Clinician-governed care-gap outreach · RI FQHCs
Let's catch those who slip through the gaps.
CATCH turns existing community-health-center data into transparent, clinician-approved outreach, delivered in each patient’s preferred language and channel. Deterministic rules decide who needs follow-up; staff decide what gets sent.
Computed offline from SyntheticRI (Synthea) synthetic records, not real patients. CATCH demonstrates an auditable method; it does not report real Rhode Island prevalence. Every record shows the exact rule that flagged it. Methodology.
Audience
Who it's for
Federally Qualified Health Centers: care-management leaders, population-health and clinical-operations teams who own outreach capacity.
Nurses, medical assistants, care coordinators, and community health workers who review and approve outreach.
People with overdue chronic or preventive follow-up, initially prioritizing RI Hispanic/Latino communities.
How it works
The workflow, end to end
One path from data a clinic already has to an auditable outcome logged back to the care team. A human reviews and approves before anything is sent.
- 1Existing data in
CSV / FHIR / registry export a clinic already maintains. Minimum-necessary fields only.
- 2Transparent rules
Versioned, deterministic eligibility and exclusion logic decides who has a care gap.
- 3Prioritized worklist
Each patient ranked with a plain-language reason and last-contact context.
- 4Tailored draft
An approved template, adapted for the patient’s language and community variety.
- 5Staff review
A person reads the draft, checks the back-translation, and edits if needed.
- 6Approve & send
SMS or email, only after human approval and only to a consented channel.
- 7Outcome logged
Status, version, consent, and timestamps written back for the care team.
- 8Patient in control
Patient chooses language and channel and can opt out or reach a person anytime.
Clinical & community need
Why this matters in Rhode Island
When outreach capacity is tight, follow-up that never happens turns into a clinical risk. Federally Qualified Health Centers (FQHCs) see a large share of the Hispanic/Latino and lower-income patients in the state, and often with a small staff. CATCH helps that staff do more with the hours they have: it shows who needs follow-up, the evidence behind each flag, and a draft message in the patient's language and channel.
- Reaches patients who tend to slip between visits
- Outreach in the patient's own language, not only English
- Wording checked by people from the community, so it reads right
Try it in the demo: in the outreach queue, use the Language access lens to surface the patients flagged for interpreter support, an access-based proxy for reaching LEP patients, routed to bilingual community health workers. It filters on documented interpreter need, never on race or ethnicity.
National figures are labeled as national. CATCH's own counts come from synthetic data and do not report a real Rhode Island prevalence rate.
Why this patient?
Explainable rules, not a black box
- Rule fired
- Treated but uncontrolled
- Eligibility
- Hypertension diagnosis + antihypertensive on file, and 2+ systolic readings ≥ 140 after medication start.
- Evidence fields
- 24 elevated readings (peak 158), diagnosis code, active medication, encounter history.
- Exclusion checks
- Adult (age ≥ 18); same-day readings de-duplicated; ambiguous medications not counted.
- Reason ranked
- Repeated highs while on treatment, with stacked cardiometabolic risk.
- Version / time
- engine v0.1.0 · reference date 2026-05-15
Synthetic record. See the live version, with the decision path highlighted, in the queue.
Rules decide who needs outreach. AI helps adapt an approved message. Staff decide what gets sent.
Eligibility and prioritization are 100% deterministic and auditable, you can read the exact criteria for every flag. Generative assistance is constrained to wording, tone, and reading level. It never decides who is contacted and never sends on its own.
Open a real record in the queue →Language & community congruence
One language is not one community
Generic Spanish can miss differences in vocabulary, tone, health literacy, and trust. CATCH treats each community variety as governed configuration: the clinical meaning is locked, the wording is community-reviewed, and the patient chooses their preference. Pick a variety and channel below, the message changes, the meaning does not.
Variety is chosen by the patient or entered by staff, never guessed from a name or ethnicity. Patients can change it or opt out at any time.
Hola Maria, le escribe Sample Community Health Center. Nuestro equipo de salud revisó su historial y notó algunas lecturas de presión arterial que conviene revisar. Esto no es un diagnóstico. Nos gustaría coordinar una consulta breve de control de presión arterial. Llame al (401) 555-0100 o reserve en clinic.example.org/schedule. Si necesita ayuda antes, comuníquese con su clínica; en una emergencia llame al 911. Responda ALTO para no recibir más mensajes.
Neutral U.S. Spanish. Prototype awaiting review by a bilingual community reviewer.
{{patient_first_name}}{{clinic_name}}{{care_gap_name}}{{scheduling_phone}}{{scheduling_link}}{{preferred_language}}{{opt_out_text}}- Identifies the sender as the patient's own clinic
- States records were reviewed and blood-pressure readings are worth checking
- Explicitly says this is NOT a diagnosis
- Invites the patient to schedule a short visit (no urgency to an ER)
- Gives the same call-to-action: phone or scheduling link
- Includes emergency guidance (call 911) and an opt-out
Actions you take here are logged, with role, variety, channel, and version.
Language and variety are chosen by the patient or entered by staff, never inferred from name or ethnicity.
Clinical content is locked across variants; the language layer may only adapt tone, vocabulary, and reading level.
Draft → community reviewer (native speaker) → clinical reviewer → clinically approved for production.
If no reviewed variety exists, the clinic's approved neutral template is used automatically.
No PHI is sent to an unapproved model provider; generative help is constrained to wording, never eligibility.
Every variant targets a plain-language reading level and is checked before approval.
This short list does not represent every Hispanic/Latino or Portuguese-speaking identity, and varieties are never inferred from ethnicity. Non-English variants are prototypes awaiting review by speakers from each community; an “other / patient-preferred wording” fallback always exists.
Existing tools vs. CATCH
How CATCH compares
| Capability | EHR / registry | Bulk messaging | General-purpose AI | CATCH |
|---|---|---|---|---|
| Uses existing clinical data | Yes | Sometimes | Not inherently | Yes |
| Transparent care-gap eligibility rules | Varies | No | No | Yes |
| Prioritized outreach worklist | Varies | Limited | No | Yes |
| Community-reviewed language variants | Limited | Limited | Ungoverned | Designed in |
| Human approval before outreach | Workflow-dependent | Sometimes | Not inherently | Required |
| Rule / message version audit trail | Varies | Limited | Limited | Designed in |
| Replaces existing datasets | No | No | No | No, it complements |
Categories, not vendors: EHR / registry care-gap modules, bulk patient-messaging platforms, and general-purpose chat assistants. “Varies”, “limited”, and “workflow-dependent” reflect that capabilities depend on the specific product and configuration.
Business model
Who pays, and how
Rhode Island FQHCs, and the Medicaid managed-care and value-based programs they contract with, who carry the quality measures CATCH helps close.
A per-attributed-patient subscription, a one-time integration and onboarding fee, and an optional paid community-language review service.
Closes documented hypertension care gaps that feed quality measures (e.g. HEDIS Controlling High Blood Pressure), and helps a small team cover more patients.
Fixed: product & security engineering, clinical-rule governance, and template maintenance.
Variable: EHR / FHIR integration, SMS / email delivery, implementation, staff training, and paid community-language review.
Shown as a model, not a quote.
Implementation readiness
Technical feasibility & safety architecture
CSV, FHIR, or EHR export from existing registry data. Minimum-necessary fields only.
Versioned, deterministic eligibility and exclusion logic.
Approved base content with constrained language adaptation.
Draft, reviewed, and approved before anything sends.
SMS or email, only after authorization and approval.
Role, rule version, message version, consent, channel, and timestamps.
Role-based access, encryption in transit and at rest, retention controls, and vendor BAAs.
No PHI to unapproved model providers; generation limited to wording.
HIPAA-ready architecture, designed for HIPAA-aligned deployment.
Adoption plan
The first pilot
1 RI FQHC, 1 high-priority care-gap workflow
Historical / synthetic validation first, then staff-supervised outreach
8 to 12 week phased pilot with a small group of coordinators / CHWs
An explicit stop / go review for safety, staff workload, and message quality gates any expansion.
CATCH is a clinician-governed, rule-based care-gap outreach copilot. It does not diagnose, triage emergencies, practice medicine, replace clinicians, or autonomously send messages. All data shown is synthetic (SyntheticRI / Synthea). Efficiency, outcome, and equity gains are stated as pilot hypotheses and targets, not proven results.