Synthetic US Healthcare Dataset — 21M+ Rows, 10 Relational Tables (CSV + Parquet)

Description:

 A production-quality synthetic US healthcare dataset built for machine learning,

 analytics, and data engineering practice. Entirely fictional — no real patients,

 providers, or clinical events.

 ── WHAT'S INCLUDED ──

 10 relational tables, 2.1 million+ rows total:

 • patients — demographics, insurance, BMI, smoking status, chronic condition count

 • providers — doctors/NPs/PAs with specialty, credentials, and NPI

 • facilities — hospitals, clinics, urgent care centres

 • encounters — visit records with ICD-10 diagnosis, charges, and encounter type

 • diagnoses — primary and secondary ICD-10 coded diagnoses per visit

 • medications — prescriptions with drug name, dosage, route, and frequency

 • lab_results — LOINC-coded tests with real numeric values and reference ranges

 • vitals — BP, heart rate, temperature, O2 saturation, BMI per encounter

 • procedures — CPT-coded procedures with duration and charge

 • insurance_claims — billed, allowed, paid amounts with denial reasons

 ── WHY IT'S BETTER THAN FREE ALTERNATIVES ──

 Most free healthcare datasets are a single flat table with 6 conditions and

 5 medications. This dataset has:

 ✓ 30+ ICD-10 coded conditions

 ✓ 18 LOINC-coded lab tests with real numeric values (not just "Normal/Abnormal")

 ✓ 20 CPT-coded procedures

 ✓ 60+ medications with dosages, routes, and frequencies

 ✓ Clinically correlated data — hypertensive patients have higher BP readings,

  diabetics have elevated HbA1c and glucose, COPD patients have lower O2 sat

 ✓ Comorbidity clustering — diabetes patients also tend to have hypertension

  and hyperlipidaemia

 ✓ Age-adjusted insurance — Medicare for 65+, Medicaid for children

 ✓ Realistic dirty data — ~8% missing insurance, ~5% missing phone numbers,

  ~3% still-admitted encounters

 ── FORMAT ──

 Delivered as a ZIP containing:

 • CSV (universal — works in Excel, Python, R, SQL)

 • Parquet (columnar — ideal for Spark, DuckDB, BigQuery)

 • 25-row preview CSVs for every table

 • schema.md — full column reference for all 10 tables

 • LICENSE (MIT)

 ── GREAT FOR ──

 • Healthcare ML model training and benchmarking

 • EHR system prototyping and testing

 • Data engineering pipeline development

 • Analytics dashboards and BI tool demos

 • University coursework and research

 • Interview prep and portfolio projects

 ---