How to build an app to facilitate the process of decision making for placement of a leadless pacemaker or tranvenous pacemaker , both with Conduction System Pacing (CSP) in a patient with already an Aveir VR.
A clinical decision support web application for pacemaker placement decisions in patients with existing Aveir VR devices
Below is a blueprint that teams at device companies or cardiology practices have found workable when they set out to turn clinical know-how into a web-based decision-support app. I’ve broken it into five layers— clinical logic, data, software architecture, workflow/UX, and compliance—so you can decide where to prototype first.
| Decision gate | Key inputs (from Aveir VR interrogation, EHR, or manual entry) | Branches & typical thresholds |
|---|---|---|
| 1. Is additional atrial support needed? | Sinus node dysfunction, symptomatic pauses, episode log (Aveir VR), HR variability | Yes → Consider Aveir AR or a transvenous atrial lead. No → go to 2. |
| 2. Is physiologic (conduction-system) pacing indicated? | Expected RV pacing burden > 40 %, LVEF < 50 %, QRS > 120 ms, existing bundle-branch block | Yes → CSP (His-bundle or LBB area) via transvenous lead or leadless upgrade once available; No → go to 3. |
| 3. Venous access / infection risk? | Central vein occlusion, prior CIED infection, dialysis fistula | High risk → stay leadless; Low risk → either pathway |
| 4. Adequacy of current Aveir VR | Capture threshold trend, R-wave amplitude trend, projected longevity (Abbott battery estimator) | Poor metrics OR battery < 18 m ≈ “replace/upgrade” bucket |
| 5. Need for CRT-class resynchronization? | NYHA class II–IV, LVEF ≤ 35 %, wide LBBB | Yes → CSP+CRT strategy or CSP alone (per 2023 HRS/LAHRS)(PMC, PMC) |
Why start with a rules table, not ML?
The 2023 HRS/APHRS/LAHRS physiologic-pacing guideline and the 2025 ACC/AHA Appropriate-Use Criteria emphasize explainability in CDS. A clear decision tree lets you map citations directly to UI labels.(JACC, PMC)
FHIR resources
Patient, Condition, Device, Observation, Procedure cover 95 % of what you need.
Device telemetry ingest
Option A (today): CSV export from Merlin™ or PaceView™ → secure upload.
Option B (future): Direct REST pull once Abbott exposes a FHIR-endpoint.
Clinical rules engine
Open-source openCDS or a lightweight Python engine (e.g., business-rules) keeps logic versionable and testable.
┌────────────┐ CDS Hooks "patient-view"
│ React/TS │ ───────► API Gateway ─► Rules svc (FastAPI + openCDS)
│ Next.js UI │ ◄───────┴──────────── ◄─┐
└────────────┘ FHIR R4 store │ (PostgreSQL)
▲ │
└─────🔒OAuth2 / SMART-on-FHIR────┘
Frontend: React (or Next.js) + Tailwind for a responsive dashboard that can be embedded in Epic/Cerner sidebars.
Backend: FastAPI (Python) or NestJS (Node) with pydantic schemas → keeps clinical rules in plain JSON.
Deployment: Docker Swarm or Kubernetes on AWS (Shielded AMIs) with encrypted S3 for file uploads.
Audit trail: Append-only event log (e.g., AWS QLDB) to satisfy FDA SaMD traceability.(FDA)
| Widget | Purpose |
|---|---|
| “Traffic-light” recommendation card | Green = Stay with Aveir VR; Yellow = Add Aveir AR; Red = CSP upgrade. Shows why in one sentence plus citation link. |
| Trend panel | Spark-lines for capture threshold, R-wave, impedance—flagging ≥ 50 % rise from baseline. |
| What-if simulator | Slider to adjust projected pacing burden → instantly updates battery longevity and recommendation tier. |
Keep the CDS “non-directive”: surface guidelines and patient-specific facts; let the operator click “accept” or “override” with a mandatory comment—this is what the FDA calls “human-in-the-loop” non-device CDS.(FDA)
| Phase | Objective | Evidence artifacts |
|---|---|---|
| Analytical validation | Unit-test each rule against 30 de-identified cases. | Rule test pack, version log |
| Clinical validation (pilot) | 3-center prospective study, n = 100. Measure concordance with EP board decisions. | IRB protocol, GCP logs |
| FDA filing (if you cross into device territory) | 510(k) or Q-submission for SaMD class II. | Predicate comparison, cybersecurity risk plan |
Convert the decision table above into a DMN diagram—Camunda Modeler works well and can be executed directly.
Mock the UI in Figma and run a 15-minute hallway usability test with two electrophysiologists.
Stand up a FastAPI skeleton with one /recommendation endpoint that ingests a sample FHIR Bundle and returns the traffic-light tier; swap in the real rules engine later.
Would you like sample Python code for that /recommendation endpoint, or a deeper dive into the regulatory submission checklist?