Teamwork in healthcare is important to patient safety, including in the area of diagnosis. Consider the following scenario:
Ms. Hopkins is a 72-year-old woman with hypertension presenting to the office for her yearly checkup. On three occasions over the past month, her smart watch alerted her to an “abnormal heart rhythm.” She did not contact your office because she was asymptomatic during the episodes. In addition, she “doesn’t trust the watch’s ability to diagnose heart problems.”
An electrocardiogram is completed during the visit and is normal. Ms. Hopkins notes that her older sister has atrial fibrillation. She would like to know if additional testing should be completed for atrial fibrillation.
The call for adoption of team-based care models began more than a decade ago. In a 2012 Institute of Medicine (IOM) report, Mitchell, et al.,1 defined team-based healthcare as:
...the provision of health services to individuals, families, and/or their communities by at least two health providers who work collaboratively with patients and their caregivers—to the extent preferred by each patient—to accomplish shared goals within and across settings to achieve coordinated, high-quality care.
The 2015 National Academy of Medicine report Improving Diagnosis in Healthcare emphasized the importance of collaboration and teamwork among and between healthcare professionals, patients, and their families to reduce diagnostic errors.2
Definitions of team-based care and descriptions of core members of diagnostic teams generally include humans only, with technology serving as an ancillary tool.2 In recent years, however, significant progress has been made in artificial intelligence (AI) and machine learning (ML) in the context of healthcare. AI is the use of computers to perform tasks that typically require objective reasoning and understanding. ML is a subdomain of AI that involves using computational methods to teach computers to learn from examples.
Some of the most significant healthcare AI/ML advancements have occurred in diagnostics. AI/ML is increasingly demonstrating safety and effectiveness in healthcare as suggested by the growing number of U.S. Food and Drug Administration (FDA) approvals.3 For example, the FDA has approved ML models that can accurately diagnose breast cancer on mammograms, skin cancer on clinical images, and diabetic retinopathy on fundoscopic images.4-6
Incorporating AI/ML into broader clinical practice will undoubtedly affect healthcare teams and the patient-clinician relationship. As AI/ML with varying levels of autonomy becomes more common in healthcare, clinicians and patients will need to learn to effectively team with AI in the diagnostic process.
The diagnostic process occurs within a complex system that includes team members functioning in an interrelated fashion (Figure 1).2 Diagnostic teams are often interprofessional and interdisciplinary.1 However, how AI fits into the diagnostic team and affects the patient-clinician relationship still needs to be better defined. For example, patients and clinicians will need to understand their respective responsibilities when AI is involved in the diagnostic process. Furthermore, AI systems must be developed and implemented in a way that supports the patient-clinician diagnostic team, rather than hindering it.
Figure 1. Work system in which the diagnostic process takes place
Source: Adapted with permission from National Academies of Sciences, Engineering, and Medicine. Improving Diagnosis in Health Care. Washington, DC: The National Academies Press; 2015.
This issue brief provides a framework for patients and clinicians to successfully partner with safe and effective AI when making diagnostic decisions. At times this partnership will involve a tridirectional exchange among patients, clinicians, and AI that is not typically seen with tools and technologies used in healthcare.
Rather than viewing AI as a diagnostic tool to be wielded by human agents, we view it as a member of the diagnostic team capable of understanding, interpreting, reasoning, relating, responding, and ultimately collaborating with clinicians and patients in the diagnostic process.7 The following sections describe the strengths and limitations of the dyadic relationships between patients and clinicians, patients and AI, and clinicians and AI. Finally, using the team-based care framework, we describe the triadic patient-clinician-AI diagnostic team.
