Voice-Driven EHR Flowsheet Population Using Generative AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Healthcare professionals face challenges in efficiently documenting patient data in electronic health records (EHR) flowsheets due to the time-consuming nature of navigating through numerous rows and columns, leading to delayed accessibility and increased inaccuracies.
Innovation Solution
A system utilizing generative artificial intelligence (AI) and retrieval augmented generation (RAG) to automatically populate relevant rows in flowsheets by processing speech inputs, segmenting meaningful information, and applying similarity-based searches to match transcript segments with appropriate key/value pairs in the flowsheet schema.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If nurses manually navigate and fill flowsheet rows in EHR, then data can be documented, but the process is time-consuming and reduces productivity
Solution Approach 1:
The system pre-processes clinical data during patient care delivery, transcribing speech to text and structuring information before the nursing documentation task. This preliminary processing eliminates the need for nurses to manually navigate flowsheets later, as data is already prepared and waiting for insertion.
Solution Approach 2:
An AI-based intermediary system acts as a bridge between clinical data generation and EHR documentation. This intermediary automatically transcribes speech, segments information, matches it to flowsheet rows using similarity search, and prepares structured data, freeing nurses from manual data navigation and entry.
2Reliability
If nurses manually document patient data in flowsheets, then data can be recorded, but inaccuracies increase due to cognitive burden
Solution Approach 1:
The system enables self-service documentation by automatically transcribing clinical speech, structuring the information, and populating flowsheet rows without requiring nurse intervention for data extraction. The AI system serves itself by matching transcribed data to appropriate flowsheet categories using similarity search algorithms.
Solution Approach 2:
The manual cognitive process of nurses reviewing, interpreting, and manually entering data is replaced with an automated AI system that performs speech-to-text conversion, information segmentation, and intelligent matching to flowsheet rows, eliminating cognitive burden and improving accuracy.
3Adaptability or versatility
If comprehensive flowsheets with thousands of rows are maintained, then all possible patient data can be captured, but determining relevant rows becomes complex and time-consuming
Solution Approach 1:
Instead of requiring nurses to understand the entire comprehensive flowsheet schema, the system applies local quality by using AI to identify and match only the specific subset of rows relevant to each clinical situation. The similarity search algorithm focuses on local relevance rather than global schema knowledge.
Solution Approach 2:
An AI intermediary layer sits between the comprehensive flowsheet schema and the documentation process, automatically determining row relevance through similarity search. This intermediary handles the complexity of schema navigation, allowing nurses to work with simplified interfaces while maintaining access to comprehensive data capture capabilities.
Data Source
AI summary
A method, computer program product, and computing system for flowsheet population by voice. Dictation or conversation is transcribed from voice to text. The keys and values relevant to the documentation are extracted from the transcript and used to populate the flowsheet by selecting a subset of rows and examples relevant to the transcription of the information from a plurality of rows in the flowsheet and based on the transcript, each row corresponding to a key and a value for the flowsheet and extracting an instance of information from the transcript of information associated with a key from the subset of rows by processing a prompt including at least a portion of the transcript of information and the subset of rows with a generative artificial intelligence (AI) model using retrieval augmented generation (RAG).