A full-stack healthcare management platform with role-based access for physicians and patients. Built with Next.js frontend and a Node.js + MySQL backend that integrates public health datasets (CMS, Synthea, RxNorm, DailyMed, Orange Book). Physicians get context‑aware insights (coverage, costs, research), while patients get plain‑language guidance, adherence tools, and price transparency.

• Right info, right provider, right time: surfacing formulary tiering, PA/ST/QL, and cost signals alongside the patient’s EMR.
• Evidence & safety: auto‑ingest FDA DailyMed PDFs → extract SPL → clinical flags + HCP summary via LLM (strict JSON).
• Cost transparency: estimate out‑of‑pocket by plan/coverage level, retail vs mail, preferred/non‑preferred.
• Datasets integrated: CMS (formulary, costs), Synthea (EMR), RxNorm (drug IDs), DailyMed (labels), Orange Book (TE codes).

frontend/                  # Next.js 15 (App Router, TS, Tailwind, shadcn/ui, Clerk)
backend/                   # Express API (mysql2/promise, axios, pdf-parse, OpenAI)

• Physicians: patient lists, EMR timeline, prescription flow with safety checks, FDA/clinical feed, notifications, AI summaries.
• Patients: medication manager, recovery timeline, daily checklist, cost & coverage view, appointment tracking.

• Next.js 15 (App Router) · TypeScript · Tailwind · shadcn/ui · Clerk · React Context · lucide-react · date-fns

cd frontend
npm install
npm run dev

Create frontend/.env.local:

NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=your_clerk_publishable_key
CLERK_SECRET_KEY=your_clerk_secret_key
NEXT_PUBLIC_CLERK_SIGN_IN_URL=/sign-in
NEXT_PUBLIC_CLERK_SIGN_UP_URL=/sign-up
NEXT_PUBLIC_CLERK_AFTER_SIGN_IN_URL=/dashboard
NEXT_PUBLIC_CLERK_AFTER_SIGN_UP_URL=/dashboard

Open http://localhost:3000

Project layout mirrors the standard app/ router, components/, contexts/, types/, lib/.

• Language/Libs: Node 18+, Express, mysql2/promise, axios, pdf-parse.
• External APIs: DailyMed (SPL PDFs), RxNav (RxNorm), OpenAI (LLM JSON generation).
• Database: AWS RDS MySQL 8.x. DB name: pharmacy.

• backend/src/server.js – Express app, endpoints, health, graceful shutdown.
• backend/src/utils.js – PDF parsing, DB snapshot builder, LLM prompts and callers.
• backend/src/config/database.js – (not shown) pooled connection helpers: testConnection, initializeDatabase, executeQuery.

PORT=5001
DB_HOST=<rds-endpoint>
DB_PORT=3306
DB_USER=rxuser
DB_PASSWORD=********
DB_NAME=pharmacy
DB_SSL=true

# OpenAI
OPENAI_API_KEY=sk-****************
OPENAI_BASE=https://api.openai.com/v1
OPENAI_FETCH_MODEL=gpt-4o-mini          # for extraction
OPENAI_REASONING_MODEL=gpt-4o-mini      # for comparative report (can be a reasoning model)

cd backend
npm install
npm run dev
# API root
# -> http://localhost:5001/

• CMS (Centers for Medicare & Medicaid Services) — formulary, plan, beneficiary cost files (tiers, PA/ST/QL, coverage levels). Source: data.cms.gov.
• Synthea — synthetic EMR: patients, allergies, medications, observations. Source: synthetichealth.github.io/synthea.
• RxNorm (RxNav) — canonical drug identifiers (RXCUI), properties, brand/generic links. Source: rxnav.nlm.nih.gov.
• DailyMed — FDA drug labels (SPL PDFs). Source: dailymed.nlm.nih.gov.
• Orange Book (FDA) — approved products & therapeutic equivalence.

These power: plan coverage + costs, safe EMR prototyping, consistent drug IDs, label‑based safety parsing, and substitution guidance.

• RDS MySQL 8.x instance, parameter group with local_infile=1, max_allowed_packet=64M, UTF‑8 defaults.
• Security Group: TCP 3306 inbound from dev IP(s) / API host; automated backups enabled.
• Users: rxuser (password auth). Optional IAM DB Auth user if enabled.

allergies
cms_basic_formulary
cms_beneficiary_cost
cms_indication_coverage
cms_plan_info
cms_plan_map
feature_scope_patients
medications
observations
patients
patients_backup
payer_plan_catalog
payer_transitions
payers

-- Enable local infile on client
-- mysql ... --local-infile=1

LOAD DATA LOCAL INFILE '/path/basic_formulary.txt'
INTO TABLE cms_basic_formulary
CHARACTER SET utf8mb4
FIELDS TERMINATED BY '|' LINES TERMINATED BY '\n'
IGNORE 1 LINES;

LOAD DATA LOCAL INFILE '/path/allergies.csv'
INTO TABLE allergies
FIELDS TERMINATED BY ',' ENCLOSED BY '"'
LINES TERMINATED BY '\n'
IGNORE 1 LINES;

If you encounter ERROR 1300 (encoding), convert input to UTF‑8 during load.

• GET /health – uptime & env
• GET / – API index & example routes
• GET /api/test – simple heartbeat
• GET /api/db-test – validates DB connectivity

• GET /api/drug-label.pdf?rxcui=RXCUI Proxy DailyMed SPL PDF by RxCUI.
• POST /api/evidence/digest-rxcui Body: { rxcui, patientId } Flow: RxCUI → DailyMed PDF → text → EMR snapshot + coverage → LLM JSON (flags + HCP summary + coverage fields).

• GET /api/coverage?patientId=...&rxcui=... Returns { covered, tier, priorAuthorization, stepTherapy, quantityLimit, quantityLimitAmount, quantityLimitDays }.
• GET /api/cost?patientId=...&rxcui=...&daysSupply=30&coverageLevel=Initial%20Coverage&channel=RETAIL&preferred=1 Returns estimated out‑of‑pocket based on plan/channel/preference.

• POST /api/alternative Body: { rxcui } Finds disease from cms_indication_coverage, enumerates alternative RxCUIs, extracts SPL JSON for each, and returns a comparative study (similarity % + doctor_note) via LLM.

Note on Views: Some sample queries reference v_patient_formulary, v_drug_plan_coverage, and v_drug_cost_matrix. If you are not using SQL views, replace these with equivalent JOINs across cms_*, payers, and payer_transitions. (You can also add lightweight views later for readability/perf.)

• pool – mysql2/promise pool using env‑driven config.
• parsePdfToText(buffer) – PDF → raw text (nulls stripped, trimmed).
• getPatientSnapshotAndCostSignals(patientId, rxcui) – returns { core, allergies, meds, coverage:{ plan, formulary, estimatedCost } }.
• buildSPLPrompt(pdfText) – extractor prompt: outputs concise SPL JSON (drug_name, indications, MoA, etc.).
• buildPrompt({...}) – clinical decision prompt: flags (high/medium/info) + HCP summary + coverage fields; strict JSON.
• buildReportPrompt(results, inputRxcui) – comparative study across alternatives (similarity % buckets + short doctor_note).
• callLLM(prompt, model) – OpenAI Chat Completions; extracts only the JSON object from response text.

MIT

Built by the CuraRX team using CMS, Synthea, RxNorm, DailyMed, and FDA Orange Book data.