notes without hand-engineered structured features or endpoint-specific classifiers.
