Home/Med Management

IVs Are Not Enough

Electronic medication management system (EMMS): click functions 1 to 9, or Play.
See AI MedAgent in action ↗
Pharmacy Nursing unit Patient room · 4 West 12 hand-off hand-off Pharmacy workstation IV room · compounding Order entry Nurse station Dispensing cabinet monitor idle ZZTEST, Ruth · 78 · synthetic patient
Swipe the scene sideways. It follows the action.
AI MedAgent · this patient Runs alongside the apps each person already uses. Advisory only.

Why the IV is not enough

The field’s founding study of adverse drug events traced where the errors behind preventable harm began, and where they were caught.

56%
of the errors that led to preventable harm began when the drug was ordered.[4]
48%
of those ordering errors were caught before they reached the patient.[4]
0%
of the errors made at administration, the last step, were caught.[4]

Australia has a name for the whole of this, from prescribing through dispensing to administration: an electronic medication management system, or EMMS.[11] In the United States the same pieces are bought separately and have no single name. The Australian name is worth adopting here.

Event-driven alerts versus patient-level reasoning

Almost everything in a hospital today is event driven. A value crosses a threshold, a rule fires, a message goes to whoever the rule was written for. The rule is the same for every patient in the building. A reasoning layer starts from the individual patient and evaluates where she is heading.

Event driven

  • Fires when a single number crosses a single line.
  • The same line for every patient, set for the population.
  • Knows nothing about what else is running, what was held, or what she takes at home.
  • Alerts late, and often for patients who do not need it.

Reasoning, one patient at a time

  • Reads the trend across sources: the pump, the flowsheet, the labs, the orders, the home list.
  • Knows this patient: her kidneys, her anticoagulant, her allergy, what changed today.
  • Says why, traceable to the source, and says what it could not check.
  • Sends one advisory to the one person who can act on it. That person decides.

A decline that never crosses the alarm threshold

Mean arterial pressure over six hours, synthetic patient, illustrative values. The dashed line is a typical low-pressure alarm setting.
0.63
How well one widely deployed, population-based sepsis model separated septic from non-septic patients at one health system (1.0 is perfect, 0.5 is a coin toss).[12]
67%
of the sepsis cases it missed entirely.[12]
18%
of every hospitalized patient triggered one of its alerts anyway.[12]
Limits. This is one model validated at one health system in 2021, not a finding about all population models. Patient-level reasoning has not yet been shown in a trial to perform better at the bedside. That trial remains to be done.

Notes and sources

Every number on this page has a source below. The workflow itself is drawn from the author’s forty years in medication management technology.

    About this page. It is a concept, not a product or a product specification. The patient is synthetic and every clinical value is illustrative. The screens are drawn for this page and are not Epic or any vendor's. Rover is Epic's mobile app for nurses and other clinicians; prescribers more often use Haiku and pharmacists work in the pharmacy application at a workstation, so each person is shown in their own surface.[13] AI MedAgent is a research demonstration of a patent-pending method. Advisory only. Not a medical device. Not for clinical use.