Deployment-Centered Evaluation: Predicting Query-Level Rejection Risk in a Clinical LLM System

Researchers at an academic medical center developed a pre-response classifier that predicts the risk of a user rejecting an LLM-generated clinical response, achieving an AUROC of 0.719 over 4.5 months of user feedback. The model leverages deployment-specific context—such as provider type, department, and language model used—alongside query content to estimate rejection risk before generation. This approach enables targeted guardrails and abstention mechanisms, addressing blind spots in static benchmarks that fail to capture real-world user acceptance in clinical LLM systems.

arXiv:2606.12702v1 Announce Type: new Abstract: Large language models LLMs are increasingly integrated into clinical systems, making it essential to evaluate the real-world utility of these systems. However, static benchmarks tend to measure correctness rather than user acceptance, aggregate performance across queries, and require densely annotated datasets -- leading to major blind spots for evaluating clinical systems. In this work, we perform a deployment-centered evaluation of an LLM system embedded within electronic health records at an academic medical center, where user feedback is sparse but closely reflects the deployment conditions. Specifically, we train a pre-response classifier that estimates the risk that a future interaction will result in the user rejecting the LLM response, based on query content and deployment-specific context available before generation. We conduct a prospective analysis of our model over 4.5 months of user feedback, finding that our prediction model achieves an AUROC of 0.719. Further, we estimate the benefit of such predictions in two downstream use cases guardrail triggering and abstention . Our key conceptual insight is that making use of deployment-specific context i.e., the provider type, department name, language model used for response , as opposed to only query content, improves the ability to predict whether the user will reject the system output. Altogether, our empirical case study demonstrates the feasibility of predicting user rejection using deployment-specific context, opening the door to targeted guardrails.