About this dashboard
This dashboard maps three interlocking dimensions of US cardiovascular care at county and
census-tract resolution: the supply of cardiologists, the burden of cardiovascular disease in the
population, and the digital infrastructure that supports remote-care delivery.
Digital-health targeting (primary layer)
The default map answers a targeting question: among US communities of greatest
cardiovascular need, which are ready for immediate digital-health deployment,
and which first require infrastructure investment?
We first define a single high-need county pool, the union of:
(a) counties with NO cardiologists in the latest panel year, and
(b) counties with a DECLINING PER-CAPITA cardiology workforce
(OLS slope of cardiologists per 100,000 < 0 over 2010-2023;
counties with mean < 3 use the last-vs-first rate)
Within that pool, at census-tract resolution, two anchored cuts form a
two-by-two matrix (the framing the manuscript uses):
high burden : burden_z > 0 (above the national-average tract)
ready : DDI ≤ national median (18.77)
DEPLOY = high burden AND ready → digital tools can reach patients now
INVEST = high burden AND not ready → infrastructure must improve first
To rank tracts within each group, burden and readiness are standardized
across the pool and summed into a priority score: burden minus readiness
for deployment, burden plus readiness for investment.
Counties are shaded by their tract mix: Deploy-lean, Invest-lean,
or Mixed (containing both target types, the within-county heterogeneity that
county-level analysis hides). Zoom in or open a county to see its individual tracts.
The two boxes are disjoint; every target tract is either a deployment or an
investment target, never both.
Data sources
- Cardiology workforce, 2010-2023. Non-federal MDs in the AMA "Cardiovascular Disease" specialty, derived from the Area Health Resources Files (HRSA).
- Disease burden, 2022-2023. Tract-level age-adjusted/crude prevalence of 10 cardiovascular outcomes and risk factors from CDC PLACES (HIGHCHOL, BPHIGH, CHD, DIABETES, OBESITY, STROKE, CSMOKING, LPA, SLEEP, BINGE).
- Digital readiness. The tract-level Digital Divide Index (DDI), composed of infrastructure/adoption (INFA) and socioeconomic (SE) sub-scores. Higher DDI = greater digital divide = lower readiness.
Workforce trend
For each county c, we fit an ordinary least squares slope βc of cardiologist counts
over the panel years. A minimum-count gate (mean ≥ 3) excludes very small counties whose trends
would be dominated by noise.
trend = growing if β > 0
declining if β < 0
stagnant if β = 0
insufficient_data otherwise
Disease burden trend
For each tract i and year t, we standardise each PLACES measure m within the year:
z(i,m,t) = ( p(i,m,t) − μ(m,t) ) / σ(m,t)
Then average across measures to form a composite z-score, and fit an OLS slope over time:
Z(i,t) = mean over m of z(i,m,t)
β_burden(i) = OLS slope of Z(i,t) on t
Quadrant classification
Each tract carries its own burden trend and its parent county's workforce trend:
| Workforce growing | Workforce declining / stagnant |
|---|
| Burden increasing | Q2 Recovering | Q1 High risk |
|---|
| Burden stable / decreasing | Q3 Low risk | Q4 Watch |
|---|
Digital readiness tiers
National tercile cutoffs of the tract-level DDI define three tiers (high readiness = best-served):
DDI ≤ 33rd pctl → Tier 1: high readiness (deploy)
33rd < DDI ≤ 66th → Tier 2: moderate
DDI > 66th pctl → Tier 3: low readiness (invest)
Limitations
- AHRF cardiology counts are county-level only; every tract within a county inherits the same workforce trend.
- Tract-level PLACES publishes only crude prevalence, not age-adjusted.
- Burden trend uses two years (2022-2023); workforce trend uses up to seven years (2010-2023).
- 78% of counties (predominantly rural) carry too few cardiologists to support trend classification under the minimum-count gate.
CarDS Lab, Yale School of Medicine. Pipeline code and methodology: see lab website.