Study: Electronic health record data could improve the accuracy of clinical AI

The authors of the article emphasize that traditional AI models in healthcare mainly analyze lab results, diagnoses, prescriptions, and other indicators of a patient’s condition. However, a physician’s behavior in the system — from clicks and transitions between sections to the response time to notifications and the appointment of additional studies — reflects the course of clinical thinking and decision-making. Incorporating this data will allow algorithms to better understand the context of treatment and more accurately predict outcomes.

In a one-year cluster randomized trial of 60,000 hospitalizations at two medical centers, integrating digital traces into an early warning system for deteriorating patients reduced in-hospital mortality by 35.6%, shortened length of hospital stay by 11.2%, and reduced the risk of sepsis by 7.5%.

Taking such data into account opens up opportunities not only for improving predictive models, but also for analyzing the impact of technologies on practice. It shows how doctors use algorithms, how their workload and decision-making routes change. This allows us to identify weak points in interfaces, prevent staff overload, and adjust the organization of work.

At the same time, the researchers acknowledged that implementing the new approach is fraught with challenges. Data on physician interactions with electronic systems varies widely across providers, is noisy, and does not capture aspects of work that occur outside the digital environment. To overcome these limitations, standards for maintaining activity logs, unified vocabularies, and data cleaning technologies are needed.

Combining behavioral cues captured in EHR interaction data with traditional patient data, the researchers say, opens the door to clinical AI that is not only more accurate but also more contextual—closely aligned with real-world health care practices.

While researchers from the US and China propose improving algorithms by analyzing digital traces of doctors’ work in EHRs, the journal Digital Medicine puts forward the idea of ​​introducing a new medical specialization – “algorithmic consultant”. It is assumed that such specialists will help doctors correctly select and interpret AI models, control their implementation and application in clinical practice, which should increase the reliability and safety of digital solutions in healthcare.