Skip to main content
Digital Scientists
Problems Work Ventures About
Healthcare data engineering

Healthcare data analytics and engineering services

Digital Scientists builds the data work behind healthcare decisions: pipelines out of EHRs, claims and on-premises databases, data marts and lakehouses, dashboards in Power BI or Tableau, and predictive models on top. For CommuniCare we moved risk adjustment data from on-premises SQL Server into AWS and built the engine on it, which delivered $10M+ in RAF improvement at 90%+ accuracy across 20K+ patients. Most clients start with a call to talk it over, then a Blueprint that maps their sources.

Updated

Claims analytics dashboard on a laptop, showing spend by diagnosis, patient priority group and place of service
A claims analytics dashboard: spend by diagnosis group, patient priority and place of service, from one validated data set.
$10M+

RAF improvement at 90%+ accuracy

20K+

Patients in the risk adjustment data, 7 states

60+

Systems audited for CommuniCare Advantage

7,020

Enrolled patients tracked on Congruity Health

Duke Health
Congruity Health
McKesson
CommuniCare
Guardian
Easterseals
Proof first

What does healthcare data engineering look like when it pays off?

It looks like CommuniCare's risk adjustment work. We moved the data out of on-premises SQL Server into AWS, cleaned it, and built an engine on it that delivered $10M+ in RAF improvement at 90%+ accuracy.

The data sat in on-premises Microsoft SQL Server databases. We built a secure pipeline over a VPN tunnel into AWS, mapped and normalized 7 tables across 3 databases, and checked data quality before any model touched it.

On that foundation the engine reads six clinical sources per patient: diagnoses, health risk assessments, medications, progress notes, labs and claims. Most of the work is plain SQL against structured data. Generative AI handles a smaller share, and every suggested code links to the evidence in the chart. Coders accept, change or reject each one. A patient review that took an hour now takes about 60 seconds, across 20K+ patients, 245 providers and 7 states. See the risk adjustment case study.

$10M+

RAF improvement

90%+

Accuracy on suggested codes

1 hr to 60 s

Review time per patient

6

Clinical data sources per patient

The hard part

Why is healthcare data so hard to use?

Because it lives in many systems that were never built to agree with each other. Before anyone can trust a report or a model, someone has to get the data out, line it up and prove it matches the source.

01

Many sources

EHRs, claims files, payer feeds, on-premises databases and spreadsheets, each with its own IDs and code sets. CommuniCare's risk adjustment work alone pulled from several payer and vendor feeds plus an EHR export.

02

Different definitions

A patient, an encounter or a member means something slightly different in each system. We write the definitions down and build them into the data model.

03

Data you can't check

A dashboard nobody trusts goes unused. We validate counts and quality against the source before a report or model goes live.

04

Patient data rules

Where PHI lives, who can see it and what gets logged are design decisions. We sign a BAA before we touch patient data.

What we build

What healthcare data systems do you build?

Pipelines, data marts, lakehouses, dashboards and the models that sit on them. Here is where we've built each one.

Healthcare data work we've shipped

What we buildWhat it doesWhere we've built it
Cloud data migrationMoves on-premises databases into the cloud in stages, validated against the sourceCommuniCare risk adjustment: on-premises SQL Server to AWS over a VPN tunnel. See HIPAA cloud migration.
Data pipelines and data martsLands raw data, models it in layers and serves reporting tablesCommuniCare and NeverAlone: Meltano and Airflow pipelines into BigQuery, with call, telehealth, facility and patient handoff marts
Public healthcare data lakehouseJoins CMS provider, ownership and cost report files with NPPES and PECOS into one queryable storeOur own lakehouse on DuckDB and Parquet: 377,000 organizations and 210 million physician referral links
Research and SDOH data platformsBrings scattered program data and social drivers of health data together on a mapA Duke University SDOH research platform: county and census tract maps, matching, grants and analytics
Claims analyticsTurns claims and eligibility data into flags for at-risk patientsCongruity Health: 7,020 enrolled patients tracked, 968 intervention triggers
BI and reportingDashboards in Power BI, Tableau or a custom app, on validated dataCommuniCare Advantage: one data and reporting layer in Power BI after a 60+ system audit

Quality reporting such as HEDIS and SNP runs on the same pipelines. See quality measures.

Predictive analytics

What is predictive analytics for healthcare, and where does it pay off?

Predictive analytics for healthcare uses past clinical, claims and operational data to flag what is likely to happen next, so staff act before it does. It pays off where a person already makes the call and the model tells them where to look first.

Our two clearest examples. The risk adjustment engine predicts which diagnoses are documented in the chart but missing from claims, and it delivered $10M+ in RAF improvement at 90%+ accuracy. For Congruity Health we built a platform that reads claims and eligibility data for musculoskeletal and Type II diabetes patterns and triggers outreach: 7,020 enrolled patients tracked, 968 intervention triggers and $109K+ in cardiometabolic savings identified.

We also built a model on public CMS data that predicts nursing home quality star ratings, tested on chains it had never seen. Off-the-shelf tools are fine when your question is common and your data is clean. Custom makes sense when the model has to fit your population, live inside your workflow and explain each flag.

The rules we apply: a person reviews every output that matters, no PHI goes into a tool without a BAA, and we start with back-office work like coding and reporting before anything clinical. More on healthcare AI development.

90%+

Accuracy on missed diagnosis suggestions

968

Intervention triggers on Congruity Health

$109K+

Cardiometabolic savings identified

How it works

How does a healthcare data project run?

Map the sources first, then move and validate the data, then build the reports and models people will use. Nobody likes a big-bang cutover, so the old reports keep running until the new numbers match.

From scattered sources to trusted reports

StepWhat happensWhat you get
1. MapList every source: EHR, claims, payer files, databases, spreadsheets. Mark where PHI lives and who owns each feed.A source inventory and a PHI data-flow diagram
2. ConnectSign the BAA. Open secure links to each source through an API, a standard interface or a reliable export.Data moving into your cloud account, encrypted
3. ModelLand raw data, then build cleaned and reporting layers with written definitions for patient, member and encounter.Data marts your analysts can query
4. ValidateCheck counts and quality against each source. Fix the mapping, not the report.Numbers that match the source system
5. ServeBuild dashboards in Power BI, Tableau or a custom app, and models where they earn their place.Reports and flags people use every week
6. Run or hand overWe monitor the pipelines and fix breaks, or hand the code and runbooks to your team.Pipelines that keep running after launch

Most clients start with a Blueprint, 4 or 6 weeks, that maps the sources and sets the plan. More on ways to start.

Tools and standards

Which data tools and healthcare standards do you work with?

We pick tools to fit your cloud, your team and your budget, and we have shipped on most of the common ones. The healthcare standards matter more than the warehouse brand.

Data from EHRs reaches the warehouse through the same connections we build for healthcare integrations. PointClickCare and Netsmart (GEHRIMED) data, with patient matching across EHRs, is in production on platforms we operate. Our Epic work is in a sandbox. We can integrate with any system that has an API, a standard interface or a reliable export.

Pipelines

Meltano, Apache Airflow, Python, SQL, R

Storage

BigQuery, PostgreSQL, Microsoft SQL Server, DuckDB and Parquet

Cloud

AWS, Google Cloud, Microsoft Azure

BI

Power BI, Tableau, custom React dashboards

Healthcare data

FHIR, HL7, X12 claims, ICD-10, CPT, CMS-HCC

Public data

CMS provider files, NPPES, PECOS, Medicare cost reports, Census ACS

Security

How do you keep PHI safe in a data pipeline?

We design it in from the first pipeline. We sign a BAA with every covered entity, keep the data in your cloud account where possible, and log who touched what.

01

BAA first

Signed before any PHI moves or anyone gets access. Our security plan is mapped to NIST CSF 2.0, and each client gets a security addendum.

02

Encryption and access

Data is encrypted in transit and at rest. Access is by role, with least privilege for our own engineers.

03

Audit trail

Every load is logged, and you can trace each report figure back to its source table.

04

Less PHI where you can

De-identified or limited data sets for analytics and model work when the question allows it. See healthcare security and PHI.

What Our Partners Say

Trusted by healthcare leaders.

“The team at Digital Scientists was a huge help in not only developing our analytics platform, but understanding the competitive space and defining our true value proposition for our customers.”
Justin Davis

Justin Davis

CEO, Congruity Health

“Digital Scientists provided immense value to our plan administrators, providers, and patients. The team was able to come in and understand our core needs, while remaining product agnostic. They helped identify the technical debt and pitfalls, and to get us to a point of stabilization and growth.”
Robert Hager

Robert Hager

CEO, CommuniCare Advantage

FAQ

Questions buyers ask about healthcare data analytics

What is healthcare data analytics?

Healthcare data analytics is collecting, joining and analyzing data from clinical, financial and operational systems to make better decisions. It runs from basic reporting to predictive models that flag at-risk patients or missed diagnoses. The foundation is data engineering: the pipelines and data models that make the numbers trustworthy.

What is the difference between data analytics and predictive analytics in healthcare?

Data analytics tells you what happened and why. Predictive analytics uses that history to flag what is likely to happen next, such as a diagnosis missing from claims or a patient who needs outreach. Predictive work only holds up when the data underneath has been validated, so we build the pipeline first.

What's the difference between a data lake and a data warehouse?

A data lake stores raw data in its original format at low cost. A data warehouse stores cleaned, structured data built for fast queries and reports. Healthcare teams usually need both: raw data kept for audit and new questions, and modeled layers for dashboards. We build the raw, cleaned and reporting layers as one design.

How long does a healthcare data project take?

It depends on the number of sources and how clean they are. A Blueprint of 4 or 6 weeks maps the sources and sets the plan. First reports on validated data usually follow within a few months, and we add sources in stages after that.

Can you migrate our on-premises databases to the cloud?

Yes. For CommuniCare we moved risk adjustment data from on-premises SQL Server into AWS over a VPN tunnel, mapped 7 tables across 3 databases, and validated it before any model used it. We run old and new side by side until the numbers match. See HIPAA cloud migration.

Can you integrate with our existing EHR system?

Yes. PointClickCare and Netsmart (GEHRIMED) data integration, with patient matching across EHRs, is in production on platforms we operate. Our Epic work is in a sandbox environment. We can integrate with any system that has an API, a standard interface or a reliable export. See healthcare integrations.

How do you keep data pipelines HIPAA compliant?

We sign a BAA with every covered entity before we touch patient data. Pipelines encrypt data in transit and at rest, use role-based access and log every load. Our security plan is mapped to NIST CSF 2.0, and each client gets a security addendum.

Tell us what's holding your platform back

Where should we reply?

No commitment. No sales deck. We respond the same day, and you talk to senior leadership.

We modernized a 20-year-old mission-critical operations platform under NRC oversight, with no big-bang cutover.