The use of artificial intelligence in hospitals is booming. Over nine out of ten hospitals (93 percent) are applying AI to a greater or lesser extent. As yet, monitoring of AI use is not keeping pace with this. Barely one in three hospitals (29 percent) knows how frequently they test, train and calibrate AI models in order to prevent errors.
All this emerges from the "AI Monitor Hospitals" conducted annually by M&I/Partners. For the study, the consulting firm surveyed 43 hospitals. The number of hospitals not yet using AI has dropped dramatically from 33 percent to 7 percent. In many hospitals, AI use is still in its early stages. Nearly seven out of 10 hospitals are experimenting with AI, an increase of 26 percent. Yet the number of implementations has also grown by 6 percent to 60 percent.
Generative AI
This advance is largely due to the advent of generative AI. More than half of hospitals are using large language models (LLMs), such as Chat GPT. More than half of hospitals (57 percent) are using generative AI for automatic transcription of call records, document summarization and text generation, among other things.
Radiology continues to lead the way
Diagnostics, more specifically image diagnostics, remains the domain where AI is most widely used. More than six out of ten hospitals (63 percent) use AI for diagnostic purposes. Logically, radiology thus remains the specialty with the most AI implementations and AI experiments. Following in second place comes - also like last year - AI to support prognosis (23 percent). Strong risers are: AI serving communication (21 percent) and transfer and dismissal (19 percent).
Reduce workload
Unlike before, hospitals are no longer citing quality of care, but rather reducing workload as the primary driver for using AI. Nearly nine out of 10 hospitals (89 percent) hope to use AI to ease the burden on staff. Furthermore, 93 percent of respondents expect AI to have a positive impact on the employee experience over the next five years.
Buy or develop?
The survey also addresses the choice to self-build, co-create or purchase AI applications. Of the 24 hospitals that indicated they have implemented AI applications, 64 percent have purchased them. In addition, 23 percent indicated they developed AI applications through in-house development, possibly with outside support. The remaining 14 percent have AI applications that were created in co-creation with other hospitals. According to M&I/Partners, the latter proves that AI is no longer only for large hospitals: "By using external vendors and collaborations between hospitals, AI has also become accessible to smaller hospitals."
Anchoring
Growing AI maturity is also reflected in the allocation of people and resources. 42 percent of hospitals now have an AI team. This has an average size of 5.7 FTE, with a data engineer, data scientist and medical specialist as permanent members. That said, AI is not yet embedded in strategy and policy everywhere. A quarter of hospitals have a strategy. Half indicate it is under development. At 30 percent of hospitals, there are frameworks, standards and working agreements for the deployment of AI. Forty percent are in the process of developing AI policies.
One in three hospitals (29 percent) know the frequency with which AI models are retested, trained and calibrated to prevent errors due to hallucinations and drifting. In 11 percent of hospitals, this does not happen. Of more than half of the hospitals (54 percent), it is unknown if this is happening. "With that, more attention to monitoring of AI models in use seems to be needed," reads M&I/Partners' economical conclusion.
This is an article by Dutch Health Hub. Want to keep up with all the news from the healthcare industry? Then take a look at the hub and sign up for the weekly newsletter.
↗
