clinosim

clinosim is a clinically realistic hospital data simulator designed to generate synthetic Electronic Health Record (EHR) data in HL7 FHIR R4 format. Unlike traditional tools that sample from static disease distributions, clinosim employs a forward simulation model where every patient possesses a hidden 14-variable physiological state. All observations, including laboratory results, vital signs, medications, and diagnoses, are derived from this dynamic state, ensuring that the generated data is clinically coherent by construction. For instance, a patient on warfarin will exhibit therapeutic PT-INR levels, which drop upon medication removal, rather than just being sampled from a normal range. The tool is particularly useful for creating training data for medical AI and machine learning models, testing EHR systems, developing clinical research methods, and generating educational case datasets. It offers native support for both US and Japanese healthcare standards, complying with the JP Core profile for 16 primary FHIR resource types and including JLAC10 and MHLW YJ codes, as well as Japanese names and addresses. The system is extensible via YAML-driven configurations, covering 32 inpatient diseases and 46 ED/outpatient conditions without requiring code changes. clinosim also supports longitudinal service lines, including oncology (with chemotherapy cycles and tumor markers) and obstetrics (modeling pregnancy as a time-boxed lifecycle with prenatal visits and delivery encounters). The output is fully synthetic and does not ingest or reference any real patient data, making it safe for use in development and testing environments. The project is open-source under the MIT license and requires Python 3.11 or newer.