Future prospects
Successfully translating AI proof of concepts into clinical practice
remains pivotal for fully realizing AI’s impact.
While practical guidelines and best practices are emerging in medical
AI, they are not always adhered to and require frequent reassessment due
to the pace at which the AI field is moving forward. When implementing
AI, it is strongly recommended to verify available guidelines to ensure
applications are reliable and provide meaningful outcomes. We here
propose a set of minimal requirements for good practice in AI (Table 1)
based on published guidelines of the FDA, literature on best-practice
model development in biomedicine, or expert-based checklists for
developing and reporting algorithms (e.g., STARD-AI, TRIPOD checklist,
and awaited TRIPOD-AI adaptation).
In the allergy and immunology field, research beyond proofs-of-concept
is relatively scarce, let alone meaningful clinical applications. We
provide an expert outlook on noteworthy AI trends. Firstly, the
ever-increasing accessibility, automation, and transferability ofML tooling are expected to drive AI adoption further, enabling
non-specialized researchers to apply novel techniques. Secondly, we
expect an increase in the use of unstructured data. Innovations
such as AI-based image analysis, NLP, and generative AI are at the
forefront of academic efforts in computer science while being
underutilized in our field. For clinicians, an AI clinical assistant
akin to readily available ‘home assistants’, could quietly listen in on
consults and subsequently support in documentation in EHRs, diagnosis,
and therapy suggestions. Clinical solutions that leverage speech
recognition are entering the market, aiming to improve the clinical
workflow and efficiency, although adoption and showcases of tangible
impact are still limited. Thirdly, the emerging trend of multi-modal
learning can open new research avenues by integrating multiple data
sources and modalities in a singular analytical approach, hereby
creating more holistic models and insights.
The largest future impact from AI is expected when current
proofs-of-concept are translated successfully to clinical practice. The
US and the EU are making steps towards developing AI and algorithm
regulations, to facilitate updates and improve privacy, security and
transparency.
The developers of algorithms play a role in clinical translation, and
clinicians would need to adapt to the integration of AI within
healthcare. While most AI systems are designed as a support mechanism
rather than a replacement, it will change their work and role. Clinician
training in the fundamentals of AI is needed to gain trust in these
systems and work with them effectively. One of the common concerns
regarding AI is that these systems will replace humans in their
installment. While many studies position their analytical solution in a
head-to-head comparison with humans, most clinical applications are
designed as decision-support tools that strengthen and assist experts in
their profession rather than replacing them41. Lastly,
we foresee further developments in dynamic learning systems, which
continuously evolve based on clinical usage. Such approaches are rare,
and FDA-approved tools are generally ‘locked’, referring to a fixed
algorithm state. The FDA is working on an action plan to better assess
and support such applications.
In conclusion, the potential of AI to transform clinical medicine is
evident, but the steps from a proof of concept to clinical applications
are not easily made. Innovations from the field of AI can address many
important open questions in allergy; we anticipate that good future
utilization of AI (Table 1) will deepen our knowledge of disease
mechanisms and contribute to precision medicine in allergy.