Our current design-prototype is made through Figma, design-based and low-fidelity
Home page where every patient interaction is recorded as a note by clinical professionals.
Each note can contain unstructured notes or notes structured by the AI.
Version History:
Each note has a version history for reviewing unstructured notes post AI-alteration.
Facilitates easy reference to past patient interactions.
Interaction:
Create new notes or open old notes by clicking on database entries.
Displays an unstructured note, often written quickly after urgent patient interactions.
Note could contain spelling mistakes and shorthand.
AI Feedback:
Clickable option to get AI feedback for structuring notes, fixing errors, and ensuring clinical standards.
Helps in better referencing and form-filling by making notes organized and accurate.
Adelaide provides feedback to elevate notes to clinical standards.
Examples:
Prompts for missing information (e.g., blood pressure measurement).
Corrects mistakes (e.g., confusion between ‘vertigo’ and the correct term).
Corrected Notes:
Shows a corrected and organized set of notes generated by Natural Language Processing.
User Interaction:
Users can make further edits and corrections.
Ensures accurate and complete clinical documentation efficiently.
Accessible by clicking the ‘Fill Forms’ button.
Extensive database of all possible forms needed in the EHR.
Example:
Users can select specific forms like vital information forms.
Example Vital Records Form with fields populated by Adelaide.
Auto-population:
Adelaide uses information from notes and EHR to fill fields (in green).
User Interaction:
Users can populate or edit fields not auto-filled by AI (in gray).
Learning Capability:
Adelaide remembers manually populated fields for future forms, enhancing efficiency and accuracy over time.
Completion:
Users save and exit once satisfied, marking the form ready for review, billing, and insurance processing.
Our current prototype is design-based and low-fidelity: there are a few key steps and improvements before our tool is functional.
Speech Recognition: Transcribe audio input from spoken into a text-based unstructured note at the first step of our workflow
Entity Recognition: Distinguish between key elements in the note, such as symptoms, patient, medications, etc.
Natural Language Understanding: Extract relevant information from the note, such as a patient's condition
Text Summarization: Organize unstructured notes into structured notes
Error Detection: Identify ambiguities or potential error's in user's notes and ensure notes uphold clinical standards
There are many advanced NLP models that have similar functionalities to Adelaide in the market today. For example, Nuance's DAX Copilot is an AI assistant that automatically converts multi-party conversations into specialty-specific, clinical summaries that are available in Epic, a commonly used EHR. *Discussion into similar solutions are also presented in Team and Topic
Relevant to Form Filling Pages: The tool must be able to integrate into EHRs with different formats and levels of restrictiveness.
Customize to each version of EHR: Various Epic versions, other EHR systems.
The current design of Adelaide only provides support for notes organization and summarization as well as feedback regarding the notes content and clinical documentation standards. The NLP model can be tuned to also provide assistance in clinical-decision making.
Clinical Diagnosis: AI-based patient diagnosis based on symptoms or observations presented in the notes or in the EHR
Diagnostic Tests or Procedures: recommendation
Recommendations for Patient Care: suggest different treatment options or medication dosages or regimens