Health Samurai Blog

How to configure and fine-tune a patient matching model in MDMbox — from defining match criteria and model structure to blocking, training, and manual tuning.

How Aidbox handles concurrent requests through its HTTP queue — three load scenarios, common failure modes, and how to tune worker capacity for stable performance.

Generate typed profile classes from the US Core IG with @atomic-ehr/codegen. Build compliant Patients and BP observations with typed factories, typed extensions and slices, profile-aware validation, type guards, and typed bundles.

Aidbox 2604 brings HIPAA Safe Harbor de-identification in SQL-on-FHIR, Databricks Lakebase token auth, a reworked SQL Console, and Formbox SMART on FHIR integration.

Aidbox supports HIPAA Safe Harbor de-identification directly in ViewDefinitions. Transform FHIR data into compliant, analytics-ready tables with per-column control — and map results back when needed.

FHIR's Patient/$merge assumes the server knows how to merge. Two decades of MPI vendor configs, EHR vendor divergence, and national registry policy show why one algorithm cannot serve every organization.

FHIR R5's Patient/$merge is a start, but production MDM needs more. We built a resource-agnostic $merge with client-driven plans, atomic audit trails, and a generic $referencing operation.

We're benchmarking popular open-source FHIR servers and Aidbox across CRUD, batch processing, and search workloads. Here's what we're testing and why.

Aidbox 2603 brings BigQuery streaming, Databricks Lakebase support, FHIR package dependency overrides, and Formbox improvements.

How Aidbox moved canonical resolution from runtime to configuration time — with pinning, tree-shaking, and a deterministic candidate selection algorithm.