Capabilities
Test-driven software and medical data sciences.
Three lines of work, one standard. We build test-driven software, data pipelines and secure environments health programs run on; we turn the data those programs hold into predictive models, analysis and multimodal AI; and we bring physician-led clinical services and advice to the programs that need them.
The consoles and product figures on this page are illustrative product concepts, not screenshots of delivered systems. The respiratory, Tampa Bay and research-funding charts are CDC’s and NIH’s own data, and each names its source.
NAICS 541511· Primary
Custom software & data pipelines
Software built for health programs that have to answer for it: applications and internal tools, data and analysis pipelines with full lineage, and APIs that modernize legacy service layers. Versioned data and code, documented decisions, automated tests, and interfaces designed to Section 508 from the first sketch — so a reviewer can trace every result back to its source.
Caseload, 64 patients: change in PHQ-9 since enrolment
Each dot is one patient at their latest contact. Choose a patient on the chart or from the lists.
- Remission (PHQ-9 under 5)
- Responding
- Needs a treatment change
- Early in treatment
- Self-harm item positive
Safety follow-up today
Due for case review this week
P-21416: PHQ-9 over 22 weeks, bupropion 300 mg
| Date | Week | PHQ-9 | Contact |
|---|---|---|---|
| Nov 3 | 22 | 16 · item 9 positive | care manager (phone) |
| Oct 20 | 20 | 12 | care manager |
| Oct 6 | 18 | 12 | care manager (phone) |
| Sep 22 | 16 | 15 | care manager |
| Sep 8 | 14 | 13 | care manager (phone) |
| Aug 25 | 12 | 13 | care manager |
| Aug 11 | 10 | 14 | care manager (phone) |
| Jul 28 | 8 | 10 | care manager |
| Jul 14 | 6 | 13 | care manager (phone) |
| Jun 30 | 4 | 14 | care manager |
| Jun 16 | 2 | 18 | care manager (phone) |
| Jun 2 | 0 | 18 | care manager |
- active patients
- 64
- active patients
- responding, 50% or more better
- 48%
- responding, 50% or more better
- in remission, PHQ-9 under 5
- 27%
- in remission, PHQ-9 under 5
- not responding at 10+ weeks, for case review
- 13
- not responding at 10+ weeks, for case review
- self-harm item positive, same-day follow-up
- 2
- self-harm item positive, same-day follow-up
What we deliver
- Clinical software: measurement-based care registries and outreach tools
- Custom web applications and internal tools for health programs
- Engineered data and analysis pipelines with full lineage
- API design and modernization of legacy service layers
- Accessible interfaces designed to Section 508 / WCAG 2.2 AA
- Versioned data and code; every figure regenerates from source
- Automated testing and documentation as first-class deliverables
NAICS 541512
Health systems design & interoperability
Data exchange is where health IT programs go to die. Hale MDS works natively in the standards — FHIR, HL7v2, X12, OMOP — to connect systems that were never designed to talk. On top of that plumbing we design the intelligence: clinical decision support at the point of care, and AI-driven medical systems that survive contact with government oversight — documented, monitored, bias-tested, and explainable to a review board.
393,000 messages a day, five feeds, three destinations
Band width is daily volume. Moving dots are messages from the selected feed. Hover a band for its volume.
One VistA message, three forms
As received VistA HL7 v2.5.1 ADT^A03, a discharge
MSH|^~\&|VISTA|VAMC|HALE|ENGINE|20261024083012||ADT^A03|81920|P|2.5.1 EVN|A03|20261024083000 PID|1||ICN48213^^^USVHA^NI||||19560412|F PV1|1|I|5W^512^A|||||||||||||||V7731
Normalized FHIR R4
{ "resourceType": "Encounter",
"status": "finished",
"class": { "code": "IMP" },
"period": { "start": "2026-10-19", "end": "2026-10-24" },
"subject": { "reference": "Patient/48213" } }Delivered Decision support
Care-team alert: discharged after a 5-day stay, with an admission 22 days earlier. 30-day readmission risk 0.31, above the 0.25 outreach threshold.
Where the time goes
End to end: 385 ms at the median, about 894 ms at the 95th percentile. Bars are medians; whiskers reach each step’s 95th percentile.
Flag rate by provider, against the network rate
268 providers, 6 specialties. Hover any provider; pick one above the limit to see why its claims were flagged.
Why PRV-2923 stands out
Laboratory. 95 claims, 30 flagged by the claim-level model (31.6%) against a network rate of 6.4%: 10.1 standard errors above it.
What pushes its claims’ scores up, and down
Referred at the 99.8% limit, highest first
- Provider within the limit
- Above the limit, referred
- Network flag rate
- Control limit
- providers above the 99.8% limit
- 11
- providers above the 99.8% limit
- expected above it by chance alone
- 0.3
- expected above it by chance alone
- claims held for review
- 14,583
- claims held for review
- of all claims flagged, network-wide
- 6.4%
- of all claims flagged, network-wide
Share of attempts in the outreach tier, by tier size
100,000 patients ranked by predicted 12-month risk of a suicide attempt, highest first. The gray line is what choosing patients at random would capture.
Of every 100 patients in the top 0.1%
about 11 make an attempt within a year; 89 do not.
- patients in the top 0.1%
- 100
- patients in the top 0.1%
- attempts in the tier (2%)
- 11 of 652
- attempts in the tier (2%)
- of the tier go on to an attempt, against 0.7% overall
- 11.0%
- of the tier go on to an attempt, against 0.7% overall
- patients reviewed for each attempt the tier contains
- 9
- patients reviewed for each attempt the tier contains
Calibration: predicted against observed risk, by tenth of predicted risk
Points on the diagonal mean the model’s risk is right on average. Whiskers are 95% intervals.
Everyone: observed 1.00× predicted
Within the 0.8–1.25 band the review uses: the model’s risk is right on average for this group.
100,000 patients, 652 attempts. The model sees a risk score only, which is how an age or rurality effect it was never given shows up here.
This month’s tier, by facility: reviewed within 2 business days
Twelve facilities. The dashed line is a 90% target; facilities under it are marked.
One patient in the top 0.1%: predicted risk 17.1%, 26× the average
What raised this patient’s predicted risk
After the flag
- Flag reviewed by a clinicianwithin 2 business days of entering the tierdone Oct 28
- Structured suicide risk screenC-SSRS screener: moderate riskdone Oct 29
- Safety plan completed with the patientwarning signs, coping steps, contacts, crisis linedone Oct 29
- Lethal means counselingdocumented in the plandone Oct 29
- Follow-up contactwithin 7 days of the safety plandue Nov 5
- Caring contactsa brief message each month for 12 months, the approach behind VA’s Caring Lettersfrom Dec 1
Decision support for clinicians: the score orders a review list. It does not diagnose, and the model contacts no one.
Veterans and service members in crisis: dial 988 and press 1 for the Veterans Crisis Line, text 838255, or chat at VeteransCrisisLine.net. Anyone else: call or text 988, the Suicide & Crisis Lifeline.
What we deliver
- Clinical decision support embedded in daily workflow
- FHIR R4 servers, facades, and bulk-data pipelines
- HL7v2, C-CDA, and X12 interface engineering
- Terminology services: SNOMED CT, LOINC, RxNorm, ICD-10 mapping
- Risk models with model cards, drift monitoring, and bias testing
- Health IT modernization — incremental strangler-fig, not big-bang
NAICS 518210
Secure data platforms & hosting
Health data deserves infrastructure built for it. We design, build and run the platforms that ingest, process, and serve clinical and claims data at scale — and the research enclaves where sensitive datasets can be analyzed without leaving governed boundaries. Accreditation is treated as an engineering problem: controls as code, evidence generated automatically, environments reproducible from day one.
Streaming ingest, last two minutes
Thousands of events a second, by pipeline. Hover to read any second.
Error budget, 30-day window, 99.95% availability target
Budget remaining, day by day. Hover for each day’s burn rate.
- Budget remaining
- Even burn
- Burn-rate page
- of the month’s error budget left, day 22
- 61.9%
- of the month’s error budget left, day 22
- burn rate over the last 7 days
- 0.9×
- burn rate over the last 7 days
- projected at the end of the window
- 38.1%
- projected at the end of the window
- of full outage the 99.95% target allows in 30 days
- 21.6 min
- of full outage the 99.95% target allows in 30 days
What we deliver
- Enclave architectures that inherit their controls from the hosting environment
- Research enclaves for sensitive and limited datasets
- Streaming and batch ingestion from legacy and modern sources
- Clinical and claims data processing at population scale
- NIST SP 800-171 aligned controls with automated evidence collection
- Observability, monitoring, and documented incident response procedures
NAICS 541715
Life & health sciences R&D
Study design and analysis on data agencies already hold. From forecasting an outbreak to measuring whether a treatment worked: retrospective and observational research, clinical outcomes, epidemiological modeling, and health services research for government health programs. Scenario models that quantify what an intervention buys, time-to-event analysis that shows who benefits, and statistics rigorous enough to publish and reproducible enough to audit.
Influenza admissions per 100,000, Florida
Weeks from the start of October. Hover to compare the seasons at any week.
- This season
- Earlier seasons
- Projection, 80% band
Source: CDC, Weekly Hospital Respiratory Data (HRD) Metrics by Jurisdiction, National Healthcare Safety Network (NHSN). Laboratory-confirmed new admissions per 100,000 population. The 2023–24 season starts in December, when the per-capita series begins. Projection: a log-linear fit to the last five weeks, with an 80% band; a sketch of the recent growth, not a forecast.
Chronic obstructive pulmonary disease among adults
Crude prevalence, 2023, for 122 Tampa Bay ZIP code areas. Five equal-count classes; outlined areas are statistical hot and cold spots.
Source: CDC, PLACES: Local Data for Better Health, ZCTA estimates published in 2025, model-based from the 2023 Behavioral Risk Factor Surveillance System. Boundaries: U.S. Census Bureau TIGER ZCTAs, simplified. Hot and cold spots: Getis–Ord Gi* over each area’s eight nearest neighbours.
Readmission-free after inpatient mental health care, care-transition programme against usual care
142 and 139 veterans followed for up to 24 months. Ticks mark censored patients. Hover to read both curves at any month.
- Care-transition programme
- Usual care
- Time gained, to the horizon
- more readmission-free time per patient over 12 months, about 46 days
- +1.5 mo
- more readmission-free time per patient over 12 months, about 46 days
- average readmission-free time, programme, out of 12
- 9.8 mo
- average readmission-free time, programme, out of 12
- average readmission-free time, usual care, out of 12
- 8.3 mo
- average readmission-free time, usual care, out of 12
- more still out of hospital at 24 months
- 26 pts
- more still out of hospital at 24 months
Adjusted hazard ratios for readmission
From a proportional-hazards model of the same patients. Left of 1 is a lower hazard; bars are 95% intervals.
Suicide prevention: funding by fiscal year
Projects whose title or terms match the topic, by the institute or office that funds them.
Source: NIH RePORTER, project search, fiscal years 2016–2025. Projects whose title or terms match the topic (suicide), by funding agency; a project counts once per fiscal year it is funded. RePORTER lists VA projects without award amounts, so VA is shown by project count only.
What we deliver
- Retrospective and observational study design on existing data
- Clinical outcomes and comparative-effectiveness research
- Epidemiological modeling, surveillance and geospatial analysis
- Health services research: access, utilization and cost
- Time-to-event analysis and hazard estimation
- Publication-grade statistical analysis, reproducible end to end
NAICS 541720
Health services research & methods
Health services research on access, utilization and cost, and the methods that make secondary data answer a question defensibly — phenotyping, confounding control, bias analysis, and privacy-preserving analytics. Data collected for care and payment was not collected for the question in front of you; the method is what closes that gap, and it is written down before the first query runs. Every workflow is reproducible end to end, because research that can't be rerun can't be trusted.
Covariate balance, care-transition programme against usual care
2,514 programme patients and 2,295 usual-care patients. Absolute standardized mean difference; under 0.1 is balanced.
Propensity score overlap
Effect on 30-day readmission
Crude: +5.3 percentage points (95% CI +3.4 to +7.3)
- Before weighting
- After weighting
- Programme
- Usual care
What we deliver
- Cohort and case definition on EHR and claims data
- Confounding control: propensity methods, weighting and matching
- Bias analysis and sensitivity analysis
- Missing-data methods and measurement-error assessment
- Privacy-preserving analytics and de-identification
- Survey design, data collection and instrument validation
- Methods documentation a reviewer can audit
NAICS 621112
Clinical services
Physician-led clinical services for federal health programs: medical evaluation and screening, and behavioral and mental health services — documented for audit and for reuse, with the same measurement discipline as the rest of this page.
What we deliver
- Medical evaluation and screening examinations
- Behavioral and mental health services
- Clinical review of records and cases
- Clinical documentation built for audit and data reuse
- Quality and outcome measurement for clinical programs
- Clinical input to system and data design
NAICS 541690
Scientific & technical consulting
Physician-led advisory to federal health programs: framing the clinical question, designing the protocol that can answer it, and serving as principal investigator or subject-matter expert on the work that follows. The judgment being sold is clinical — what an outcome measure actually captures, which confounders matter in a population, and what a result does and does not license a programme to conclude.
What we deliver
- Clinical advisory and protocol design for federal health research
- Principal investigator and subject-matter expert support
- Outcome-measure selection and endpoint definition
- Study feasibility, sample-size and power assessment
- Independent review of research designs and analytic plans
- Translation of research findings into programme decisions
How we work
Compliance is part of the design, not a stage after it.
HIPAA
Privacy by design
De-identification, minimum-necessary access, and audit trails built into every data flow.
NIST SP 800-171
Controls as code
Security controls implemented as code with evidence collected automatically, so accreditation is not a retrofit.
Section 508
Accessible by default
WCAG 2.2 AA conformance designed in and tested with assistive technology.