Problem
Research questions and friction points this paper is trying to address.
clinical documentation
provider queries
missing information
automated clarification
Closing Ambient Clinical Documentation Gaps with Automated Provider Queries
🤖 AI Summary
This study addresses the issue of missing information in clinical documentation, which frequently leads to coding and recording discrepancies, noting that existing automated approaches generally overlook active querying. We propose a large language model-based "Draft-Ask-Update" (DAU) closed-loop paradigm that automatically initiates clarification requests to physicians to refine medical records, ICD-10 coding, and order extraction. This approach integrates a transcription degradation benchmark with a multitask evaluation framework to optimize confidence analysis and recall strategies. Our work reveals task-specific differences in predictors of effective questioning, emphasizing the importance of learning when not to ask to avoid redundant interruptions. Experiments identify 9% of query turns as counterproductive, demonstrating that simple recall suffices for note completion while complex multiple-choice questions benefit coding tasks, thereby providing critical insights for system deployment.
📝 Abstract
Provider queries are clarifying requests sent by clinical documentation specialists to physicians to close gaps in the clinical note and ensure accurate billing. Prior work automates note drafting, ICD-10 coding, and order extraction assuming a complete transcript, leaving these gaps unaddressed. We study whether an LLM can automate the query loop, termed DAU (Draft, Ask, Update), across those three tasks. An audit of 3,000 real visits identifies the sources of missing documentation, from which we build five transcript-degradation benchmarks on public data. Analyzing 21k clarification turns on real conversations, we find useful-question predictors are task-specific: oracle confidence dominates, but note completeness needs only simple recall questions while ICD-10 coding needs harder, multi-option ones. About 9% of turns hurt performance, driven by redundant questions and non-answers that still trigger a rewrite. Deployment depends on learning "when not" as much as "what to" ask.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Large Language Model
Clinical Documentation
Provider Queries
DAU Framework
Transcript-Degradation Benchmarks
🔎 Similar Papers
No similar papers found.
