Bridging the gap between scientists and clinicians: addressing collaboration challenges in clinical AI integration

Bridging the gap between scientists and clinicians: addressing collaboration challenges in clinical AI integration

Abstract

This article explores challenges for bridging the gap between scientists and healthcare professionals in artificial intelligence integration. It highlights barriers, the role of interdisciplinary research centers, and the importance of diversity, equity, and inclusion. Collaboration, education, and ethical AI development are essential for optimizing AI's impact in perioperative medicine.

Introduction

Artificial intelligence is transforming how healthcare tasks are performed by enabling artificial agents to process environmental inputs and execute intelligent actions, often enhancing or replacing human interventions. While definitions of AI keep evolving, recently it has been defined as 'systems that display intelligent behavior by analyzing their environment and taking actions - with some degree of autonomy to achieve specific goals'.

Recent advancements in machine learning algorithms and foundation models have significantly increased AI's capabilities and accessibility, mainly through natural language processing, multimodal AI models, efficient transformer architectures, and self-supervised learning techniques, enabling more personalized AI assistants and innovations. As a result, AI is rapidly evolving to become a cornerstone in modern healthcare, offering tools for diagnostics, predictive analytics, interventions, care, and workflow optimization.

In perioperative medicine, AI can enhance patient care by optimizing drug dosing, and predicting complications such as hypoxemia or hemodynamic instability. Moreover, AI can be a powerful tool to train safely new generations of anesthesiologists and surgeons. However, successful AI integration into clinical settings depends not only on technological advancements but also on seamless collaboration between AI scientists and healthcare professionals.

As such, introducing a new AI technology within healthcare is more than just adding a technical system, it impacts the distribution of tasks, professional identity and use of resources.

Consequently, AI scientists and healthcare professionals bring their own distinct expertise to the table: AI scientists excel in algorithm development and data modeling, while healthcare professionals focus on patient safety, clinical workflows, and ethical considerations. Inherent differences in their professional views and key performance indicators may lead to communication challenges and misaligned priorities.

This editorial examines the key barriers to interdisciplinary collaboration, explores strategies to enhance cooperation, and discusses the role of DEI (diversity, equity and inclusion) in creating ethical AI tools.

Challenges in interdisciplinary collaboration

AI scientists and physicians operate within vastly different intellectual frameworks and their priorities often diverge.

AI scientists focus on machine learning algorithms, data processing, and computational efficiency, often working in spaces far removed from clinical environments. Many of the challenges they tackle have AI

solutions that work "most" of the time, building on data where there are correct ("right") and incorrect ("wrong") datapoint assignments. In contrast, anesthesiologists prioritize real-time decision-making, patient safety, the nuances of human physiology, and ethical dilemmas. Many of the challenges they tackle, do not have a "right" or "wrong"- but each option comes with advantages and disadvantages that might be assessed by different people in different ways and / or differently by the same person over time.

These differences can lead to misunderstandings when translating clinical needs into technical specifications. For instance, an AI scientist might design a predictive model for intraoperative complications without fully understanding the dynamic and high-stakes nature of perioperative medicine, as well as the ultimate consequences of these complications. Additionally, causal relationships between different factors might not be well captured in the AI model, leading to spurious correlation. These considerations might lead to predictions that are not adequate for the complexity of the clinical task. Conversely, physicians may struggle to grasp the limitations of AI systems, such as the reliance on high-quality data or the risk of overfitting. This might lead to unrealistic expectations and disappointments on the side of the healthcare professionals, when these AI tools are developed in a vacuum. An example of this might be the reliance on AI for predicting operating room turnovers without keeping in mind human factors and limitations.

Bridging this knowledge gap is essential to ensure that AI tools are both technically sound, clinically relevant, and can be integrated into healthcare as a complex socio-technical system.