ESCORT Synthetic FHIR R4 Dataset: 500 European Chronic-Care Patients with Risk Tiers and Wearable Time-Series

This dataset contains 500 synthetic European patient records in HL7 FHIR R4 format, designed to demonstrate chronic-disease cohort management, AI-based Early Warning Score (EWS) classification, tele-medicine pathways, and emergency-dispatch triggering. The data are fully synthetic and contain no real patient information. Each patient is a self-contained FHIR Bundle (type=collection) combining Patient, Encounter, Condition (SNOMED CT), MedicationRequest, Observation (LOINC/UCUM vitals and labs), Immunization, Procedure, and — for a subset — a wearable Device resource. Chronic-disease prevalence, age/sex structure, and geographic spread across 15 European countries are calibrated against Eurostat / WHO European Region burden-of-disease estimates. Key features: every patient carries a risk tier (high / moderate / stable) with vitals sampled to reflect that tier, giving a ground-truth label for evaluating EWS classifiers; 60 patients include 14 days of wearable time-series (daily steps, sleep duration, resting heart rate, heart-rate variability) with deterioration patterns in the high-risk group; a patient-level manifest (MANIFEST.csv) and a device-assignment lookup (DEVICE_ASSIGNMENTS.csv) support cohort selection and pipeline integration. Coding systems: SNOMED CT (conditions, medications, vaccines, procedures, device type), LOINC (observations), UCUM (units), and HL7 v3 / FHIR terminology for encounter class, observation category, and condition status. Generation is reproducible via a fixed random seed. Known simplifications: addresses carry city + country only; observation timestamps are UTC; device telemetry is aggregated daily.