Client overview
- Industry focus
- Healthcare
- Portfolio segment
- Healthcare
- Organization profile
- Regional IDN with 42 clinics and 2 hospitals, ~6M patient encounters/year
Patients bounced between EHR silos; scheduling staff maintained shadow spreadsheets for specialty queues. Care managers lacked prioritized worklists tied to HEDIS gaps. Nursing leadership wanted visibility into LOS risk without exporting PHI to unsecured analytics tools.
Problem
Fragmented scheduling and care management tools obscured population health priorities and inflated admin load.
Call center teams toggled five systems to book, confirm, and route patients; duplicate records appeared when MRN merges failed overnight. Specialty access metrics were reported from static PDFs weeks after decisions were needed.
Care managers escalated manually to PCPs when payer authorizations stalled, with no SLA tracking. Social determinants intake lived in Word forms scanned to shared drives — unusable for risk scoring.
Compliance officers worried about PHI in email threads and unapproved cloud drives when teams chased missing records during transitions of care.
Solution
Unified patient engagement layer atop EHR APIs with identity resolution, prioritized care gaps, encrypted collaboration, and governed analytics extracts.
A patient 360 index links MRNs probabilistically with human review queues; scheduling microservice enforces capacity rules per provider templates while surfacing estimated wait fairness to patients via secure portal.
Care gap engine consumes claims + clinical feeds, producing ML-ranked worklists integrated with Epic/Cerner SMART on FHIR launch contexts. Messaging uses topic-based subscriptions with PHI redaction policies.
Snowflake-hosted analytics sits behind row-level security; PHI never leaves approved environments; Tableau embedded for clinic leaders with cell-level auditing.
Implementation
- 1
Clinical workflow ethnography
Shadowing revealed 140 min weekly per scheduler on phone tag; we digitized deferrable intents to self-service chat with nurse escalation. RACI clarified who owns duplicate resolution.
- 2
Phased clinic rollout
Pilot sites in two counties validated no-show interventions and consent capture on tablets. Feedback loops tuned SMS cadence within TCPA constraints.
- 3
Population analytics trust
Joint validation with analytics governance council on cohort definitions; differential privacy experiments for research partners. BAA amendments aligned with AWS and Snowflake vendors.
Tools & platforms
- FHIR R4
- SMART on FHIR
- AWS (HIPAA)
- Snowflake
- Twilio HIPAA
- Tableau embedding
Engineering challenges addressed
- Latency when aggregating longitudinal records from two EHR vendors.
- Consent management for outreach when state laws differed for adolescents.
Program artifacts & environments
Tech stack
- Next.js
- Node.js
- PostgreSQL
- AWS
- Snowflake
- FHIR
- Kafka
- Tableau
- Terraform
Results
- 23% reduction in no-shows after omnichannel reminders and self-reschedule
- 31% faster median time to close HEDIS care gaps in pilot cohorts
- Nurse admin time on care coordination down 17 hours FTE/week across pilot clinics
Quantified impact
23% fewer no-shows
Six-month pre/post with weather and flu season controls.
4.2 NPS improvement for scheduling experience
Post-visit surveys, n>1.8k.
Key takeaways
- Patient engagement platforms must earn clinical trust — start with narrow, measurable workflows before promising full population AI.
- Identity resolution needs human-in-loop review; fully automated merges are unsafe in healthcare.
- Analytics adoption follows data dictionary clarity; invest in glossary before dashboards.