Search papers, labs, and topics across Lattice.
This paper introduces CAST, a novel framework that enhances the robustness of clinical language models by addressing the issue of note-specific artifacts that can skew predictions. By employing Sparse Autoencoders to identify and suppress these artifacts while providing human-auditable features, CAST not only improves prediction accuracy on MIMIC-IV discharge-note mortality but also maintains competitive performance against leading LLMs. The key result demonstrates that CAST not only enhances model reliability but also offers a transparent audit trail for clinical decision-making.
Clinical language models can be made auditable and robust against deployment shifts by effectively suppressing misleading artifacts, revealing a clear path to trustworthy AI in healthcare.
Clinical language models can achieve strong in-hospital accuracy yet fail under deployment shifts because they exploit note-specific artifacts (e.g., templates, separators, boilerplate) that do not reflect patient state. We propose CAST (Concept-guided Artifact Suppression Tuning), an SAE-based framework for auditable clinical text classification. CAST uses Sparse Autoencoders to expose sparse, human-auditable features from intermediate Transformer activations, labels SAE latents with an LLM-assisted interpretation pipeline and ICD-10 retrieval constraints, suppresses verified artifact latents via residual subtraction during fine-tuning, and provides post-hoc per-concept attributions for auditing model decisions. On MIMIC-IV discharge-note mortality prediction, CAST improves over its corresponding fine-tuned encoder baselines and remains competitive with strong LLM baselines, while producing a feature-level audit trail of the clinical concepts that support each prediction and the artifact concepts suppressed during training.