A consultant pharmacist covering 800 long-term care beds is reviewing, on a slow week, 8,000 active medication orders. The clinically meaningful question on each one — is this drug, at this dose, in this patient, with these comorbidities, still appropriate? — does not scale by hiring. It scales by structure.
Three sources do most of the heavy lifting for that structure: the STOPP criteria (potentially inappropriate prescribing), the START criteria (potentially omitted prescribing), and the AGS Beers Criteria (drugs to avoid or use with caution in older adults). Together they describe most of what a careful geriatric pharmacist would flag if they had unlimited time per chart.
The problem is they have to live somewhere a computer can actually use them.
What "encoded" means
A rule pack is not a PDF of the criteria. It is the criteria broken down into atomic, testable rules. Each rule has:
- A trigger condition expressed against structured data — RxNorm codes, ICD-10 diagnoses, lab values, age, renal function, length-of-therapy windows
- A severity — informational, advisory, or actionable
- A citation — the specific STOPP/START/Beers rule ID and the source publication
- A recommended action template — phrasing the pharmacist can accept, edit, or reject
When a chart has the structured data needed to evaluate a rule, the rule fires or it does not. When the data is missing, the system records a "data gap" rather than guessing.
Our rule packs cover STOPP, START, and Beers criteria, organized into versioned packs so that when the AGS updates Beers — typically every three years — the prior version stays intact for ongoing reviews and the new version applies forward.
Why this is harder than it sounds
The criteria are deceptively simple to read and deceptively hard to encode. Two examples:
STOPP F1: "Sulfonylureas with a long duration of action (e.g., glibenclamide, chlorpropamide, glimepiride) with type 2 diabetes mellitus — risk of prolonged hypoglycemia."
To encode F1 you need RxNorm-level resolution on "sulfonylureas with long duration of action" (not all sulfonylureas qualify), an active ICD-10 for T2DM, and exclusion of recent dose modifications that suggest the prescriber is already actively managing the risk. A naive rule that fires on every sulfonylurea + diabetes pair produces too many flags and trains the pharmacist to ignore them.
Beers 2023: "Avoid use of proton pump inhibitors beyond 8 weeks without justification."
The "without justification" clause is the entire game. The rule needs to know whether GERD, Barrett's, or another long-term indication is on the problem list, whether the patient is on chronic NSAID or anticoagulant therapy, and whether the chart contains a documented justification from the prescriber. A flag that fires every time a PPI hits 56 days produces alert fatigue. A flag that fires only when none of the justifications are present produces an actionable finding.
A good rule pack is judged by what it doesn't flag, not by what it does.
The pharmacist still decides
A flag is a candidate intervention. It is never a recommendation in the patient's chart until a pharmacist has reviewed it. The workflow:
- The rule pack runs against a resident's chart on a defined cadence (daily for new admits and order changes, monthly for stable residents).
- Candidate interventions appear in a triage queue, ranked by severity and recency.
- The pharmacist opens each candidate, reviews the rule citation, the patient data that triggered it, and the draft recommendation language.
- The pharmacist accepts, edits (with the edit captured), or rejects (with a reason code).
- Accepted recommendations are sent through the standard MRR workflow with the rule provenance attached.
The final clinical document the facility sees carries the pharmacist's voice and the pharmacist's name. The rule pack is the screening layer.
What this changes for the consultant pharmacist
The consultant pharmacist is not freed from clinical judgment. They are freed from clerical labor — from scanning long medication lists for the obvious flags, from manually cross-referencing the Beers update PDF, from being the only line of defense against a thousand small rule-violations that any thoughtful clinician would catch given time.
The rule packs do the scan. The pharmacist does the medicine.
A panel of 1,500 LTC residents, reviewed thoughtfully every month, is a workload that has historically required a team. With the rule pack doing the first pass, a single experienced consultant can cover it with their judgment intact and their head clear. That is what scale should mean — not less attention per chart, but the attention focused where it matters.