HealthKey

HealthKey's core service is automated prescreening of existing patients against trial eligibility criteria. It reads structured records and clinical documents, evaluates criteria across each patient's history, and returns a ranked list with supporting evidence. The Oregon Urology deployment applied 24 criteria to 5,360 prostate cancer patients, including information in imaging, pathology, laboratory results, and medication histories.

At Tekton, the workflow also supported a custom prioritization score for chart review. Research staff still determine qualification, but the product's central work is the model assisted search and comparison. Without that component, the site would be left with the manual record review the service is intended to reduce.

Doctors and coordinators review the source evidence before deciding whether a patient qualifies. At Oregon Urology, six full matches were separated from 42 near matches that needed judgment. The Flourish workflow similarly flagged boundary cases involving BMI, laboratory timing, or ambiguous history for manual review. Those distinctions preserve a concrete human decision step rather than treating every automated match as final eligibility.

The system can prioritize candidates and identify the relevant records, while automatic exclusion thresholds, a route to revisit rejected patients, and mandatory escalation rules remain unspecified. Review of returned candidates does not by itself establish how the team discovers an eligible patient omitted from the list.

HealthKey translates inclusion and exclusion criteria into checks across structured fields and unstructured records, then ranks likely candidates. At Oregon Urology, the logic connected imaging findings, pathology, laboratory values, and medication history across 24 criteria. Tekton added a site specific prioritization score to help direct chart review, while founder descriptions say doctor feedback improves the system.

These details explain the work the software performs and the information it uses. The model family, suppliers, training population, extraction architecture, production version, and process for evaluating updates remain unspecified. A count of criterion evaluations does not identify the technical model behind them.

Patient information moves from the clinic's EHR into HealthKey's matching service and is returned to research staff as candidate evidence. Named EHR integrations identify the record systems involved, but do not identify who supplies or operates the matching models. A hosting company, external model supplier, complete processor register, and downstream retention or training terms remain unspecified. Whether the model is operated entirely in house or calls an external inference service also remains unclear.

The April 2026 Oregon Urology case reports 5,360 patients screened, 5.2 million medical records processed, six eligible matches, 42 near matches, three consents, and one enrollment. It identifies the trial, EHR, 24 criteria, and 21,115 criterion evaluations, making the workflow more interpretable than a headline percentage. A May Flourish case describes 323 prescreen matches, 175 patients subsequently screened by coordinators, 74 passes, and two enrollments within 25 days.

Those stages are different denominators. The analyses remain vendor case studies without independent adjudication, controlled comparison, or a measurement of eligible patients the software missed. Estimated manual hours are not directly observed savings.

The matching workflow processes structured patient data and narrative clinical documents, including faxed PDFs, and returns record evidence to research staff. Doctor feedback can improve the system, but the terms governing that improvement are unspecified.

A patient data retention schedule, restrictions on model training, deidentification method, downstream provider handling, and deletion behavior remain unclear, as do technical hallucination safeguards and procedures for investigating AI safety incidents.

The EHR connection and record analysis carry a HIPAA compliance claim. The actual workflow handles clinical notes, medication histories, laboratory information, imaging reports, and other patient documents before returning evidence to research staff. A readable business associate agreement and substantive service privacy terms remain unavailable. The applicable customer agreement, subcontractor obligations, participant data rights, and exact scope of business associate handling are therefore unspecified despite the compliance claim.

The service carries a HIPAA compliance claim for handling EHR data. Research teams access a separate dashboard and receive links to supporting clinical evidence, making access controls and the handling of copied records relevant to deployment. Independent security audit status, certification type, auditor, reporting period, penetration testing scope, and a retrievable assurance artifact remain unspecified. The compliance claim alone does not identify an outside examination of the service's operational security.

The software identifies potential trial participants from existing records and leaves qualification to doctors and research coordinators. Its work applies a sponsor's inclusion and exclusion criteria, surfaces relevant chart evidence, and helps a site prioritize screening. The named oncology and obesity cases describe subsequent consent, screening, and enrollment steps performed through the research operation.

These are recruitment support functions rather than independent diagnosis or treatment selection. No device clearance is claimed for this stated role. The fact that a trial evaluates a regulated drug or device does not confer that product's regulatory status on the software used to find potential participants.

The cases measure candidate discovery, screening activity, and enrollment, with clinicians reviewing evidence behind proposed matches. Those operational results do not evaluate missed eligible patients across demographic or physiological groups. A formal process for approving model changes, comparative subgroup performance, and independent assessment of matching behavior remain unspecified. Screening against protocol criteria establishes the intended selection rules, but does not demonstrate that the system applies them equally well across patient populations or documentation styles.

Clinicians can inspect the records behind a match and decide whether a patient qualifies, while near matches can receive additional review. This provides a practical route to challenge an automated selection before enrollment. The founder's launch description says clinics pay only when qualified patients are identified, but the current scope of that promise is unclear alongside the expired setup promotion.

A fee arrangement without operative terms does not specify compensation for a false match or a missed candidate. An output accuracy warranty, patient remedy, contractual correction obligation, and the allocation of responsibility after an erroneous screening decision remain unspecified.

Oregon Urology's deployment connected to athenahealth and ingested structured information plus 652,000 clinical documents or PDFs. Flourish's obesity project connected four clinics using Practice Fusion and eClinicalWorks, while Tekton's rheumatoid arthritis workflow also used Practice Fusion. These are named production routes for reading records into the matching service.

Broader compatibility claims include Epic, Veradigm Allscripts, NextGen, and other systems, but do not establish the same deployment scope for each. Results are reviewed in HealthKey's dashboard with evidence links. A production EHR writeback workflow, bidirectional chart update permissions, and the interface details for every listed connector remain unspecified.

Research teams use a hosted dashboard to review candidates and their supporting records, with EHR connections bringing clinical information into the matching workflow. Multi-clinic deployments can process records across different systems in parallel. The hosting provider, committed storage country, model inference locations, customer isolation design, disaster recovery geography, and private deployment options remain unspecified. Named clinic locations describe where customers operate, rather than establishing where their data is processed.

The stated charging model is a success fee per patient that scales with sites, studies, and subjects. Implementation involves connecting the clinic's records and configuring the protocol's inclusion and exclusion logic, with support for research teams reviewing the results. The homepage's $0 upfront offer expired July 31, 2026; its assigned values for implementation, setup of five trials, and support do not establish current charges.

An untimed founder launch description also says clinics pay on successful identification, but the operative fee rate and definition of success remain unspecified. Current setup costs, minimum commitments, and whether payment depends on qualification or enrollment are therefore unresolved.

Named deployments support research practices and networks in urology, oncology, rheumatology, and metabolic studies. Oregon Urology used HealthKey for the PSMA-DC prostate cancer trial and achieved an enrollment. Flourish screened an obesity study across four Chicago area partner clinics using two EHR systems, with coordinators reviewing the candidate pipeline and two patients enrolling within 25 days.

Tekton's rheumatoid arthritis case describes 18 eligibility criteria and clinician review of a narrowed candidate list. These examples establish specific settings and specialties, including both a practice and a distributed clinic network. They do not establish identical performance in every indication, trial phase, or patient population the platform could address.