Proof, not promises

Case studies

Real data challenges and how we approached them — different sources, different methods, same goal: something a public health partner can act on.

Cross-system linkage

Linking health records across Minnesota health systems — without sharing patient data

A working demonstration of how 11 health systems can contribute to statewide prevalence estimates while patient records stay on-site.

Problem
A statewide consortium needed timely, stratified condition estimates across multiple health systems. No single organization could hold all patient records, and privacy and governance rules ruled out centralizing identifiable data.
Approach
Standardized clinical data at each contributing site, privacy-preserving record linkage, administrative enrichment, federated cohort queries, and small-cell suppression before any aggregate is published.
Result
A runnable end-to-end pipeline and interactive walkthrough — evidence of the infrastructure Stratum Health designs for consortia and public health networks.

Skills demonstrated

  • Data pipeline engineering
  • Privacy-preserving linkage
  • Federated analytics
  • Governance & suppression
  • Dashboard-ready outputs
View case study →

Health equity geography

When housing policy becomes a health map

Interactive map comparing 1930s HOLC redlining grades with CDC tract-level health estimates in the Minneapolis–St. Paul metro.

Problem
Partners needed to demonstrate measurable present-day health differences tied to historical housing investment policy — not just cite the literature.
Approach
1930s HOLC redlining grades spatially joined to modern census tracts and CDC PLACES model-based estimates, with an interactive public map and correlation chart.
Result
A public-facing tool comparing formerly redlined and highest-graded neighborhoods across diabetes, asthma, and mental distress in the Twin Cities metro.

Skills demonstrated

  • Geospatial analysis
  • Historical data integration
  • Health equity framing
  • Public dashboard design
  • CDC open data
View case study →

Population estimation & synthesis

Estimating health condition burden among people experiencing homelessness

A reusable methodology for county/CoC-level burden estimates when standard surveillance does not cover homelessness intersected with chronic disease — illustrated with public HUD PIT data and literature-derived prevalence rates.

Problem
Homeless service providers and public health agencies know population counts from PIT and HMIS, but lack reliable estimates of health condition burden in that population. Standard surveillance datasets like CDC PLACES do not cover homelessness intersected with chronic disease — making it hard to justify resources, target outreach geographically, or support grant applications with credible numbers.
Approach
Combine HUD Point-in-Time counts for local denominator and geographic distribution; apply published peer-reviewed prevalence rates with explicit confidence ranges (labeled as literature-derived estimates, not measured counts); overlay illustrative shelter and clinic locations to flag service deserts; and document optional triangulation against discharge or ED data where homeless status is recorded.
Result
A reusable methodology for producing defensible, geographically specific health burden estimates for a population standard surveillance systems do not capture — applicable to any county or Continuum of Care region.

Skills demonstrated

  • Population estimation & synthesis
  • HUD data integration
  • Literature-derived modeling
  • Service access mapping
  • Grant-ready reporting
View case study →

Public data synthesis

County health snapshot in one page

Public lookup tool that pulls CDC PLACES, Census demographics, and County Health Rankings into a shareable county profile — with PDF export and optional email delivery.

Problem
Community health planners and grant writers need a credible county-level picture fast — but the underlying sources live in different portals, use inconsistent formats, and take time to assemble by hand.
Approach
Automated fetch from CDC PLACES, U.S. Census ACS, County Health Rankings, and (for Minnesota) MDH vital statistics; cached per county with stale-data refresh; single-county map boundary; formatted PDF and Resend email on request.
Result
A no-login public tool at a stable FIPS URL — suitable for CHNA scoping, grant appendices, and partner conversations, with a path to tract-level infrastructure for organizations that need more.

Skills demonstrated

  • Multi-source ETL
  • Public open data APIs
  • Geospatial visualization
  • PDF generation
  • Lead capture & rate limiting
View case study →

Food systems geography

Mapping food access across north-central Minnesota

Interactive census-tract map of USDA food access indicators for Beltrami, Hubbard, Clearwater, Cass, and Itasca counties — with tribal nation boundaries and county agricultural context — built for food-systems grant applications and food council presentations.

Problem
A food-systems consulting partner working with farmers and food councils in north-central Minnesota needed to show grant reviewers where food access need is concentrated — but the evidence lives in separate federal datasets that none of their audiences can easily combine or map.
Approach
USDA Food Access Research Atlas tract flags joined to Census boundary files and ACS poverty and vehicle-access estimates, overlaid with Red Lake, White Earth, and Leech Lake reservation boundaries and county-level context from the 2022 Census of Agriculture and County Health Rankings — pre-processed into static map layers so the page loads fast with no live API calls.
Result
A single shareable map showing that 13 of the region’s 30 census tracts are flagged low-income and low-access — ready for grant appendices, food council meetings, and partner conversations.

Skills demonstrated

  • Geospatial ETL
  • USDA & Census open data
  • Tribal land context
  • Choropleth design
  • Grant-ready reporting
View case study →

Traditional foods & habitat discovery

Traditional foods habitat maps

One habitat-capability product for traditional foods in north-central Minnesota (Beltrami, Cass, Clearwater, Hubbard, and Itasca). Live screens cover Manoomin (Wild Rice), Makizin-waabigwan (Showy Lady’s Slipper), Bagaaniminzh (Hazelnut), Wiigwaasaatig (Paper Birch), Ininaatig (Sugar Maple), plus trailing arbutus (Mayflower) and chokecherry — always capability, never occurrence. Manoomin leads with the 1850s presettlement reference, present-day basin scoring, and named barriers; other guilds share the same CARE-aligned framing.

Problem
Food-as-medicine and stewardship work in north-central Minnesota keeps arriving at the same place: the region’s traditional food and plant base was displaced, and communities need a way to ask — at regional scale — which waters and biomes could still support Manoomin (Wild Rice), Makizin-waabigwan (Showy Lady’s Slipper), and related guilds, and what is actually in the way. Occurrence maps are the wrong tool; capability screens are the right one.
Approach
Start with a restoration-potential model that subtracts rather than predicts: Marschner’s reconstruction of presettlement vegetation establishes what the land held before 1850; DNR hydrography supplies basin morphometry and the drained/impounded classification that names present-day barriers; current DNR-identified rice waters are held out of scoring as an independent check. The same capability-only pattern extends across guilds on one shared map shell. Next phases replace morphometry proxies with Minnesota Gen2 lidar derivatives (topographic wetness, flow, canopy) and public/partner orthomosaics, with taxa and thresholds confirmed with tribal natural resources partners.
Result
A partner-review habitat product for tribal natural resources staff. Manoomin (Wild Rice): 6,598 DNR water basins in five counties, 1,024 biophysically capable with mapped presettlement evidence — 32 with a barrier identified, including MPCA sulfate listings — and a published spatial-holdout validation (bands set on Beltrami and Hubbard rice waters; on the held-out counties 89% of DNR rice waters rate capable against a 58% base rate, AUC 0.80). Plant guilds: 31,750 one-kilometre land hexagons scored on present-day DNR Native Plant Community surveys where they exist and the 1850s map elsewhere, with the historical screen validated against the survey. Built to CARE Principles for Indigenous Data Governance: capability, never occurrence.

Skills demonstrated

  • Geospatial ETL
  • Habitat suitability modeling
  • Lidar / remote-sensing roadmap (Gen2 DEM, TWI, orthomosaics)
  • Indigenous data governance (CARE)
  • MN DNR & Geospatial Commons data
  • Food-as-medicine framing
View case study →

Forest health & hydrology

Black ash swamp vulnerability to emerald ash borer

A warning map for north-central Minnesota (Beltrami, Cass, Clearwater, Hubbard, and Itasca): where black ash is abundant, how soon emerald ash borer is likely to arrive, and where losing the ash would change the hydrology most. Three axes kept separate — host, exposure, consequence — each hexagon reported as a category with a named driving factor.

Problem
Black ash is a foundation species of northern Minnesota wet forests: it moves enormous volumes of water, and where it dies the water table rises and forested swamp converts to sedge marsh. Emerald ash borer kills essentially every ash it reaches, and its front has not yet arrived in these counties. Partners need to know which stands matter most before it does, not a post-mortem afterwards.
Approach
MN DNR forest inventory species volumes (state land) and DNR Native Plant Community ash-swamp classes describe the host, with a replacement-capacity term from the non-ash species present; MDA emerald ash borer detections, quarantines and the MDA introduction-risk model describe exposure through a recency-weighted distance-decay fitted and tested statewide on a temporal holdout; SSURGO drainage, Histosol and hydric shares and the Minnesota wetland inventory describe hydrologic consequence. CDC PLACES cardiovascular and respiratory estimates are joined by tract as an adjacency, never a causal claim.
Result
A partner-review build on 1 km hexagons across five counties with the validation numbers on the page: the spread model is scored on detections it never saw (rank AUC and base rate reported together), the two independent ash datasets are cross-checked and their agreement rate published, and the consequence axis is labelled as a modelled projection because no post-mortality outcome data exist for Minnesota black ash swamps.

Skills demonstrated

  • Geospatial ETL
  • Invasive-species spread modelling
  • Temporal holdout validation
  • Soils & wetland hydrology
  • MN DNR, MDA & USDA open data
  • Health adjacency framing (CDC PLACES)
View case study →