Research & Development
Practical research that improves care operations.
Healthzee R&D turns front-line signals into safer, faster workflows - standards-first (FHIR/HL7), privacy-aware, and vendor-neutral.
Mission
Our mission is to learn from real-world operations and help healthcare organizations improve access, safety, and efficiency - without vendor lock-in and without compromising privacy. We focus on standards, measurable outcomes, and repeatable playbooks that any clinic or hospital can adopt.
Commitments
Guardrails we never compromise on
Patient safety first (non-diagnostic, crisis redirects always on).
Standards-first (FHIR/HL7) with a small-practice calendar fallback.
PHI minimization, role-based access, and auditability.
Transparent metrics; no inflated "AI" numbers.
Share what we learn - de-identified and ethics reviewed.
Focus Areas
Where we are deploying research pilots
Each program is a modular stack that you can pilot independently or combine into an integrated operating picture.
Adverse Drug Reactions (ADR)
Earlier signal, cleaner case assembly, faster review.
Operational Insights
Intent analytics, staffing peaks, channel mix, bilingual access.
Reminder Science
Timing and phrasing experiments that reduce no-shows and late cancels.
Methods
How we work with your data
Modular components let you start with CSV exports and graduate to standards-based pipelines without rewriting everything.
Data Ingest (CSV -> FHIR/HL7)
Start with CSV; upgrade to standards when exposed by your systems.
De-identification
PHI minimization at source with role-based unmask when needed.
Analytic Pipeline
Event stream -> feature store -> dashboards -> export (CSV/PDF).
Human-in-the-Loop
Review queues, checklists, sign-off, and audit trail.
Research guardrails
Guardrails and governance
Operational guardrails
- Non-diagnostic: Research modules never provide medical advice.
- Emergency redirects: 911/988 are always on and non-configurable.
- Consent and notices: Clear consent banners for reminders and study participation.
- Retention: Configurable windows (90/180/365 days) with auto-purge.
Outcomes we measure
- Time-to-review drops for ADR cases and operational incidents.
- Capture rate increases for suspected ADRs and appointment confirmations.
- No-show rate decreases via optimized reminder windows.
Get involved
What a pilot looks like
Every pilot has a clear entry point, data-sharing plan, and governance pathway so you can move fast without sacrificing compliance.
Pilot length
6-12 weeks with weekly check-ins.
What you provide
A single point of contact, sample datasets (CSV acceptable), and feedback.
What you get
Dashboards, exports, playbooks, and a de-identified summary.
Ready to participate?
Start a pilot and choose which module to activate.
Research - ADR (Adverse Drug Reaction)
ADRs spotted sooner - reviewed faster.
More signals: Intake, conversations, staff forms, and patient-reported outcomes feed one queue.
Cleaner cases: Guided fields enforce the minimum dataset (suspect drug, event, timing, outcome, reporter).
Faster review: Timelines, checklists, and assignments in one place with full audit.
Data model (minimum dataset)
- Suspect medication(s): name, strength, route, start/stop dates.
- Adverse event(s): description, onset date/time, seriousness flags.
- Patient factors: age band, sex, comorbidities (optional), allergies (optional).
- Co-medications: concurrent therapies (optional).
- Outcome: recovered, recovering, not recovered, fatal/unknown.
- Reporter: role (patient/clinician/pharmacist), contact channel.
- Artifacts: labs, notes (de-identified when possible).
We support CSV today; FHIR/HL7 objects where available. Mapping templates are provided.
Data ingest
- Claims, dispensing logs, encounter summaries, patient and clinician reports.
- CSV now; FHIR/HL7 when exposed.
- Optional de-identification at source.
NLP triage (optional)
- Extract likely medication and event mentions from free text.
- Normalize to controlled vocabularies (e.g., RxNorm/SNOMED) when feasible.
- Always human reviewed; never auto-final.
Causality aids (non-diagnostic)
- Naranjo-style prompts and minimum dataset checks.
- Temporal association plus de-challenge and re-challenge views.
- "Data sufficiency" meter for reviewers.
Signal board
- Frequency shifts and disproportionality flags (pilot).
- Filters by drug class, age band, sex, and time window.
- One-click assignment to a reviewer with SLA timers.
Case review
- Assign -> comment -> redact -> finalize.
- Export de-identified case packages (CSV/PDF).
- Full audit trail and reviewer workload dashboard.
Alerts and governance
- Threshold emails to the pharmacovigilance team.
- Consent, masking, and retention windows.
- Role-based access with least privilege.
KPIs
- Suspected ADRs captured increases.
- Time from report to first review decreases.
- Percentage of cases with the minimum dataset increases.
- Mean time-to-finalization decreases.
- Reviewer throughput increases.
Implementation plan (6-12 weeks)
- Week 0-1: Governance and data access (CSV acceptable); KPI baseline.
- Week 2-4: Pilot queue live; reviewers trained; calibrate minimum dataset.
- Week 5-8: Tune thresholds; first summary; publish playbook v1.
- Week 9-12: Stabilize pipelines; final report; go/no-go for scale.
FAQ (ADR)
- Is this a diagnostic tool?
- No - information organization for human reviewers.
- Do you store PHI?
- Only as permitted; PHI minimization and retention controls apply.
- Can we use our terminology?
- Yes - upload mappings and lists.
- How do you handle duplicates?
- Hash-based matching plus reviewer merge tools.
Research - Operational Insights
Operational clarity from everyday conversations.
What you will see
- Top intents: hours, directions, insurance, booking, refills.
- Time-of-day peaks: staff at the right hours.
- Language split: EN/ES volumes; bilingual access ROI.
- Channel mix: voice vs SMS; opt-out trends.
- Missed vs captured: delta after bot activation.
Dashboards and exports
- Time-series charts with comparison windows (pre/post pilot).
- Heatmaps for peak staffing windows.
- Intent taxonomy editor (merge/split).
- CSV/PDF export and weekly email digest.
How it works
- Event stream -> feature store -> charts and exports.
- No PII in aggregates; de-identify where possible.
- Benchmarks for clinics vs hospitals; EN/ES split.
KPIs
- Abandoned calls decrease.
- First-response time decreases.
- Coverage during peaks increases.
- Bilingual answer rate increases.
- Campaign conversion increases (if enabled).
Research - Reminder Science
Fewer no-shows without guesswork.
Experiment tracks
- Timing: T-72h vs T-48h vs T-24h vs T-2h (alone or combined).
- Channel: SMS vs email vs voice fallback with quiet-hours compliance.
- Phrasing: neutral vs supportive vs action-oriented; bilingual.
- Personalization: visit type (new/return), prep required, travel time.
Sequences (examples)
- Standard: T-48h SMS -> T-2h SMS (with confirm/cancel links).
- High-prep: T-72h email (instructions) -> T-24h SMS (checklist) -> T-2h SMS.
- High no-show risk: T-48h SMS (confirm now) -> T-24h voice fallback if no response.
- Bilingual: auto-detect language; send in EN/ES accordingly.
Analysis and ethics
- Randomized cohorts; significance checks with small-N safeguards.
- No coercive wording; clear opt-out (STOP/HELP).
- Calendar fallback for rescheduling; consent records in audit.
KPIs
- No-show rate decreases.
- Late cancel rate decreases.
- Confirmation rate increases.
- Re-booking time decreases.
- Staff call-backs decrease.
Methodology & governance
How we keep research ethical and transparent
Document how we run research ethically and transparently.
Principles
- Minimal necessary data; privacy by design.
- Non-diagnostic posture; crisis redirects mandatory.
- Role-based access; least privilege.
Governance
- Study brief and goals approved by the organization.
- Data Processing Addendum and BAA (if required).
- Retention schedule defined up front.
Security
- Encrypted at rest and in transit.
- Access logging plus anomaly alerts.
- Redaction tools for exports.
Publication and IP
- Results shared de-identified unless otherwise agreed.
- Participants may be acknowledged (opt-in).
- Reusable playbooks licensed for participating orgs.
Participation & agreements
What collaboration looks like
Clarity on roles, agreements, and hand-offs keeps the research cadence predictable for your team.
Pilot structure: 6-12 weeks; weekly check-ins; one SPOC.
What you provide
- A sample dataset (CSV is fine).
- A contact for operations and compliance.
- Policy references (reminders, crisis scripts).
What you get
- Live dashboards, weekly summaries.
- De-identified final report.
- A production-ready playbook if you choose to scale.