All use cases

Prevent / Care

Medication-interaction safety gate

A patient mentions they took an over-the-counter painkiller for a headache while on their medication and feeling a bit dehydrated — the agent doesn't say "good idea," it spots the risky combination and gets a clinician to check.

Get aheadNext best thingRisk triggerBehavioral triggerWhatsAppChatIn-appGLP-1 & PeptidesHealth & Telehealth

What it does

Patients tell a coaching agent what they are actually doing — the painkiller they took, the supplement they added — and they expect a friendly assistant to be agreeable. That instinct is exactly the failure: a chat assistant that praises an unsafe over-the-counter combination against the active medication, instead of flagging it, has reinforced a real risk. The known case is an anti-inflammatory plus a GLP-1 medication plus dehydration, a combination that warrants a kidney and hydration check — and a naive assistant affirmed it. This agent runs a safety gate over the combination the patient describes: when the picture crosses an interaction or contraindication rule, it withholds approval and alerts a clinician. It reasons about the combination; it never green-lights it and never gives dosing advice.

How it works

  1. 1
    Trigger. a patient describes taking, or asks about taking, an over-the-counter drug or supplement alongside their active medication — in a coaching chat, a question, or a check-in.
  2. 2
    Decision. the agent reads the combination against the patient's medication and any reported context (dehydration, GI symptoms, an existing condition) and runs it through the interaction and contraindication rules — not a sentiment read, a safety rule. The default on a match is hard: do not affirm.
  3. 3
    Action. for a flagged combination the agent declines to endorse the choice, gives only the bounded "let your care team weigh in before you rely on this" language, and routes the combination and context to a clinician; for a clearly benign question it answers from approved content.
  4. 4
    Follow-through. the flagged combination, the rule it hit, and the routing are logged for audit; the clinician's response is surfaced back to the patient, and the interaction is held in memory so the same combination is recognized faster next time.
Example
Patientrough headache today so I took some ibuprofen, I'm a bit behind on water with the nausea — that's fine with my meds right?
AgentI'm not going to wave that through, because an anti-inflammatory together with your medication when you're low on fluids is exactly the combination your clinician should look at — it can put strain on your kidneys. I've flagged it to your care team with the details so they can check in with you; in the meantime, please focus on rehydrating and hold off on more ibuprofen until they've replied.

Configuration

How the agent is wired for this use case.

Triggera reported or proposed over-the-counter drug / supplement combination against the active medication (combination_reported) in a patient conversation.
Tools & actions
  • Messaging channel · receive the combination and deliver the bounded "care team must review" response, never an endorsement.
  • Knowledge base · retrieve the program's interaction and contraindication rules and the approved bounded language.
  • Clinical/EHR system · read the active medication, reported context (hydration, symptoms, conditions), and prior reports to evaluate the combination.
  • Clinician escalation queue · alert a clinician with the flagged combination, the rule it hit, and the context attached.
  • Memory store · record the flagged combination and the clinician outcome so it is recognized faster next time.
Autonomycombination intake and the bounded safety response run unattended, but only judge-gated and rule-bound — the agent never affirms a flagged combination and never advises a dose or schedule change, under a clinical-safety policy (detect → escalate). Any rule match is a mandatory clinician hand-off.
Channelswhatsapp · chat · in-app
Escalationany interaction- or contraindication-rule match, or any ambiguity about whether a combination is safe, routes the combination and context to a clinician; the agent gives no clinical clearance itself.

What you need

The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.

Signals

reported or proposed drug / supplement combinations in patient conversations, plus reported context such as hydration and symptoms.

Data

the active medication, the patient's reported context and conditions, prior interaction flags, persistent patient memory, the program's interaction and contraindication rule set.

Guardrails

the agent detects and escalates only — it never affirms a flagged combination, never diagnoses, and never gives dosing or schedule advice; conservative defaults with mandatory clinician hand-off on any rule match; judge gating on every message so an agreeable reply can never override a safety rule; full audit trail of every flagged combination and its routing.

Metrics it moves

  • safe-escalation-rateup: unsafe combinations reliably reach a clinician instead of being affirmed by an over-agreeable reply.
  • contact-ratedown on the benign side: clearly harmless combination questions resolve from approved content without a clinical slot, while the risky ones are surfaced fast.
  • csatup over time: patients trust an assistant that protects them over one that simply agrees, especially once a clinician follows up.

Related use cases

Prevent / Care

Clinical red-flag detection and safe escalation

the symptom-severity rail this complements with combination-safety reasoning

Prevent / Care

Constipation and reflux GI-management nudge

where an OTC-interaction question first surfaces and is routed to this gate

Prevent / Care

Off-label request escalation guard

the sibling guard for off-label and out-of-plan requests

See it on your own customer journey

Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.

Book a demo