An antibiotic review can involve a medication from another prescriber, an incomplete allergy list and a chart note that has not been reconciled. A useful EHR alert should make those gaps visible and let the podiatrist inspect the evidence behind it.
This guide explains the workflow demonstrated by the PhiClaw fictional-patient medication-safety preview. It is a software workflow guide, not patient-specific prescribing advice.
Start with a reconciled medication and allergy list
Keep each medication as a structured entry with its ingredient, documented dose, route, whether it is active, and where the information came from. Record allergy substance, reaction and verification separately. Unknown allergy history must stay distinguishable from an explicitly reviewed no-known-allergies result.
Free-text notes and AI-generated summaries are useful context, but they should not silently replace verified medication facts. Stopped entries, duplicate entries and unsupported ingredients need explicit handling.
Make each alert traceable
The DailyMed TMP-SMX label documents interactions with warfarin, methotrexate and dofetilide. PhiClaw's pilot uses version 4, published January 29, 2026, checked October 22, 2026. It stores that identity with each evaluation.
A label-based pilot is narrow coverage, not a comprehensive interaction service. The screen should show the affected pair, review priority, source section and limitations. A no-match result cannot establish that a combination is safe.
Keep clinician review visible
A reviewer should see the saved medication and allergy snapshot, unresolved data gaps and source details before recording a decision. An override needs a substantive reason, clinician confirmation, actor and time. It should preserve the original alert and never silently change a prescription.
When a list changes, a new evaluation should replace the earlier result as the current review target. A source outage or expired snapshot should produce an unavailable result, not a reassuring green check.
Why general AI chat is a separate tool
A conversational answer is not an interaction-screening result. Medication checks need a defined data source and reproducible rules or a validated service. NLM states that RxNav's drug-interaction features were discontinued in January 2024; using RxNorm ingredient identifiers alone does not restore that interaction service.
Prediction needs a measurable outcome and validation
A dashboard count or an AI-generated risk description does not validate a clinical prediction. Define the population, outcome and time horizon first. For this project, the proposed endpoint is a first new diabetic foot ulcer within 12 months among adults with diabetes and no active baseline ulcer.
Before activation, require the named model and version, deployment rights, external validation in the intended population, calibration and discrimination evidence with confidence intervals, a missing-data policy and clinician oversight. PhiClaw currently returns no risk probability for this endpoint.
PhiClaw's current status
Available for fictional demonstration: structured reconciliation, three limited TMP-SMX interaction rules, the specified label allergy scope, missing-data warnings and clinician review/override history in the authenticated demo. These deterministic checks run separately from generative chat.
Still required for real-patient clinical use: broader maintained knowledge coverage and appropriate use rights, independent clinical validation, the exact EHR connection, and the practice's privacy, access and activation approvals. No validated patient-risk model is active. No prescribing, treatment automation, patient outreach or regulatory approval is implied.
Explore the podiatry EHR preview, review EHR integration requirements, or see current pricing.
Bring your medication-review workflow to a demo.
Book a podiatry demo