Based on the EHC replay, this article analyzes how generative AI lightens the administrative burden on physicians to restore time for patient care.
Administrative overload represents a major structural challenge for healthcare facilities today. During his presentation at our event " Data, AI & UX at the service of your Business," Frédéric André, Chief Information Officer (CIO) of the Côte Hospital Group (EHC), detailed an initiative aimed at integrating generative artificial intelligence into the daily work of practitioners. The objective is to automate the production of complex clinical documents, thus freeing up time for patient care teams.
This article reviews his intervention and analyze the methodology used, the technical challenges encountered, and the results observed, distinguishing the reported facts from the perspectives they open up for the hospital sector.
The diagnosis: the doctor's cognitive overload
The initial problem stems from a precise quantitative observation. The studies cited by Frédéric André indicate that physicians spend nearly 50% of their working time interacting with the Electronic Patient Record (EPR). This administrative burden directly contributes to the phenomenon of professional exhaustion, or burnout, observed among healthcare professionals.
Among the administrative tasks, writing the discharge summary is a particular point of contention. This document is mandatory to close the hospital stay and ensure continuity of care with community-based physicians. Its preparation requires a significant effort to synthesize information, necessitating the compilation of the patient's medical history, test results, and treatments administered.
Analysis of this process reveals that the difficulty lies not only in text entry, but also in structuring the dispersed information. Therefore, the added technological value must be found in cognitive synthesis rather than simple word-for-word transcription.
From speech recognition to generative artificial intelligence
The EHC already used conventional speech-to-text technologies. However, this project marks a technological breakthrough by coupling speech capture with large language models (LLMs).
The process described by Frédéric André unfolds in several fluid stages:
- The context: The system automatically retrieves the patient's administrative data (age, gender, hospitalization dates) from the EHR.
- Voice instruction: The doctor dictates the key elements of the stay, without necessarily worrying about the form or perfect order of the sentences.
- Structured generation: AI processes these elements to produce a formatted document, respecting the hospital's standardized structure (anamnesis, evolution, discharge treatment).
This approach demonstrates an evolution in the use of voice. Dictation is no longer used to produce the final text, but to provide instructions and raw material that AI then formats. This allows practitioners to express themselves in a more natural and less constrained language.
The imperative of data sovereignty and security
The deployment of such solutions in a critical environment like a hospital imposes strict regulatory constraints, including the absolute necessity of guaranteeing the confidentiality of patient data.
EHC has opted for an architecture where data never leaves Swiss territory. The use of private instances via Microsoft Azure Switzerland allows the CIO to ensure that information is not used to train global public models. This precaution complies with the requirements of the Swiss Federal Act on Data Protection (FADP) and aligns with European standards such as the GDPR.
This technical vigilance demonstrates once again that the adoption of AI in healthcare depends as much on legal trust as on algorithmic performance. Mastering the data lifecycle remains a prerequisite for any clinical innovation.
User experience (UX) as a lever for adoption
The project's success largely hinges on its user-friendly integration. This new tool must not be an additional application requiring separate authentication. The AI is accessible directly from the doctor's existing interface via a simple button.
The workflow is designed to keep the physician in control. After the AI generates the text, the practitioner must review and validate the document. The AI sometimes makes errors, described as hallucinations, or may omit subtle details. Maintaining human involvement in the validation process ensures medical safety and legal accountability.
This design choice promotes team buy-in. By minimizing the number of clicks and integrating into existing habits, the IT department reduces the friction to change often observed when introducing new digital tools.
Results and observed benefits
Initial feedback from the field highlights both qualitative and quantitative gains. Frédéric André reports estimated time savings of between 5 and 10 minutes per discharge summary, depending on the case. Over the annual volume of a hospital, this represents thousands of potential medical hours that can be reallocated to patient care.
Beyond the timer, the impact on mental workload is significant. The tool relieves the physician of the effort of formatting and wording, allowing them to focus on the clinical content. Furthermore, the quality of the documents improves: the letters are more complete, standardized, and always polite, as the AI does not experience fatigue or frustration at the end of a shift.
Activation guide: the keys to replicating this model
For institutions wishing to embark on this path, the EHC experience makes it possible to identify several actionable guiding principles.
Targeting painful processes
It is helpful to begin by identifying the documents whose drafting is the most time-consuming and standardized. The discharge summary or the operative report are ideal candidates for an initial implementation, as their structure is predictable and their length is significant.
Build a robust "prompt"
The quality of the result depends directly on the quality of the instructions given to the AI (the prompt). Frédéric André explains that developing this prompt system, which defines the AI's role and the expected format, required numerous iterations to achieve a satisfactory result. Investing time in prompt engineering is essential.
Managing change through evidence
The EHC's approach involved working with pilot groups of volunteer physicians. The concrete results obtained by these early adopters then served as a communication tool to convince the rest of the institution. This strategy of dissemination by example facilitates the tool's cultural acceptance.
Anticipating the evolution of uses
Beyond this new tool, new uses are already being considered, notably the use of AI on mobile devices for bedside visits. This suggests a future where AI will become an omnipresent assistant, capable of capturing clinical information in real time, further reducing the gap between care and its documentation.
In conclusion, the EHC initiative conducted with Smile and Synotis illustrates a pragmatic transition to the digital hospital. Generative AI is used as a concrete lever for operational efficiency, framed by strong security requirements, confirming that the technology is now mature enough to assist healthcare professionals in their most demanding tasks.
To learn more, watch the replay or contact us !