Safer prescribing in hospitals: what a modern EHR changes

The essentials in 30 seconds

Introduction

Medication error remains one of the most closely watched risks in any hospital, and one of the most frustrating, because it so rarely comes from a lack of skill or care. Far more often, the error grows in the gaps: information that is missing, arrives too late, or gets lost as a patient moves from the emergency department to a ward, from one software tool to another, from the hospital back to their community pharmacist. The medication travels through many hands, and every handover is a place where something can quietly go wrong.

This is why medication safety is, at heart, an information problem before it is a clinical one. The prescriber, the pharmacist and the nurse are all competent; what they need is the right information, complete and at the right moment. When the record is fragmented, they are asked to make safe decisions on incomplete data, which is an unfair thing to expect of anyone. When the record is unified and the circuit is traceable, the same professionals catch far more before it reaches the patient.

Galeon builds an AI-native electronic health record (EHR, in French DPI) alongside caregivers since 2016. It is used in 19 hospitals, including 2 university hospitals (CHU), across more than 3 million patient records, on HDS-certified infrastructure aligned with ISO 27001:2022. Securing the medication circuit is therefore not a feature we describe from the outside; it is part of how we design the tool, with the people who use it every day.

One thing to remember: a modern EHR does not remove risk, it removes a large share of its avoidable causes.

How big is the problem, really?

Big enough that the World Health Organization made it a global priority. Its "Medication Without Harm" challenge set the goal of reducing severe, avoidable medication-related harm, precisely because so much of it is avoidable rather than inevitable. The point of that framing is important: we are not talking about the irreducible risk of medicine, but about the share of harm that better information and better processes could prevent.

In a hospital, that avoidable share concentrates at a few predictable points: admission, where a patient's usual treatment has to be captured accurately; internal transfers, where the prescription is re-read and sometimes re-entered; and discharge, where the hospital's changes have to be reconciled with what the patient was taking before. Get those transitions right and you remove a disproportionate amount of the risk, which is exactly where a well-designed EHR earns its place.

Where does the medication circuit break down?

At every transition. Prescription, pharmacist validation, preparation, administration: the information has to flow without interruption for the circuit to be safe. Yet breaks are frequent. A treatment is re-entered by hand from one system into another, and a digit changes. A community prescription is captured incompletely on admission, and an interaction becomes invisible. An allergy noted in one tool never makes it into the one used at the bedside. None of these is a failure of competence; each is a failure of information flow.

The insidious part is that these breaks are usually silent. Nobody sees the moment a piece of information fails to travel, only the consequence, later, when an interaction or a duplication surfaces. That is why the goal is not to add a human check at every step, which only adds workload, but to remove the breaks themselves, so the right information simply stays with the patient throughout the stay.

What safeguards does a modern EHR add?

At its core, a modern EHR brings computerised prescribing with decision support: as the prescriber orders, the system checks interactions, flags dose issues, and takes account of allergies and, where relevant, renal function. Done well, this is quiet and contextual. Done badly, it becomes a wall of pop-ups that everyone learns to dismiss. The whole art is relevance, because too many alerts kill the alert.

Around that core sit the other safeguards that actually change outcomes: pharmacist validation on a shared record, guided medication reconciliation, closed-loop administration that checks the right drug reaches the right patient, and end-to-end traceability. None of these is exotic; what makes the difference is that they live in one system rather than being stitched together across several.

The safeguards that matter most

Not every feature carries equal weight. A few safeguards do most of the work, and it is worth being concrete about them rather than listing everything a brochure can list.

1. Contextual interaction and allergy checking

The system cross-checks the new order against current treatments and known allergies, and surfaces only what is clinically meaningful. The measure of quality here is not how many alerts fire, but how few false ones do.

2. Dose adjustment for the individual patient

Weight, age and renal function change what a safe dose is. A modern EHR brings these into the prescribing moment, so the adjustment happens before the order is placed, not after an incident.

3. Medication reconciliation at every transition

Comparing the current and prescribed treatment at admission, transfer and discharge is the single highest-yield safeguard, because it catches exactly what isolated steps miss. On a shared record it becomes a guided reflex rather than a manual chore.

4. Closed-loop administration

Checking, at the bedside, that the right patient receives the right drug at the right dose closes the last and most dangerous gap in the circuit: administration. It turns the final step from a leap of faith into a verified action.

5. Prioritised, contextual alerts

An alert that fires for everything protects no one, because it trains people to click through. Prioritising alerts by real risk is what keeps the safety net credible over months of daily use.

Why is medication reconciliation the key link?

Because it catches what every isolated step misses. Reconciliation means sitting the patient's actual current treatment next to what is being prescribed, at admission and again at discharge, and resolving every discrepancy on purpose rather than by accident. It is where omitted drugs, silent duplications and dropped chronic treatments are caught before they cause harm.

Done on paper, reconciliation is slow and easily skipped under pressure. Backed by a single shared record, it becomes a guided step: the system lays the two lists side by side, flags the differences, and asks for a decision on each. The work does not disappear, but it moves from reconstruction to judgement, which is where the professional's time is actually worth spending.

Alert fatigue: the paradox of doing too much

There is a failure mode that looks like safety but is its opposite: drowning clinicians in alerts. When every order triggers three warnings, most of them irrelevant, people stop reading them, and the one alert that mattered is dismissed with all the others. More alerts do not mean more safety; past a point, they mean less.

A serious system is therefore judged as much on what it stays silent about as on what it flags. Calibrating alerts to real risk, suppressing the noise, and letting the important warning stand out is harder to build than a firehose of pop-ups, but it is the only version that survives contact with a busy ward.

What about traceability and the closed loop?

A traceable circuit means every step is recorded and attributable, from the order to the administration at the bedside. It is what lets a team understand an incident after the fact, and, more importantly, prevent the next one by seeing where the process bent. Traceability is not bureaucracy; it is organisational memory.

The closed loop takes this further by verifying the final step in real time rather than trusting it. When administration is checked against the order at the point of care, the circuit no longer relies on the last handover going perfectly; it confirms it. That is the difference between hoping the right thing happened and knowing it did.

Bolt-on prescribing module versus safety built into the EHR

A prescribing module bolted onto an existing system can add decision support, but it inherits the fragmentation around it. If allergies live in one tool, the current treatment in another and administration in a third, even a clever module is checking against a partial picture. Its alerts are only as good as the data it can see, and it can rarely see everything.

When medication safety is built into the EHR itself, the checks draw on the whole record at once: allergies, current treatments, results, renal function, administration history. Reconciliation is guided from the same place the prescription is written, and the closed loop closes inside the same system. The lesson mirrors the rest of clinical software: safety is decided by how unified the record is, far more than by the sophistication of any single module added at the end.

Is medication-safety AI automatically "high risk" under the AI Act?

Not automatically, and it depends on the use. Decision support that surfaces an interaction or a dose issue for a prescriber to weigh is not the same as a system that would decide a prescription autonomously. The level of regulatory obligation follows the degree of autonomy and the potential impact on the patient, so a supervised support tool and an autonomous decision-maker do not sit in the same box.

What holds regardless of classification are the fundamentals: transparency about what the support does and does not do, traceability of every alert and decision, human oversight that is real rather than nominal, and strict protection of health data. A medication-safety tool worth trusting is built on those principles from the outset, with the clinician firmly in the loop.

What does it change for the DSI, the pharmacist, the prescriber and the nurse?

For the hospital CIO (DSI)

The priority is unification and interoperability: one record that carries allergies, treatments and administration without re-entry, and that can exchange with community care and pharmacy. The DSI's real lever on medication safety is upstream, in the coherence of the information system and in hosting that meets HDS requirements.

For the pharmacist

Validation happens on a complete, shared record rather than a partial view, and reconciliation is guided rather than reconstructed by hand. The pharmacist's expertise is freed from chasing missing information and refocused on the decisions that genuinely need it.

For the prescriber

Decision support arrives in the flow of prescribing, contextual and prioritised, not as a wall of pop-ups. The aim is that the right consideration surfaces at the right second, so the safe choice is also the easy one.

For the nurse

Closed-loop administration confirms, at the bedside, that the right drug reaches the right patient, turning the most exposed step of the circuit into a verified one. It is support at the sharp end, where the consequences are most immediate.

Limits and challenges to be aware of

• Alert fatigue: poorly calibrated alerts get ignored. Relevance beats volume, always.
• Input data quality: allergies and community treatments must be reliable, because no check compensates for a wrong or missing entry.
• Clinical judgment: the tool equips it, it does not replace it. The prescriber and pharmacist keep the decision.
• Interoperability: without exchange with community care and pharmacy, blind spots remain at the hospital's edges.
• Change management: safety features only work if teams trust and adopt them, which is a human process as much as a technical one.

FAQ

Does an EHR remove medication error?
No, and no honest vendor should claim it. What it removes is a large share of the avoidable causes, such as information breaks, re-entry and lost allergies, so the residual risk is smaller and better understood.

Aren't too many alerts counterproductive?
Yes, and it is one of the most common failure modes. A good system favours contextual, prioritised alerts and stays silent on the rest, because credibility is what keeps the important alert from being ignored.

Is reconciliation just extra time?
It is time invested that avoids reworks and incidents, and a single shared record speeds it up considerably. The alternative, catching the same problems later, costs far more time and carries real risk.

Who stays in control?
The prescriber and the pharmacist. The tool informs and verifies; humans decide and remain responsible, exactly as they should.

Where should a hospital start?
With reconciliation at admission and with the highest-risk drug classes. Those two moves capture a large part of the avoidable risk before any broader roll-out.

What about data security?
The medication circuit handles sensitive health data, so it must run on infrastructure meeting health-data hosting requirements (in France, HDS certification) with strict access control and traceability. Security is a precondition, not a bonus.

In summary

Making prescribing safer is not about adding barriers or piling on alerts. It is about moving the right information to the right person at the right moment, and removing the breaks where information used to get lost. Relevant decision support, guided medication reconciliation, closed-loop administration and an end-to-end traceable circuit do most of the work.

A modern EHR lowers the team's workload and keeps the patient safer, without ever replacing clinical judgement. And, as everywhere else in clinical software, the decisive factor is how unified the record is: safety is built into a coherent circuit far more effectively than it is bolted on at the end. That is the logic of an AI-native EHR like Galeon.

Read next: why clinical AI depends first on data quality.

Sources

• WHO, Medication Without Harm
• Haute Autorité de Santé (HAS)
• ANSM, bon usage du médicament
• Galeon, smart EHR vs traditional EHR