Presentation
Human-Centered Design in AI-Based Clinical Decision Support Systems for Predicting Patient Readmission: Bridging Clinical Practice and AI Interpretability
DescriptionIn healthcare, patient care decisions are critical, complex, and time-sensitive, particularly when predicting the risk of hospital readmissions. As data-driven tools and predictive models become more common, a pressing challenge emerges: how can healthcare professionals trust and effectively use these tools in their daily practice? AI-driven clinical decision support systems (CDSS) are often based on complex algorithms, which despite their increasing accuracy, can be difficult to interpret and understand. The more complex the system, the less likely it is to be transparent, understandable, and consequently, trusted. While there has been a push towards integrating explainable AI (XAI) to make predictions more understandable, current solutions often fall short. Simply providing explanations is not enough, these explanations need to be meaningful and aligned with the needs of healthcare professionals. Trust in CDSS depends not only on the predictive accuracy of these systems but also on how well they fit into the workflows and decision-making processes of clinicians, nurses, and care coordinators.
By understanding healthcare professionals’ workflows and decision-making processes in discharge planning, this study aims to bridge the gap between standard generic XAI outputs and the needs of healthcare providers. This work aims to develop comprehensive design guidelines that ensure the information provided by predictive systems are not only accurate but also meaningful and supportive for their decision making.
This study focuses on collecting data from healthcare professionals—such as physicians, nurses, social workers, and discharge planners—responsible for patient care and discharge decisions Interviews with 20 diverse healthcare professionals provide key insights into how they assess readmission risks and make discharge decisions. The focus is on understanding their decision-making processes, the information they rely on, team interactions, and challenges in managing readmissions. Thematic analysis of the interviews identifies essential data sources and types of data clinicians rely on, and highlights common challenges faced by them, such as lack of data transparency or comprehensive patient information. The findings are then translated into guidelines by mapping the types of data clinicians prioritize and the decision-making challenges they face to tailor XAI features that directly address these needs. From a user-centered point of view, these guidelines specify how AI explanations should be structured and presented, ensuring they are relevant, actionable, and supportive for clinicians. By addressing the specific needs of healthcare professionals, these guidelines aim to bridge the gap between AI-driven CDSS outputs and the requirements of clinical decision-making.
By understanding healthcare professionals’ workflows and decision-making processes in discharge planning, this study aims to bridge the gap between standard generic XAI outputs and the needs of healthcare providers. This work aims to develop comprehensive design guidelines that ensure the information provided by predictive systems are not only accurate but also meaningful and supportive for their decision making.
This study focuses on collecting data from healthcare professionals—such as physicians, nurses, social workers, and discharge planners—responsible for patient care and discharge decisions Interviews with 20 diverse healthcare professionals provide key insights into how they assess readmission risks and make discharge decisions. The focus is on understanding their decision-making processes, the information they rely on, team interactions, and challenges in managing readmissions. Thematic analysis of the interviews identifies essential data sources and types of data clinicians rely on, and highlights common challenges faced by them, such as lack of data transparency or comprehensive patient information. The findings are then translated into guidelines by mapping the types of data clinicians prioritize and the decision-making challenges they face to tailor XAI features that directly address these needs. From a user-centered point of view, these guidelines specify how AI explanations should be structured and presented, ensuring they are relevant, actionable, and supportive for clinicians. By addressing the specific needs of healthcare professionals, these guidelines aim to bridge the gap between AI-driven CDSS outputs and the requirements of clinical decision-making.
Event Type
Oral Presentations
TimeMonday, March 3111:30am - 12:00pm EDT
LocationPier 2/3
Digital Health (DH)