apparent
US Physician Referral Networks · 2014 – 2017
Topology-aware exploration
of apparent.
A prototype companion to the Aidos Lab dataset. Explore referral networks by state, by Hospital Service Area, or as force-directed graphs with Forman–Ricci edge curvature — all from SQLite, no cloud graph DB required. Advanced users can optionally wire in self-hosted Neo4j.
Box-plot every HSA in a state across the three years. Network size, Forman curvature, clustering, income, unemployment, no-HS rate.
Pick any one of 3,239 HSAs and watch its physician network, demographics, and curvature evolve.
Force-directed graph of the physician referrals inside a single HSA, with edges coloured by Forman–Ricci curvature.
Read-only SELECTs over all seven SQLite tables. Inherits the preset queries from the original Datasette site.
Optional graph queries — hub physicians, k-hop neighbourhoods, triangle motifs. Requires a local or hosted Neo4j.
The dataset
3,239 Hospital Service Areas across 51 states, mapped to ~1 M physicians and 39 M dyadic referral interactions from Medicare claims (2014 – 2017).
Network features (Forman curvature, clustering, density, centrality) are pre-computed and joined with US Census demographics and Dartmouth Atlas metadata.
What's new vs. Datasette
The original site (Datasette + datasette-dashboards) showed aggregate metrics but never the underlying graphs.
apparent v0.1.0 adds a force-directed network view per HSA, edge-level Forman–Ricci curvature, hub and bottleneck rankings, and read-only SQL — runnable locally with just the SQLite file.