Artificial intelligence (AI) is entering general practice (GP) through documentation tools, digital assistants, and diagnostic support systems. Even though AI is a powerful prediction technology with impressive potential for health care applications, the key question is how AI tools will be designed and whether clinicians will be able to use them in ways that improve care.
New evidence from a countrywide survey experiment in Denmark shows potential challenges of AI adoption in general practice. Around one-third of general practitioners are not comfortable following AI predictions even if they contain meaningful information on disease risk extracted from data on patients’ medical histories.
Those who do use the AI-generated information use this information differently than information generated by a well-known diagnostic test that is comparable in quality to the AI tool. These findings point to the importance of how diagnostic information is presented and explained for achieving the potential of new prediction technologies.
Therefore, the challenge is not the development of the technology alone, but the implementation so that it is adopted in practice and generates value for physicians and patients. To achieve this, efforts to advance implementations should involve empirical evaluations of how AI-generated information is communicated to clinicians, accounting for the diagnostic tools and routines already in place, for organizational and behavioral components of clinical practice, and how their interaction affects treatment outcomes.
