Collaborative strategies for deploying AI-based physician decision support systems: challenges and deployment approaches

Abstract

AI-based prediction models demonstrate equal or surpassing performance compared to experienced physicians in various research settings. However, only a few have made it into clinical practice. Further, there is no standardized protocol for integrating AI-based physician support systems into the daily clinical routine to improve healthcare delivery. Generally, AI/physician collaboration strategies have not been extensively investigated. A recent study compared four potential strategies for AI model deployment and physician collaboration to investigate the performance of an AI model trained to identify signs of acute respiratory distress syndrome (ARDS) on chest X-ray images. Here we discuss strategies and challenges with AI/physician collaboration when AI-based decision support systems are implemented in the clinical routine.

Clinical Decision Support Requirements for Ventricular Tachycardia Diagnosis Within the Frameworks of Knowledge and Practice: Survey Study

“…Notably, the accuracy of etiological diagnosis was relatively high, aligning with the lower knowledge demand for an etiological diagnosis ( Figure 3 ). In terms of CDSS needs, the cardiologists favored direct decision support over knowledge support, including explanatory diagnoses and executable evaluation processes, which has also been recognized in recent studies [ 57 , 59 , 60 ].…”

“…Moreover, the cardiologists revealed a preference for automatically prompted relevant knowledge during the diagnostic and therapeutic processes, which can provide more targeted knowledge support ( Figure 2 ). The challenge lies in ensuring that the CDSS accurately identifies the current diagnostic and therapeutic tasks; determines user knowledge gaps; and automatically retrieves, integrates, and presents knowledge support rapidly and accurately [ 57 ]. The results of the practice competence highlighted the need for improvement in the interpretation of diagnostic tests, etiological diagnosis, and diagnostic evaluation, suggesting the need for decision support in these three aspects, which were also highlighted as key clinical reasoning [ 58 ].…”

Development and Testing of a Novel Large Language Model-Based Clinical Decision Support Systems for Medication Safety in 12 Clinical Specialties

“…Widespread adoption of NKB CDSS systems is met with barriers, including lack of transparency, uncertainty relating to the evidence and lack of trust in the system, or disruptions to the clinical work ow that adds time to routine clinical practice. 36,37 Till date, ML-based CDSS are fairly narrow in applications, most being disease domain speci c. LLM grounded with contextual knowledge present with various advantages over ML-based models, including the ability to integrate and process vast amounts of varied data types including unstructured clinical texts, easy updating of clinical knowledge corpus without the need for explicit retraining, offer explanations in natural language that are more comprehensible to human practitioners. We demonstrate that a RAG-LLM CDSS performed equally across different clinical disciplines and medication classes, being agnostic to disease state.…”

AI-Driven Clinical Decision Support Systems: An Ongoing Pursuit of Potential

“…Biases inherent in training data or algorithmic design can lead to erroneous predictions and contribute to inequities in patient outcomes. Mitigating such bias, as emphasized by Obermeyer et al and Ferrara, requires diligent and continuous data curation, algorithmic auditing, and the implementation of fairness-aware machine learning techniques to ensure that AI-CDSS uphold principles of fairness and equity across diverse patient populations [ 13 , 27 , 28 ].…”

“…In essence, while AI holds immense promise for transforming CDSS, addressing the inherent challenges, and charting future directions require a concerted effort from stakeholders across the healthcare ecosystem. By acknowledging technical limitations, addressing workflow alignment and attitudinal barriers, implementing strategies for successful adoption, and embracing interdisciplinary collaboration, we can navigate the complexities of AI-CDSS and unlock its full potential in revolutionizing healthcare delivery [ 27 ].…”