The anatomy of a weak prescribing answer is consistent: it ends at drug and dose, as if the prescription were the decision's last act rather than its first, and the whole apparatus that makes prescribing safe, baseline assessment, early review, surveillance, sick-day rules, stopping criteria, lives in the part the answer never reached. Monitoring is where non-medical prescribers demonstrate the framework's competencies most visibly, and it is also where AI can genuinely help, because a monitoring plan is a structured object, exactly the kind of thing a well-prompted system assembles and a clinician then verifies against the SmPC and guideline the assembly must cite.
The seven-section framework
Before treatment: confirmed indication, contraindications checked against the exact product's SmPC, baseline observations and investigations appropriate to the medicine's known effects, and the counselling conversation that makes the patient a monitoring partner rather than a subject. Initiation: the starting approach and titration logic per current guidance, interaction management against the full reconciled list, and what the patient should expect and report in the first days. Early review: a scheduled check of efficacy and tolerability with the initial laboratory surveillance the medicine's profile requires, at the interval guidance specifies rather than the interval the diary permits. Maintenance: the long-term review rhythm, ongoing surveillance for the adverse effects that emerge late, and periodic re-confirmation that the indication persists, the check that prevents this year's review inheriting last year's assumptions. Intercurrent illness: sick-day and temporary-withholding rules where they apply, the section most often absent from weak plans and most valuable to patients, because dehydration and acute illness change the risk calculus of many common medicines. Stopping: whether discontinuation needs tapering, what withdrawal or rebound looks like, and the criteria for restart. And escalation: the named circumstances in which the prescriber seeks specialist or medicines-information advice, written into the plan so escalation is a designed step, not a failure.
Stress-testing the framework on real medicine classes
The framework proves itself on the medicines where monitoring is the treatment's other half: renin-angiotensin agents, where renal function and electrolytes anchor the before, early and sick-day sections; SGLT2 inhibitors, whose intercurrent-illness rules are the framework's poster case; methotrexate and lithium, where the surveillance schedule is inseparable from the prescription and shared-care documents join the source list; direct oral anticoagulants, where renal trajectory, interactions and review discipline carry the safety load; and long-term corticosteroids, where the stopping section, tapering, withdrawal, restart criteria, does the heaviest work. In every case the pattern holds: the specific values, intervals and thresholds come from the current SmPC, national guidance and any shared-care protocol, never from a synthesis's memory, and a plan whose every section cites its source is a plan a colleague can check, which is the definition of a good one.
The prompt template, and the verification it demands
The AI use that works: request the monitoring plan in the seven sections explicitly, for the exact product and formulation, for this patient's context named in clinical categories, and require direct links to the relevant guideline sections and the current SmPC beside every section's claims. Then verify by construction: open the SmPC's monitoring and warnings sections and check the plan against them; open the guideline's surveillance recommendations; confirm any local shared-care requirements, which operationally govern; and fill the sections the answer left thin, thin sick-day and stopping sections being the reliable tell of a synthesis that stopped at the easy parts. The finished artefact is doubly valuable: a safer plan for this patient, and, recorded with its sources and reflection, precisely the portfolio evidence the prescribing competencies ask for, the answer-to-learning loop doing its quiet work.
Frequently asked questions
Should the plan be shared with the patient?
The patient-facing translation of it, yes: what will be checked, when, why, and what symptoms warrant contact, in plain language reviewed against the PIL, which is counselling and safety-netting in one artefact.
How do local shared-care protocols interact with the framework?
They populate and override: where a shared-care document specifies the surveillance split between specialist and primary care, its schedule governs the maintenance section, and the framework's job is exposing any gap between it and current national guidance for escalation.
What does AI add if every value must be verified anyway?
Structure, completeness and speed: the seven-section skeleton assembled with candidate content and sources in a minute leaves the clinician's time for the verification and judgement only they can supply, which is the correct division across this entire cluster.
Which section do audits find missing most often?
Intercurrent-illness rules and stopping criteria, consistently: prescribing culture documents starting far better than it documents pausing and ending, and the seven-section skeleton exists precisely to make those absences visible before they matter.