Prescription safety
96%of medication alertsare overridden.
PMC systematic reviewThe check already exists. Every record system flags interactions, and almost all of it is dismissed, because almost all of it is noise. Being ignored is the problem worth solving.
- Dismissed
- Read
- Changed the order
Illustration. The proportion comes from published override research, not from the log of any one hospital.
The size of it
Four figures, and not one of them is ours.
Accu-Rx has no deployment and no outcome study, so there is no measurement of it to put on this page. What follows is published research on what getting medicines wrong costs, each figure carrying the name of whoever counted it.
64.1
passive alerts per intensive care patient, per day. Only 4.5 percent are genuine emergencies.
J. Patient Safety1.3M
patients a year experiencing harm related to their medicines.
WHO, 2024$42B
estimated global annual cost of medication errors.
WHO, 2024$7.3B
estimated annual US cost of preventable adverse drug reactions.
WHO, 2024Worth an interruption
The pairing is rarely the question. The decision is.
Detection is not the hard part and has not been for years. Every record system already matches drug pairs, and the matching is mostly right: three of the four pairs below are real interactions, and all three are noise at this bedside. What separates the fourth is that something is still left for the prescriber to decide.
- 01
Is there anything to do about it?
A pairing the prescriber can do nothing about at this hour is a keystroke, not a warning. How serious an interaction is in general, and whether there is a decision in front of this prescriber, are two different questions. Only the second one earns an interruption.
- 02
Has the record already answered it?
Most flagged pairs are answered somewhere in the chart already: by the dose that was chosen, by the reason the two drugs are together, by the hours between them. Asking a question the record has settled is how a warning turns into furniture.
- 03
Was it asked yesterday?
The same pair, the same doses, raised on Monday and answered on Monday. Raising it again on Tuesday teaches a prescriber to dismiss before reading, and that habit does not stay with the alert that taught it.
- Surfaced
Being orderedCiprofloxacin400 mg IV q12h
Already activeTizanidine4 mg PO, day 3
WhyCiprofloxacin blocks CYP1A2, which is most of how tizanidine leaves the body. Exposure rises several-fold, and the fall in blood pressure and the sedation that follow are why the label calls this pair contraindicated. There is a decision here: confirm it, or pick another antibiotic.
- Held back
Being orderedSimvastatin20 mg PO nightly
Already activeAmlodipine10 mg PO daily
Why notAmlodipine raises simvastatin exposure, and the label answers that by capping the daily simvastatin dose. The order beside it is already at that cap. The interaction is real and it has nothing left to ask for.
- Held back
Being orderedClopidogrel75 mg PO daily
Already activeAspirin81 mg PO daily
Why notTwo antiplatelets together raise the bleeding risk, and every interaction database says so. Here the two together are the treatment, written up with the stent on admission. The alert is arguing with the plan it is reading.
- Held back
Being orderedCalcium carbonate500 mg PO, 11:00
Already activeLevothyroxine100 mcg PO, 07:00
Why notCalcium binds levothyroxine in the gut and less of the dose is absorbed. What the interaction asks for is four hours between the two, and the chart already has them four hours apart.
Three of these four are real interactions. Being real is not the same as being worth an interruption.
Where it runs
Two places, and the second one is the unusual half.
Checking of this kind is normally a module inside a hospital record system and nothing else, which leaves out every clinic that does not have one, and every person trying to keep track of their own medicines.
In the record system
The interaction and allergy check runs inside the record system the prescriber already has open, at the moment the drug is ordered.
On its own
It also stands alone, for the clinics that have no record system and for a person checking their own medicines.
Where it stops
What it will not do.
A check that never refuses anything has not thought about being wrong. These four are written in the negative on purpose. A sentence about what software does not do is one a hospital can hold us to today, with no study and no pilot in between.
It does not decide.
The prescriber writes the order and signs it. What the check produces is something read on the way there, and it has no other standing.
It does not say what is wrong with the patient.
That is a different question and a different piece of software. This one reads a prescription, not a person.
It does not stand in for the pharmacist.
A pharmacy review reads things a check on pairs of drugs never sees. The two overlap. Neither one is the other.
It does not measure itself.
There is no deployment behind this page and no outcome study, so there is no figure of ours anywhere on it. Every number here was counted by somebody else and says whom.
Talk to us
Bring a week of your own override log.
We will read it with you, pair by pair: what fired, how often the same pair fired twice at the same doses, and how many were answered the same way every time. That conversation needs no pilot and no integration, and it will tell you more about your own noise than a demonstration would.
Email us