Industry: Healthcare — Digital Health / Diabetes Management
Location: North America & UK (multi‑site rollout)
Services: Integration Engineering, Interoperability (FHIR/HL7), Backend, DevOps, QA, Compliance

Technologies: FHIR R4/R5, HL7 v2 (ORU^R01), SMART on FHIR (user & backend services), Epic Connection Hub, Oracle Health APIs, Kafka, PostgreSQL, Redis, Kubernetes, Terraform, Keycloak, Python/TypeScript, Grafana/Prometheus, S3/Parquet, dbt

Client

A scale‑up in chronic condition management expanding beyond diabetes. The platform offered CGM‑driven coaching and outcomes analytics and needed to embed results directly into clinician workflows in Epic and Oracle Health (Cerner) to win payer/provider contracts.

Business Challenge

The client’s previous EHR integrations averaged 16–20 weeks per site. Variability in site capabilities (read‑only vs partial write‑back), partial FHIR coverage, and legacy HL7 v2 feeds led to rework, duplicate records, and stalled go‑lives. Without write‑back of outcomes (e.g., Time‑in‑Range, hypoglycemia events), clinicians had to leave the EHR, and payer teams challenged auditability.

Key pain points:

• Inconsistent write paths across sites; some read‑only, some with limited flowsheet writes.
• Fragmented feeds (FHIR + HL7 v2) → duplication and reconciliation issues.
• TIR math drift from unit mismatches (mg/dL↔mmol/L), sensor warm‑ups, and backfills.
• No runtime way to adapt to site‑specific throttles, quotas, and filtering rules. 

Solution

We delivered a reusable integration accelerator: a capability‑driven EHR broker, a standardized outcomes engine for TIR/TBR/TAR, and a clean, auditable path from vendor CGM data to EHR Observations.

What We Built

• Capability Matrix per Site — A runtime config that switches behavior among read‑only, partial write‑back, and full write‑back; controls bulk/export availability, rate limits, and patient‑context quirks.
• EHR Broker — Unified interface for Epic and Oracle Health with SMART (user and Backend Services), FHIR R4/R5, and HL7 v2 fallbacks (ORU^R01). Includes idempotent message keys and duplicate suppression.
• Outcomes Engine — Canonical TIR/TBR/TAR computation with device‑specific warm‑up exclusions, short/long gap policies, timezone normalization, and explicit unit provenance (mg/dL↔mmol/L).
• Observation Model — Standardized FHIR Observation for aggregated windows with effectivePeriod, method, and Provenance linking to raw data. Optional write‑back to flowsheets or notes when allowed.
• Bulk & Cohort Path — SMART Backend Services + Bulk $export for nightly payer dashboards and backfills.
• Observability & QA — Quotas, end‑to‑end latency, error budgets, and a validation suite covering unit/clock/gap edge cases and negative write tests. 

Team Composition

One Engagement Manager, one Solution Architect, two Integration Engineers, one Backend Engineer, one QA Automation, one DevOps, one Security/Compliance Lead, and a part‑time Data Analyst.

Implementation

Phase 1 — Readiness & Design (2 weeks)
We assessed target sites, mapped permitted operations, and populated the capability matrix. Data contracts defined Observation shapes and provenance.

Phase 2 — Broker & Outcomes (3 weeks)
We built the broker with FHIR/HL7 adapters and added the outcomes engine. Synthetic datasets simulated warm‑ups, unit switches, clock drift, DST transitions, and late backfills.

Phase 3 — Site Enablement (1 week/site)
For each site, we registered SMART apps, validated patient‑context flows, exercised negative write cases, and tuned quotas and batch sizes for bulk jobs. Legacy HL7 v2 routes were reconciled with FHIR writes using de‑duplication keys.

Value Delivered

• Integration Cycle Time: reduced from 16–20 weeks to ≈6 weeks per site (median), including registration and smoke tests.
• Clinical Workflow Adoption: outcomes visible in‑EHR; read‑only sites captured clinician intent and queued deferred writes.
• Data Integrity: zero unit‑mix defects in production; provenance attached to 100% of aggregated Observations.
• Contract Velocity: payer proposals strengthened with reproducible TIR reporting and cohort exports; fewer audit follow‑ups. 

Product Overview

Client Goals

Embed outcomes where clinicians work, preserve a single source of truth across mixed feeds, and prove ROI to payers with standardized, auditable metrics.

Implementation Highlights

• Patient‑context SMART launches inside Epic/Oracle Health, with site‑specific filtering validated in UAT.
• Backend Services for nightly cohorts and backfills; $export pagination and retries with manifest verification.
• HL7 v2 ORU mapping maintained for legacy venues; conflict rules prefer FHIR writes when both exist. 

Key Features

• Data Aggregation & Normalization: CGM (Dexcom/Abbott/Medtronic), wearables, and app events.
• Outcomes: TIR/TBR/TAR, hypoglycemia events, coefficient of variation; configurable per population.
• Auditability: end‑to‑end provenance from raw to posted Observation.
• Security & Compliance: TLS, at‑rest encryption, PHI minimization, consent ledger, RBAC. 

Architecture Snapshot

• Ingestion → Normalization → Outcomes → EHR Broker → EHR & Analytics
• Canonical model with explicit units and timestamps; idempotent identifiers across retries; site capability toggles; two‑lane (near‑real‑time + batch) data flow. 

Project Results

• Go‑Live Certainty: predictable per‑site playbook; fewer surprises in production.
• Operational Resilience: automatic degradation to read‑only; queued edits synced when capability enables.
• Scalability: connectors reused to add adjacent conditions (hypertension, obesity) without re‑architecting. 

What’s Next

Next sites onboard via the same capability matrix. Upcoming scope includes device write‑backs (where allowed), clinician annotation APIs, and expanded payer integrations.

Why Jelvix

We combine healthcare integration engineering with compliance‑by‑design. Our accelerators standardize math, provenance, and site behavior so you can ship bi‑directional EHR sync faster — with fewer risks and clearer ROI.

At a Glance

• Industry: Healthcare (Digital Health)
• Location: North America, UK
• Services: Interoperability, Backend, DevOps, QA, Compliance
• Technologies: FHIR R4/R5, HL7 v2, SMART on FHIR, Epic, Oracle Health, Kafka, PostgreSQL, Redis, Kubernetes, Terraform, Keycloak, Python/TypeScript
• Team: 8 specialists

Timeline: First site in ~6 weeks; subsequent sites ~4–6 weeks each