Skip to main content
HEA-02 · Healthcare Providers

Patient Navigation Assistant

Outcome​

Guide a patient through the next steps of their care plan by combining appointments, referrals, benefits, and approved education content into plain-language instructions, escalating clinical questions to the care team.

Portfolio attributeValue
Primary personaPatient
Primary aggregatePatientJourney
Cosmos DB containers2
Grounding dataappointments, care plans, benefits, education content
Development estimateUSD 60-220 monthly dev estimate
Runtime optionsPython FastAPI and C# ASP.NET Core

Productivity hypothesis​

These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.

Impact typeMeasureConservative internal assumptionCustomer measurement
ProductivityRequest-to-approved-response time7-12% reduction(baseline median request-to-response minutes - assisted median minutes) / baseline median minutes * 100
Quality or riskCorrected responses2-5% reduction(baseline corrected or reopened response rate - assisted rate) / baseline rate * 100

Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.

Rationale: Appointments, care plans, benefits, and approved education content support faster patient-navigation responses.

Pilot guardrail: Do not diagnose, prescribe, or make autonomous care decisions; keep patient guidance cited.

Customer validation plan​

  1. Baseline the current workflow with representative users and tasks.
  2. Run the accelerator on the same task mix with human review retained.
  3. Measure request-to-approved-response time and corrected responses with the documented formulas.
  4. Replace internal assumptions with customer-observed results before a business case.

Why Azure Cosmos DB​

This AI solution needs an operational data plane, not only a vector index:

  • Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
  • Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
  • Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
  • Change feed reacts to source and operational updates without polling.
  • Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.

Domain workflow​

Domain workflow

StageActorInput → outputApproval
1. Ask a question about the next care stepPatientPlain-language question in the patient portal or SMS → Classified navigation intentPolicy controlled
2. Resolve the active pathway and due stepsCare plan servicePatient identifier and pathway state → Ordered CarePlanStep list with due windowsPolicy controlled
3. Check coverage and preparation requirementsBenefits agentPlanned service and payer coverage facts → Cost and prior-authorization expectationPolicy controlled
4. Explain the step in approved plain languageNavigation assistantCare step, appointment, coverage facts, education content → Cited plain-language instruction at the patient's reading levelPolicy controlled
5. Escalate clinical or urgent concernsNurse navigatorSymptom report or clinical question → Triage decision and human response to the patientHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
patientJourneys/tenantId, /patientIdPatientJourney, CarePlanStep, AppointmentSlot, BenefitCoverageFact, CareTeamEscalation/patientMessageVectorEmbedding, /patientMessageVector
patientEducationLibrary/tenantId, /contentIdEducationContent/bodyVectorEmbedding, /bodyVector

Query contract​

QueryContainerScopePurpose
Next steps timelinepatientJourneyssingle-partitionAssemble the patient's upcoming steps and appointments in one partition read.
Overdue care gap detectionpatientJourneyscross-partition-boundedBounded cross-partition control-plane query: find steps past their due window so the navigator can intervene. It cannot carry the second partition key because population health outreach is a panel-level view across patients, so it runs as a tenant-scoped scheduled sweep off the interactive path, bounded by steps whose due window closed inside the last 30 days, executed with MaxItemCount paging and continuation tokens.
Prior authorization exposurepatientJourneyssingle-partitionIdentify planned services that need payer approval before scheduling.
Open escalation queuepatientJourneyscross-partition-boundedBounded cross-partition control-plane query: show the care team escalations still awaiting a human response by urgency. It cannot carry the second partition key because the nurse triage queue is shared across the panel rather than per patient, so it runs as a tenant-scoped scheduled sweep off the interactive path, bounded by escalations awaiting response inside the 24-hour response standard, executed with MaxItemCount paging and continuation tokens.

Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.

Azure deployment​

Azure deployment

Clone the source repository, authenticate with Azure, and run:

Set-Location accelerators/healthcare-providers/patient-navigation-assistant
az login
./Deploy.ps1 -Runtime python -Location eastus2

Change python to csharp to deploy the .NET implementation.

Explore the implementation​

Return to all accelerators