Requirements analysis for an AI-based clinical decision support system for general practitioners: a user-centered...

General practitioners often encounter diagnostic uncertainty due to non-specific patient symptoms. Current tools may not fully support diagnostic workflows in primary care. Prior research has shown that decision support systems can aid diagnosis, but their design often overlooks the specific needs of primary care physicians. This gap motivated a study focused on user-centered design for AI-based systems. No prior work had resolved how to integrate AI into daily diagnostic routines. The study aimed to bridge this gap by analyzing GP workflows. Existing literature lacks detailed insights into GP-specific requirements for such systems. This paper addresses that by focusing on user needs in primary care. The findings aim to guide future system development in this area.

The study aimed to identify user requirements for an AI-based clinical decision support system tailored for general practitioners. The goal was to improve diagnostic accuracy and efficiency in primary care. The project, called SATURN, focused on developing a diagnosis support system for German GPs. The study used a user-centered design process to ensure relevance to daily practice. It sought to understand current workflows and digital tool usage in cases of uncertainty. The purpose was to derive requirements for a smart physician portal. The study aimed to inform the design of an AI system that aligns with GP needs. The findings would guide mockup development and system engineering.

The study involved an iterative process with five general practitioners. First, interviews explored current workflows and digital tool use. Second, a workshop collected and prioritized tasks for an ideal system. A task model was developed based on these findings. User requirements were derived from the identified tasks and subtasks. The process included analyzing data entry, result evaluation, and communication needs. The system was expected to handle patient data entry and result review. Features like differential diagnosis probabilities were proposed. Communication with colleagues and data transfer from patient records were emphasized.

The study identified several GP-specific user requirements for an AI-based system. Tasks included data entry, result evaluation, and communication with patients and colleagues. The system should prioritize atypical patterns of common diseases. It should display differential diagnosis probabilities and ensure transparency. Marking ruled-out diagnoses was suggested as a key feature. Communication with colleagues via the platform was strongly recommended. Direct data transfer from electronic patient records was emphasized. These findings formed the basis for mockup development. The results highlight the need for a transparent and user-friendly design. The system should support decision-making while fitting into existing workflows.

The study derived essential user requirements for an AI-based diagnosis support system. These requirements form the foundation for system design and mockup development. The findings emphasize the need for transparency and user-friendliness. The system should prioritize common disease patterns and differential diagnosis probabilities. Communication features and data integration were highlighted as important. The results suggest that the system should align with current GP workflows. The study supports the development of a tool that addresses primary care needs. The authors propose that these requirements guide future system engineering.