Skip to main content

Patient follow-up reminder agent

A sent reminder is not a completed follow-up.

A patient follow-up reminder agent sends approved communications tied to a current care-team request and routes replies or failed contact to responsible staff. It can summarize replies and scheduling preferences while software controls recipients, timing and message content. Werkon would validate the complete follow-up path; clinicians retain care instructions, urgency judgments and decisions about completion.

Start from an active follow-up request

Preserve the governing plan or order, intended recipient, representative authority, purpose, content, timing, channel and staff owner. A proposed plan is not an active communication request. Recheck current instructions, permissions, confidential-contact preferences and relevant record state before preparing or sending a reminder.

Hold or cancel reminders when the source is revoked, replaced, completed or marked not to perform. Stop stale messages after transfer, death, relationship changes or superseding instructions. Keep content minimal and use approved language and accessible formats. Similar records and household relationships cannot establish patient or proxy identity.

Treat delivery and replies honestly

Record generated, queued, provider-accepted, sent, delivered, bounced, opened, replied and acknowledged separately. Delivery cannot prove reading or understanding. Bound retries and frequency, honor current channel restrictions under applicable rules and retain original reply wording, sender, time and uncertainty.

No response does not establish refusal, nonadherence, incapacity or worsening condition. Route ambiguous, sensitive or inaccessible replies to staff. Symptom, medication, result, deterioration, self-harm and emergency language follows approved clinical or urgent routes without model diagnosis, scoring or advice.

Keep responsibility after escalation

An escalation needs a named owner, due time, policy-derived priority and fallback. Record staff acceptance separately from task creation, and do not turn an automated threshold into a clinical priority judgment. Authorized staff record disposition, corrections and whether the original follow-up obligation is complete.

Reconcile failed contact, wrong-recipient corrections, complaints and later care-process evidence with the original request. Task completion is not necessarily completed care, and communication volume cannot establish adherence or health improvement.

Follow-up boundary

Coordinate the reminder without claiming the care is complete.

Follow-up crosses clinical intent, private communication and staff work. Four boundaries preserve which source authorized contact, what the channel did and who remains responsible.

01

Authorized request, patient, recipient, and purpose

Bind one active request to the organization, patient, requester, intended recipient, information provider, purpose, governing order or plan, priority, occurrence window, cadence, approved content, response expectation and stop condition.

Required evidence: Organization and tenant, patient and recipient identifiers, patient-match state, representative authority, request identifier and intent, requester, information provider, order or plan reference and version, reason, priority, occurrence, cadence, content source, response contract, do-not-perform and stop state.

02

Current context, privacy, and channel

Re-read relationship, permission, confidential-contact and communication preference state; minimize the payload; choose the permitted destination, language and accessible format; and block stale, revoked, completed or clinically changed instructions.

Required evidence: Relationship and permission source, consent where applicable, confidential-contact request, channel and destination, language and accessibility needs, sensitive-content class, minimum fields, source versions, last contact, quiet hours, frequency, opt-out, revocation, replacement, completion and review result.

03

Communication attempt, channel result, and response

Prepare the exact message from approved content; persist an attempt before dispatch; capture provider and delivery states without overstating them; preserve patient replies; and route symptoms, urgency, ambiguity or sensitive content to approved human review.

Required evidence: Template and content versions, destination preview, message and attempt identifiers, timestamps, channel provider, accepted, sent, delivered, bounced, opened, replied and acknowledged states, response sender and content, classifier version and confidence, urgent-rule result and handoff.

04

Non-response task, disposition, and outcome

Apply approved attempt and timing thresholds; create an owned staff task without inferring clinical risk; preserve acceptance and completion state; reconcile the follow-up requirement; and require qualified disposition, correction or new instructions before closure.

Required evidence: Attempt count, due and escalation times, threshold version, non-response state, task identifier, owner, priority source, accepted and work states, staff contact and evidence, disposition, follow-up requirement status, correction, complaint, incident, completion owner and observed process outcome.

Request-to-disposition path

Recheck authority before every contact and closure.

A follow-up request can become stale between creation and delivery. Each stage checks the sources that can revoke, replace or complete it.

  1. 01

    Activate one approved follow-up

    Resolve the patient and recipient, bind the requester and current order or plan, confirm the request intent, purpose, priority, timing, content, response and stop contract and reject incomplete or unauthorized instructions.

    Owner
    Clinical, care-coordination, privacy and records owners
    Evidence
    Request and intent, patient and recipient, requester and information provider, governing source and version, reason, occurrence window, priority, content reference, response expectation, expiration, do-not-perform and activation decision.
  2. 02

    Qualify current patient and channel state

    Re-read patient match, relationship, representative authority, confidential-contact request, communication preference, language, accessibility, prior attempts and current clinical-source status; stop or hold changed, revoked, completed or unsafe work.

    Owner
    Patient-service, privacy, identity and clinical-source owners
    Evidence
    Patient and representative state, permission and preference versions, destination, language and format, sensitive class, quiet hours, frequency, opt-out, previous attempts, source status, replacement, completion, contradiction and hold or stop reason.
  3. 03

    Prepare and send minimum content

    Render the approved template and exact source instructions, preview recipient and destination, validate links and tokens, write the attempt, send once and record each channel state without converting transport evidence into patient understanding.

    Owner
    Communications, privacy, security and channel owners
    Evidence
    Template, source fields, rendered content digest, disclosure review, recipient and destination, credential and token scope, attempt and idempotency keys, queued time, provider acceptance, sent, delivery, bounce, open and correction events.
  4. 04

    Capture response or create an owned task

    Preserve the reply and sender context, classify only for permitted routing, escalate urgent or clinical content under fixed rules and send ambiguous cases to staff; if no response reaches an approved threshold, create a task with owner and due time.

    Owner
    Patient-service, nursing, clinical-escalation and task owners
    Evidence
    Response time, sender identity, channel and content, routing candidate and confidence, urgent-rule result, human handoff and acceptance, attempt threshold, non-response status, task type, owner, due time, priority source and escalation path.
  5. 05

    Reconcile and close with qualified evidence

    Track task acceptance and work, record staff contact and disposition, update only authorized source fields, fence late replies, correct wrong-recipient or stale-content errors and close the follow-up only when the responsible owner confirms its completion condition.

    Owner
    Care-coordination, clinical, records, privacy and quality owners
    Evidence
    Task states and history, staff action, evidence viewed, patient response, disposition, new order or instruction, follow-up source status, authorized update and receipt, late-message handling, correction, complaint, privacy incident, final owner and completion evidence.

Authority map

Separate communication mechanics, reply assistance, and clinical responsibility.

A model can route a reply candidate. It cannot establish the need for follow-up, interpret symptoms or decide that silence means a patient is safe.

01

Deterministic communication software

Software owns tenant and identity checks, source versions, request state, exact timing, minimum templates, destination rules, quiet hours, frequency, attempts, idempotency, delivery events, task thresholds, permissions, audit, late-signal fencing and correction history.

  • Patient, request, message, attempt, response and task identifiers
  • Occurrence, due, expiry, cadence, quiet-hour and escalation calculations
  • Permission, preference, do-not-perform, source-status and threshold checks
  • Send, delivery, response, task, correction and closure receipts
02

Bounded AI assistance

Models can extract scheduling preferences, classify a response for approved routing, summarize source-linked context or draft accessible message and staff-task language, but every output remains uncertain and unable to diagnose, prioritize clinically or close care.

  • Scheduling, administrative and clinical-review routing candidates
  • Preference, availability and missing-information candidates
  • Source-linked patient-response and task-summary candidates
  • Clarification, reminder, acknowledgment and handoff drafts
03

Human clinical and operational authority

Qualified people own follow-up purpose, instructions, timing, priority, patient and representative ambiguity, sensitive disclosure, symptom interpretation, urgent action, exceptions, task disposition, completed-care judgment, correction and complaint response.

  • Request, order, plan, clinical-priority and stop decisions
  • Patient match, proxy, privacy and confidential-contact decisions
  • Symptom, result, medication, urgency and care judgments
  • Escalation, disposition, closure, correction and incident authority

Follow-up components

Build a communication and task ledger, not a reminder counter.

The clinical request, channel provider, patient response and staff task have different owners. Four components keep their states separate and reconcilable.

01

Patient, authority, and follow-up-request registry

Bind organization, patient, requester, intended recipient, information provider, representative context, active relationship, consent or applicable permission to the exact request, intent, purpose, order or plan, priority, occurrence, content and stop condition.

Operating contract: Matched demographics are not certain identity, relationship is not proxy authority, proposed request is not active order, active request is not permission for every channel, priority code is not independent clinical judgment and a revoked or replaced source must stop old work.

02

Content, channel, and communication ledger

Version source instructions, minimum template, language, accessibility format, confidential destination, channel, links and tokens; record message attempts, provider acceptance, sent, delivery, bounce, open, response, acknowledgment and correction.

Operating contract: Generated is not queued, queued is not sent, accepted by provider is not delivered, delivered is not read, opened is not understood, reply is not verified clinical fact and one successful channel must not override a confidential-contact restriction.

03

Response, routing, and escalation ledger

Preserve reply sender, time, content and source; bind routing model and rules, uncertainty, urgent indicators, human handoff, attempt cadence, no-response threshold, escalation reason, task owner, due time and priority source.

Operating contract: No response is not refusal, delayed response is not nonadherence, symptom text is not diagnosis, model classification is not triage, escalation threshold is not clinical severity and creating a task is not staff acceptance or action.

04

Task, disposition, and outcome ledger

Track requested, received, accepted, in-progress, on-hold, failed, canceled, completed and entered-in-error work; preserve staff actions, disposition, source updates, late replies, correction, privacy incidents, complaints and observed follow-up outcomes.

Operating contract: Accepted task is not started, completed task is not necessarily completed care, recorded contact is not patient comprehension, patient response is not adherence, workflow closure needs the defined source evidence and technical completion must not be presented as a health outcome.

Delivery path

Prove one follow-up through non-response and staff closure.

Start with one approved follow-up type whose source, content, timing, reply handling and completion condition are already owned.

  1. 01

    Choose one bounded follow-up

    Select one recurring communication request with a named clinical and operational owner, stable approved content, explicit timing and stop conditions, manageable urgency, current manual baseline and staffed exception route.

  2. 02

    Map authority, sources, and channels

    Inventory patient and representative paths, request and governing care sources, confidential-contact and preference rules, templates, languages, accessible formats, provider receipts, reply channels, task owners and closure evidence.

  3. 03

    Build state-safe contact and escalation

    Implement current-source rechecks, minimum content, destination preview, durable attempts, bounded frequency, exact channel states, reply capture, fixed urgent routing, non-response thresholds, owned tasks and late-signal fencing.

  4. 04

    Test privacy and care failure

    Exercise ambiguous identity, unauthorized proxy, changed destination, revoked request, stale instruction, quiet hours, duplicate send, channel outage, false delivery, clinical reply, language or accessibility failure, no response, late reply and unowned task.

  5. 05

    Release narrowly and reconcile outcomes

    Begin with a bounded cohort and staff oversight, compare contact and workload with the current process, reconcile every message and task to the follow-up source, inspect access differences and rehearse correction, complaint and shutdown.

Release controls

Six controls before a follow-up can contact a patient.

A reminder can expose private information or delay needed staff attention. These controls keep source authority, disclosure and escalation current.

Every follow-up has a current authoritative request
Bind patient, recipient, requester, intent, purpose, order or plan, priority, occurrence, content, response and stop conditions; recheck current status; and stop draft, held, revoked, replaced, completed, entered-in-error or do-not-perform work.
Recipient and destination are verified for the action
Resolve patient identity without merging ambiguity, establish representative authority separately, use the current permitted destination, expose it for review where required and block cross-patient, cross-tenant and stale-contact disclosure.
Messages carry minimum approved content
Render exact source instructions through versioned templates, language and accessible format; apply confidential-contact and sensitivity rules; preview destination and payload; protect links and tokens; and keep model-generated clinical language out.
Delivery evidence is never overstated
Record generated, queued, provider-accepted, sent, delivered, bounced, opened, replied and acknowledged states from their owners; use bounded retries; and never convert transport evidence, silence or model classification into comprehension or clinical disposition.
Clinical and urgent replies reach qualified people
Preserve patient words, apply fixed escalation rules, make no diagnosis or advice, route symptoms and sensitive ambiguity promptly, show handoff ownership and acceptance and keep emergency information available through the approved path.
Non-response creates owned work, not conclusions
Use approved timing and attempt thresholds, assign a named task owner and due time, expose aging and failure, fence late signals and require qualified staff evidence to close, correct or replace the underlying follow-up requirement.

Outcome evidence

Measure resolved follow-up obligations, not reminders emitted.

Higher message volume can increase noise and privacy risk. Evidence must show whether approved follow-up reached the right person or became owned staff work and was reconciled correctly.

Baseline

  • Follow-up types, requesters, recipients, purposes, priorities, occurrence windows, channels, languages, accessible formats, attempt policies and stop conditions
  • Current time and human effort from request through contact, response, clinical handoff, non-response task, disposition, correction and closure
  • Current stale requests, wrong destinations, failed delivery, duplicate contact, opt-out conflicts, unclassified replies, late messages and unowned tasks
  • Current response, staff contact, unresolved obligation, privacy incident, complaint, access difference and observed care-process outcomes

Outcome evidence

  • Correct request, patient, recipient, destination, content, timing and stop handling against authoritative sources by follow-up and exception type
  • Time and human effort to achieve verified response, accepted staff task, qualified disposition, correction and reconciled closure
  • Minimum-disclosure, confidential-channel, delivery-state, duplicate-send, urgent-routing, non-response ownership and late-message results
  • Observed contact, response, staff action and follow-up completion against current practice with non-contact reasons, access differences and confounders visible

Guardrails

  • Patient mismatch, unauthorized representative, cross-tenant or wrong-destination disclosure, excessive content, insecure link and ignored confidential-contact request
  • Stale or revoked request, invented instruction, wrong timing, duplicate or excessive contact, inaccessible or untranslated message and opt-out conflict
  • Provider acceptance called delivery, delivery called understanding, silence called refusal, reply called clinical truth, model classification called triage and missed urgent handoff
  • Unowned or aging task, task completion called completed care, late message after stop, hidden correction, unresolved complaint and communication activity presented as health outcome

Fit test

Use this pattern when the request and staff disposition can both be observed.

Good reason to begin

  • One follow-up type has a named clinical requester and operational owner, current source record, approved minimum content, explicit timing, response expectation and stop condition.
  • Patient and representative identity, confidential-contact state, channel events, replies, tasks, staff actions and completion evidence can remain separate linked records.
  • The organization has staffed urgent and clinical handoffs, bounded attempt policies, accessible alternatives and qualified correction and complaint paths.
  • Contact, response, task aging, staff effort, unresolved obligations and observed process outcomes can be compared without causal health claims.

Resolve before beginning

  • The requester, authoritative instruction, intended recipient, approved content, timing, confidential channel, escalation owner or completion condition is undefined.
  • The process cannot distinguish request, message attempt, provider acceptance, delivery, response, task acceptance, staff action and care completion.
  • Success is defined by messages sent, clicks or calls avoided without privacy, accessibility, failed delivery, staff workload, unresolved follow-up and outcome evidence.
  • The agent is expected to invent follow-ups, diagnose or triage replies, message through unsafe channels, infer refusal from silence or close clinical work autonomously.

Source basis

Sources behind the control model.

  • 01

    Health Level Seven International

    HL7 FHIR Release 5 CommunicationRequest resource

    FHIR R5 version 5.0.0 remains the current published release. The trial-use CommunicationRequest resource records a request to convey information and distinguishes status, intent such as proposal, plan, directive or order, priority, do-not-perform state, channel, subject, requester, recipient, information provider and occurrence. It does not prove patient identity, representative authority, legal permission, clinical correctness, source freshness, actual sending, delivery, understanding, completion or compliance.

  • 02

    Health Level Seven International

    HL7 FHIR Release 5 Communication resource

    The current published FHIR R5 Communication resource represents a record of information conveyed and supports preparation, in-progress, not-done, on-hold, stopped, completed, entered-in-error and unknown event states plus sender, recipient, sent and received context. It does not define a messaging provider's delivery semantics, prove the intended person received or understood content, validate clinical truth, grant communication authority or establish a care outcome.

  • 03

    Health Level Seven International

    HL7 FHIR Release 5 Task resource

    The current published FHIR R5 Task resource supports healthcare workflow work queues and state histories including creation, readiness, in-progress work, completion, failure, cancellation, suspension and resumption, with wait and completion timing available for improvement work. A task record does not prove staff acceptance, clinical priority, completed care, correct disposition, legal compliance or a favorable patient outcome; local workflow and authority still apply.

  • 04

    United States Department of Health and Human Services

    HIPAA Privacy Rule guidance on appointment reminders and safeguards

    HHS states that the HIPAA Privacy Rule permits covered providers to communicate with patients about healthcare, including reminders, while advising reasonable privacy safeguards, limited answering-machine content and accommodation of reasonable confidential-communication requests. This US guidance applies within covered contexts and does not decide other jurisdictions, patient or representative identity, the permitted content of every follow-up, channel security, delivery, understanding or implementation compliance.

[ WORKFLOW / SYSTEMS AUDIT ]
THE FIRST ENGAGEMENT

Start with one real workflow

A Systems Audit is the usual starting point. If the opportunity is already clear, we can move directly into a focused build.

Show Us the WorkflowStart with the free automation readiness checklist

OBSERVEQUANTIFYDECIDEBUILD

Continue with the work

Choose the next useful path.

Stay with the operating question. The technology can wait until the work is clear.

01

Explore governed AI agents

Compare other bounded agent patterns and the evidence, authority and recovery contracts they require.

Open path02

Design healthcare administration

Frame patient access, follow-up and communication around protected information, qualified care authority and measurable practice outcomes.

Open path03

Govern customer communications

Coordinate purpose, permission, source truth, approved content, channel delivery, replies, handoffs and corrections across one accountable conversation.

Open path04

Control the follow-up workflow

Connect authorized requests, current state, guarded communication, owned exceptions, staff tasks, receipts and end-to-end reconciliation.

Open path