Can AI Really Save Healthcare Professionals Time?

A doctor opens the medical record of their next patient.

They know the patient’s name and the reason for the visit. Now they need to find out what has happened since their last appointment: the latest lab results, a hospital discharge report, current medications, a specialist consultation, perhaps a recent change in prescription.

All of this information exists.

The challenge is finding it, reviewing it, and quickly identifying what will be useful for the consultation.

This is precisely the kind of task where artificial intelligence is beginning to find its place in hospitals.

Part of a healthcare professional’s day is spent in front of a screen

Healthcare professionals do much more than spend time with patients.

They need to review and update medical records, write reports, search for information, prepare consultations, document care, and share relevant information with other professionals.

These tasks are essential to the functioning of a hospital. They also take time.

And as a patient’s medical record grows, so does the amount of information that needs to be reviewed.

For a patient who has been followed for several years, a healthcare professional may need to navigate through consultations, prescriptions, lab results, examinations, letters, and reports from different departments.

AI can help navigate this information differently.

Reading several pages in seconds

Let’s take a simple example.

A patient returns to the hospital six months after their last consultation.

During that time, several things have happened: a blood test, a change in treatment, an appointment with a specialist, and a visit to the emergency department.

Instead of manually searching for each of these elements, a system using a language model could prepare a summary:

“Since the last consultation: emergency department visit on June 12 for X, treatment Y modified, latest blood test performed on August 4, cardiology consultation on August 21.”

The healthcare professional still has access to the original information and can verify it.

The benefit is straightforward: reducing the time needed to find and organize information.

The same approach can be applied to many use cases: summarizing a medical record before a consultation, finding a specific piece of information in a patient’s history, preparing a report from documented information, or facilitating handovers between healthcare professionals.

Less writing, more reviewing

Documentation is another area where AI can help.

After a consultation, part of the work involves documenting what happened: observations, decisions, prescriptions, planned follow-up…

Tools using speech recognition and language models can already produce a first structured draft of a report from a conversation or notes.

The healthcare professional then reviews, corrects, and validates the document.

The same principle can be used to prepare certain letters, summaries, or administrative documents.

This gives the professional a structured first draft that they can then review, correct, and approve.

Saving a few minutes on one task may seem modest. Repeated several times a day, across multiple professionals and departments, those minutes can add up to a considerable amount of time.

AI needs to know where to look

Asking AI to summarize a medical record requires it to understand the information contained within it.

And not every medical record looks like a perfectly organized database.

One piece of information might be buried in a report written several years ago. Another might be found in a lab result. A third in a prescription. Some information is structured, while other information exists as free text.

To produce a genuinely useful summary, AI needs access to information that is sufficiently complete, properly contextualized, and usable.

This is one of the major challenges of medical AI: model performance alone is not enough. The quality and structure of the data it can access matter just as much.

Saving time does not mean delegating decisions

It is also important to distinguish between different levels of AI use.

Summarizing twenty pages of a medical record is not the same as recommending a treatment.

Finding a patient’s latest lab result is not the same as interpreting that result.

Preparing a report is not the same as validating it.

The closer AI gets to a medical decision, the greater the requirements for reliability, traceability, validation, and accountability.

For the use cases discussed here, time can be saved on tasks surrounding the medical decision itself: finding, organizing, summarizing, and documenting information.

The healthcare professional remains responsible for interpretation and decision-making.  

What about hallucinations?

This is one of the best-known limitations of language models.

An LLM can produce information that sounds perfectly plausible while being incorrect.

In a medical context, a convincing answer is obviously not enough.

A hospital tool must therefore allow healthcare professionals to trace information back to its source: which lab result? Which prescription? Which report? Which date?

A useful summary should not become another opaque layer between the doctor and the medical record.

Instead, it should make the underlying information easier to access.

This traceability will be essential in building trust in these tools.

More time for what?

The first gains may come from tools that make better use of information that is already available: finding an element of the medical record more quickly, preparing a summary, structuring a report, or facilitating handovers.

For these use cases to be genuinely useful, models need to rely on medical data that is structured, contextualized, secure, and integrated into the tools healthcare professionals already use.

At Galeon, this is also why we work to structure data directly within the electronic health record. The more usable the information is, the more opportunities there are to develop relevant AI applications for healthcare professionals.

The first benefit of AI in hospitals may ultimately be very practical: making information easier to find, understand, and use.

‍