From Electronic Records to AI-Enabled Workflows: A Scoping Review of Real-World EHR Information Processing in Hospitals and Implications for AI Readiness in China
This scoping review utilizes an AI-agent screening approach to map the real-world integration of AI-enabled information processing into hospital EHR workflows, highlighting that while process-level feasibility is evident, achieving safe value and aligning with China's AI-readiness conditions requires rigorous human verification, calibration, and interoperability.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Hospitals today run on a vast, continuous stream of information. Every time a doctor writes a note, a nurse checks a vital sign, or a machine processes a lab result, that data enters a digital system known as an electronic health record. For years, the promise of artificial intelligence in medicine was that computers could learn from this data to predict diseases or suggest treatments. But a computer model, no matter how smart, is useless if it sits in a lab while the doctor is busy trying to find a patient's file. The real challenge is not just building a smart algorithm, but weaving it into the daily, messy flow of hospital work so that it helps rather than hinders. This is the difference between a tool that works on paper and a tool that works in a busy emergency room.
A team of researchers from Jiangyin People's Hospital in China set out to map exactly how this integration is happening in the real world. They did not simply ask if artificial intelligence is good or bad; instead, they looked at the specific steps where computers and humans meet. They examined how these systems handle the acquisition of information, the sorting of data, the prediction of risks, and the final writing of medical notes. Their goal was to see if the conditions needed for these tools to work safely actually exist in hospitals, with a special focus on how these conditions compare to the current state of hospitals in China.
The researchers began by gathering thousands of reports from medical journals and public records. They used a systematic method to sort through nearly 2,300 records, looking for studies where artificial intelligence was actually used inside a hospital's electronic record system, rather than just tested on old data. They found that while there is a growing body of evidence showing these tools can be integrated, the benefits are often about speed and organization rather than immediate improvements in patient health. For instance, in several studies, the use of artificial intelligence significantly reduced the time doctors and nurses spent writing notes or searching for information. One study in the United States found that when doctors used an ambient system that listened to conversations and drafted notes for them, the time spent on documentation dropped by nearly 30 percent during a shift. Similarly, a pilot program in Taiwan showed that nurses could cut the time spent on handover notes from over three minutes down to about one minute per patient.
However, the researchers were careful to point out that saving time does not automatically mean saving lives. The evidence they found suggests that while these tools make the workflow smoother, they do not yet prove that they lead to better diagnoses or fewer deaths. In fact, the study highlights that a tool can be fast but still dangerous if it is not set up correctly. The researchers found that the success of these systems depends heavily on how well they are calibrated to the specific hospital using them. A model that works well in one hospital might fail in another if the data is slightly different or if the doctors do not trust the alerts it sends. In one instance, a change in the computer template used to display information caused the accuracy of medical coding to drop dramatically, from nearly 80 percent down to 35 percent, until the system was redesigned and the staff was retrained. This shows that the technology is fragile and requires constant human oversight.
The review also looked at the readiness of hospitals to adopt these technologies, specifically examining the situation in China. The researchers found that while many Chinese hospitals have moved past the basic stage of having digital records, there is a significant gap between simply having a system and having a system that is ready for advanced artificial intelligence. Public data suggests that while a large majority of major hospitals have reached a level where their records are digitized, very few have reached the highest levels of maturity where information flows seamlessly between different departments and supports complex decision-making. The researchers noted that persistent gaps in how information moves between systems, often called "silos," remain a major hurdle. Without a smooth flow of data, even the smartest artificial intelligence cannot function properly.
Ultimately, the study concludes that the path forward is not about waiting for a perfect algorithm, but about building a reliable environment for the tools we already have. The researchers argue that for artificial intelligence to be safe and useful, hospitals must focus on the details of implementation: ensuring the data is accurate, training the staff to trust and verify the system, and having clear rules for what happens when the system makes a mistake. They found that the most successful deployments are those where the technology is treated as a partner in a carefully managed process, rather than a magic solution that fixes everything on its own. For China and hospitals worldwide, the lesson is that the readiness for artificial intelligence is not a single switch to flip, but a layered foundation that must be built step by step, starting with the quality of the data and the clarity of the workflow. Until these foundations are solid, the full potential of artificial intelligence in healthcare will remain out of reach, regardless of how advanced the models become.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.