Arbor

Psychiatry-native trial eligibility.

A psychiatry-native eligibility and screening rail for CNS and ADHD clinical trials. Parses psychiatric EHR data to flag trial-eligible candidates for human review — it never auto-enrolls anyone.

Open the screener All four workspaces

ArborPrecision · Trials

Recruitment where the psychiatric detail actually matters.

General clinical-trial matching systems were built for oncology and cardiology, and it shows. They handle structured billing codes well. Psychiatric nuance is where they miss:

The comorbid ODD that changes an ADHD trial's inclusion criteria
The medication history that decides washout
The symptom-severity subtype hiding in a free-text note

Miss those, and your recruitment funnel fills with candidates who can't actually enrol.

Arbor is built for psychiatry. It parses structured data and free-text psychiatric notes, applies trial-specific eligibility logic, and surfaces candidates to a human reviewer who confirms before anyone is contacted. There is no auto-enrolment — Arbor is a search light, not a shepherd.

Features

Recruitment infrastructure that respects the specialty.

Deterministic eligibility engine

Trial criteria compile into an explicit ruleset — eight operators, three verdicts: eligible, needs review, excluded. No model decides who qualifies. Every verdict cites the source span that produced it.

Human-in-the-loop

Every candidate flag lands in a reviewer queue. No auto-enrolment, ever — the clinician confirms before outreach.

PHI redaction firewall

Identifiers are stripped and tokenised before any text leaves the ingestion boundary, with the token map held in a separate schema. A bypass detector re-scans redacted text and blocks the operation if an identifier survives.

IRB-friendly audit trail

Every parse, flag, and review step is logged with HMAC-chained provenance, sequenced under an advisory lock so the chain cannot fork. A verifier tool walks the chain and exits non-zero on a break.

FHIR R4, HL7v2 and CSV

Three ingestion connectors are implemented and tested. EHR-specific mapping (Epic, Cerner, and others) is not — that is roadmap, not shipped.

Psychiatry-native NLP

The extraction service targets SOAP notes, discharge summaries and med-history free text. The model server itself is not yet deployed — extraction currently runs on the deterministic rules alone, and the NLP tier degrades cleanly rather than guessing.

Open Arbor

Four workspaces — patient, reviewer, sponsor, operations.

A patient screens themselves. A clinician confirms every flag. A sponsor watches the funnel. Operations run the pipeline. All four are live against a demo dataset; a confirmed pilot site is next.

Side A

For patients

A neurodivergent-optimised e-screener: one question per screen, no timers, progress saved as you go, and a resume link if you need to stop.

Open the screener

Side B

For screeners & reviewers

Candidate queue with per-criterion verdicts and the source span behind each one. Confirm, override with a reason, or send back for more information.

Open the reviewer console

Side C

For sponsors & CROs

Enrolment funnel, site performance, feasibility insights, and criteria submission. Export to Medidata Rave or Veeva Vault.

Open the dashboard

Side D

For study operations

Queue depth, quarantine review, user and role administration, criteria approval, and the audit trail.

Open the ops console

Signed in with a demo dataset of two sponsors and three ADHD protocols. Ask us for credentials — no real patient data is present.

Arbor is one of the research tools built inside the Synapse Spark Foundation. Everything we make is part of the same mission — understanding neurodivergent minds and supporting the researchers advancing the science.

← Back to the Foundation