Methods & Data Sources

Back to maps

01Hospital Identification

United States: 1,248 PCI-capable hospitals

I identified U.S. hospitals with demonstrated PCI capability from Medicare inpatient claims data using Diagnosis-Related Group (DRG) codes 246 through 251, which capture percutaneous cardiovascular procedures with and without drug-eluting stent, with and without major complication or comorbidity.

This billing-based approach differs from the methodology used by Nallamothu et al. (2006), who relied on the American Hospital Association Annual Survey — a self-report instrument. Billing-based identification captures only hospitals with demonstrated procedural activity, eliminating hospitals that report capability but have discontinued active programs. The tradeoff is that some hospitals serving predominantly non-Medicare populations may be underrepresented, and the method cannot distinguish between hospitals maintaining 24/7 primary PCI capability and those performing only elective PCI during daytime hours.

Each hospital was geocoded to latitude and longitude coordinates using the Medicare Provider of Services file. Hospitals in U.S. territories were excluded.

Canada: 41 PCI centers

I identified Canadian PCI centers through multiple provincial sources because Canada has no single public database analogous to Medicare claims for cardiac procedure activity:

• Canadian Institute for Health Information (CIHI) — Cardiac Care Quality Indicators report
• CorHealth Ontario — Ontario's cardiac, stroke, and vascular network
• Alberta APPROACH Registry — Alberta Health Services Cardiac Sciences Program
• BC Cardiac Registry — British Columbia cardiac services network
• Provincial equivalents for Quebec, Nova Scotia, Manitoba, and Saskatchewan

This yielded 41 PCI centers. The count should be understood as the best available approximation from public sources. I found no PCI-capable hospitals in Prince Edward Island, New Brunswick (though patients may be routed to Nova Scotia), or the three territories.

02Geographic Units & Population Data

United States: 84,114 census tracts

The geographic unit of analysis is the census tract, as defined by the 2020 U.S. Decennial Census. Census tracts average roughly 4,000 residents and are designed to be relatively homogeneous with respect to population, economic status, and living conditions. They provide substantially greater granularity than the 3,143 U.S. counties, which range from 26 square miles (Arlington County, Virginia) to over 145,000 square miles (North Slope Borough, Alaska).

Population-weighted centroids for each tract were obtained from the Census Bureau CenPop2020 file, derived from 2020 block-level population data.

Canada: 293 census divisions

The geographic unit for Canada is the census division (CD), as defined by Statistics Canada for the 2021 Census. Census divisions are groups of neighboring municipalities joined together for regional planning. They are substantially larger geographic units than U.S. census tracts, which limits the granularity of the Canadian analysis. The analysis covers all 293 census divisions across 10 provinces and 3 territories, representing a total population of 36,991,981.

Population-weighted centroids for Canadian census divisions were computed from Statistics Canada's Geographic Attribute File (GAF), which provides dissemination block–level population and representative point coordinates. For each census division:

Centroid_lat = Σ(block_lat × block_pop) / Σ(block_pop)
Centroid_lon = Σ(block_lon × block_pop) / Σ(block_pop)

I computed this across 384,286 dissemination blocks from the 2021 Census. The approach places the representative point where people actually live rather than at the geographic center — a distinction that matters enormously for large northern divisions where the geometric center may lie hundreds of miles from the nearest settlement.

03Why Population-Weighted Centroids Matter

Analyses relying on geometric centroids systematically overestimate drive times for large, sparsely populated units, and thereby risk overstating the rural access problem at exactly the geographic scale where it appears most alarming.

04Drive Time Estimation

For each census tract (U.S.) or census division (Canada), I computed the straight-line (haversine) distance from the population-weighted centroid to the nearest PCI hospital using a KD-tree nearest-neighbor search. I then converted this distance to a time estimate using two approaches, both of which I present for both countries.

Approach 1: Estimated drive time

Estimated drive time measures how long it takes to drive from the community centroid to the nearest PCI hospital, accounting for the fact that roads are longer than straight-line distance and that driving speeds differ between urban and rural settings.

Drive time = (straight-line distance × 1.417 road factor) / speed × 60 min/hr

Speed = 50 mph if straight-line distance < 50 miles (urban/suburban)
Speed = 40 mph if straight-line distance ≥ 50 miles (rural/highway)

The 1.417 road factor is an empirical multiplier converting straight-line to estimated road distance, derived from the literature comparing haversine to road-network distances across a range of U.S. geographies. The speed differential reflects the observation that shorter trips tend to involve lower-speed local roads, while longer trips involve higher-speed but more circuitous highway routes.

Approach 2: Nallamothu prehospital time

The Nallamothu et al. (2006) prehospital formula estimates total time from 911 call to hospital arrival, adding ambulance dispatch delay, a round-trip driving factor (the ambulance must drive to the patient before driving to the hospital), and on-scene time to the raw drive time. The formula differs for urban and rural settings:

Urban (tract density > 5,000/sq mi for U.S.; > 213/sq mi for Canada):
Prehospital time = 1.4 min (dispatch) + drive time × 1.6 (round-trip factor) + 8 min (scene)

Rural (below density threshold):
Prehospital time = 2.9 min (dispatch) + drive time × 1.4 (round-trip factor) + 9 min (scene)

Component	Urban	Rural	What it captures
Dispatch time	1.4 min	2.9 min	911 call to ambulance departure
Round-trip factor	×1.6	×1.4	Ambulance drives to patient, then to hospital
Scene time	8 min	9 min	On-scene assessment and stabilization

What each approach measures

The two approaches answer different questions. Estimated drive time asks: how long would it take to drive from this community to the nearest PCI hospital? Nallamothu prehospital time asks: how long from the moment a patient calls 911 until they arrive at a PCI hospital? The Nallamothu formula produces substantially longer time estimates because it includes fixed overhead (dispatch, scene time) and a round-trip driving factor. I present both for both countries so the reader can evaluate the comparison under either framing.

Time bands

Band	Threshold	Clinical interpretation
< 30 minutes	Timely access	Well within the ACC/AHA guideline window for primary PCI
30–90 minutes	Transfer window	Suboptimal but within guideline window for transferred patients if networks are optimized
> 90 minutes	Beyond guideline	Primary PCI cannot be delivered within evidence-supported time windows; candidates for pharmacoinvasive strategy

05Bivariate Classification

I assigned each geographic unit to one of nine mutually exclusive categories defined by the intersection of time band and population density tier.

Density thresholds

Country	Sparse	Moderate	Dense
United States	< 500/sq mi	500–5,000/sq mi	> 5,000/sq mi
Canada	< 17/sq mi	17–213/sq mi	> 213/sq mi

I used Canada-specific thresholds because Canada's overall settlement density is sufficiently low that applying U.S. thresholds would collapse nearly all Canadian census divisions into the sparse category. Canada's dense threshold (213/sq mi) — which captures the top 15% of census divisions by count, containing 66% of the population — falls below the U.S. sparse threshold of 500/sq mi (the U.S. 30th percentile).

The 3×3 matrix — policy interpretation

	>90 min	30–90 min	<30 min
Dense	Geographic anomalies (US); infrastructure gap (CA)	Urban fringe	Urban core — well served
Moderate	Policy target for PCI expansion	Suburban fringe — transfer networks	Suburban — well served
Sparse	Pharmacoinvasive territory	Rural — within transfer window	Rural — near existing PCI

06Key Results

United States — estimated drive time

Band	Population	Share
< 30 min	259M	79%
30–90 min	62.3M	19%
> 90 min	6.4M	2.0%

The 6.4 million beyond 90 minutes of drive time: 4.7M sparse (1.4%), 1.6M moderate (0.5%), 108K dense (<0.1% — geographic anomalies: Hawaii, Florida Keys, Missoula corridor).

United States — Nallamothu prehospital time

Band	Population	Share
< 30 min	154.5M	47.1%
30–90 min	145.8M	44.5%
> 90 min	27.6M	8.4%

Under the full Nallamothu formula, the proportion beyond 90 minutes increases from 2.0% to 8.4%. The communities identified as underserved are the same — the additional 21 million people pushed beyond 90 minutes are predominantly moderate-density tracts in the 30–90 minute drive-time band that cross the 90-minute threshold once dispatch, round-trip, and scene overhead are added. The geographic pattern does not change; the threshold shifts.

Canada — estimated drive time

Band	Population	Share
< 30 min	25.6M	69.2%
30–90 min	4.4M	12.0%
> 90 min	7.0M	18.8%

Canada — Nallamothu prehospital time

Band	Population	Share
< 30 min	18.9M	51.0%
30–90 min	9.1M	24.5%
> 90 min	9.1M	24.5%

The 9.1 million beyond 90 minutes of Nallamothu prehospital time: 4.1M sparse (11.0%), 3.7M moderate (9.9%), 1.3M dense (3.5% — 11 census divisions including Francheville/Trois-Rivières, Sherbrooke, and Nanaimo).

U.S.–Canada comparison

07Limitations

Drive time estimation. Both drive time and prehospital time are estimates I derived from straight-line distance with empirical correction factors, not real-time or road-network routing. They do not account for traffic congestion, seasonal road conditions (particularly relevant in northern Canada), road closures, or EMS-specific routing.

Urban speed assignment for Canada. The Nallamothu formula assigns urban speed parameters (35 mph) to census divisions above Canada's dense threshold (> 213/sq mi). For intercity highway drives — such as Trois-Rivières to Quebec City (66 straight-line miles of predominantly highway driving) — this likely understates actual travel speed and overstates prehospital time. This bias makes the Canadian numbers conservative: the true access picture is likely somewhat better than reported. I retained the formula as specified rather than introducing ad hoc speed adjustments, accepting that this makes the comparison more generous to Canada, not less.

U.S. hospital identification. Medicare DRG billing identifies hospitals with demonstrated PCI activity but does not verify 24/7 primary PCI capability. Some identified hospitals may perform only elective daytime PCI, which would overestimate true emergency access for the U.S.

Canadian hospital identification. The 41-center count relies on multi-source aggregation from provincial cardiac network directories rather than a single standardized administrative dataset. It is the best available approximation from public sources but may miss or double-count facilities.

Geographic granularity mismatch. U.S. census tracts (~4,000 residents, ~84,000 units) are far more granular than Canadian census divisions (~126,000 residents average, 293 units). The Canadian analysis cannot capture within-division variation — a large census division may contain both a well-served urban core and an underserved rural periphery.

No incidence weighting. This analysis characterizes access for the general population, not weighted by STEMI incidence. Age, sex, race, and socioeconomic composition vary across density tiers and drive-time bands in ways that affect the expected burden of acute coronary events.

No air transport. This analysis considers ground transport only. Helicopter EMS can dramatically reduce effective transport times for remote communities.

08Data Sources

Source	URL	What it provides
Medicare Inpatient Hospitals by Provider and Service	data.cms.gov	DRG 246–251 billing → 1,248 PCI hospitals
Medicare Provider of Services File	data.cms.gov	Hospital geocoding (lat/lon)
U.S. Census Bureau, 2020 Census	census.gov	Tract population and land area
Census Bureau Centers of Population	census.gov	Population-weighted tract centroids
Statistics Canada, 2021 Census	statcan.gc.ca	CD population and land area
Statistics Canada GAF	statcan.gc.ca	Dissemination block centroids → pop-weighted CD centroids
CIHI Cardiac Care Quality Indicators	cihi.ca	Canadian PCI center identification
CorHealth Ontario	corhealthontario.ca	Ontario cardiac network
Alberta APPROACH Registry	approach.org	Alberta cardiac sciences
BC Cardiac Registry	bccardiacregistry.com	BC cardiac services

09References

1. Nallamothu BK, Bates ER, Wang Y, Bradley EH, Krumholz HM. Driving times and distances to hospitals with percutaneous coronary intervention in the United States: implications for prehospital triage of patients with ST-elevation myocardial infarction. Circulation. 2006;113(9):1189–1195.
2. O'Gara PT, Kushner FG, Ascheim DD, et al. 2013 ACCF/AHA Guideline for the Management of ST-Elevation Myocardial Infarction. Circulation. 2013;127(4):e362–e425.
3. Ibanez B, James S, Agewall S, et al. 2017 ESC Guidelines for the management of acute myocardial infarction in patients presenting with ST-segment elevation. European Heart Journal. 2018;39(2):119–177.
4. Hsia RY, Shen YC. Possible geographical barriers to trauma center access for vulnerable patients in the United States. Archives of Surgery. 2011;146(1):46–52.
5. Armstrong PW, Gershlick AH, Goldstein P, et al. Fibrinolysis or primary PCI in ST-segment elevation myocardial infarction. NEJM. 2013;368(15):1379–1387.
6. Schneider EC, Shah A, Doty MM, et al. Mirror, Mirror 2024: A Portrait of the Failing U.S. Health System. Commonwealth Fund. 2024.

10Interactive Maps & Data

All maps, data files, and analysis code are available at:

anishkoka.github.io/pci-access-maps →

The interactive bivariate maps allow filtering by any of the nine density × drive-time categories, toggling PCI hospital locations, and hovering over individual geographic units for detailed population, density, and prehospital time data.