PaperlayerView exact revision on arXiv

Source-linked AI summary

Global maps of travel time to emergency and tertiary hospitals

Johan Emilsson

arXiv:2609.12696v1physics.soc-phcs.CY

TL;DR

Global access estimates have treated facilities alike, despite a lack of global road-network estimates distinguishing documented emergency and tertiary capability. This study routes drive times to OSM hospitals by nested capability tier and weights coverage with the 2025 population grid. It finds that estimated one-hour coverage declines from 96.2% for any hospital to 78.5% for documented emergency care and 40.3% for a tertiary marker.

  • Problem

    Global access estimates treat facilities alike, and no known global estimate combines hospital-only coverage, documented emergency capability and road-network routing.

    Source: §1 · pp. 1–2

  • Method

    The study routes drive-time isochrones from OSM hospitals, dissolves them for nested capability tiers and overlays them with the 2025 population grid.

    Source: §1 · p. 2

  • Results

    96.2% of the world's population is within 60 minutes of a hospital, 78.5% of a documented emergency hospital and 40.3% of a hospital with a tertiary marker.

    Source: §5 · p. 16

  • Takeaways & Limitations

    The estimates indicate that broad access to any hospital does not extend to time-critical care, and counting facilities as equivalent overstates such access.

    Source: §4.1 · p. 14

    Travel times assume free-flow driving with an available vehicle; speeds were not sensitivity-tested, journeys were not validated against observations, and hulls can enclose unreachable pockets.

    Source: §4.4 · p. 16

Abstract

from arXiv · show

Global estimates of access to health care treat every facility alike. Drive time was routed on the OpenStreetMap network from all 217,841 of its hospitals and 15-, 30- and 60-minute isochrones dissolved for nested capability tiers, weighted with GHS-POP 2025. 96.2% of the world's population lives within an hour's drive of a hospital, 78.5% of a hospital with a documented emergency department, and 40.3% of one with a helipad or university name marking tertiary care. The 1.45 billion people within an hour of a hospital but not of an emergency hospital live mainly in South and East Asia and Nigeria; the 312 million beyond an hour of any hospital lie in a belt from the Sahel to Madagascar. Defining the tertiary tier by helipad alone leaves its estimate unchanged, and the isochrone construction shifts the 60-minute figures by a few percent. Counting every facility as equivalent overstates access to time-critical care.

1 Introduction

Travel-time access matters for conditions that worsen with delay, yet global access estimates have treated hospitals and clinics as equivalent. This paper addresses the gap with a road-network estimate that distinguishes hospital capability.

1 Introduction

  • Travel-time access matters because outcomes for severe injury, obstetric haemorrhage, myocardial infarction, stroke and sepsis worsen with time to definitive care.Benchmarks include two hours to a facility able to perform bellwether procedures, a 60-minute trauma-centre target, and condition-specific stroke and coronary-care times.Source: §1 · p. 1
  • The reference global estimate found 91.1% of people within an hour of a hospital or clinic, but treated every facility as equivalent and used a friction surface.The cited estimate's authors noted that travel times to more specialized services could be much greater than its maps indicate; regional comparisons also find raster and network methods can classify different populations as underserved.Source: §1 · p. 1
  • Capability-differentiated access has been estimated nationally and regionally, while a recent global map of essential services again used a friction surface without distinguishing facility levels.Examples include emergency hospitals in sub-Saharan Africa, major hospitals assumed capable of essential surgery, and trauma, thrombectomy and 24-hour emergency departments in high-income countries.Source: §1 · pp. 1–2
  • The paper estimates global access to OSM hospitals by road network, separating any hospital, emergency-tagged hospitals and emergency hospitals with tertiary markers.It reports coverage at 15, 30 and 60 minutes using the 2025 population grid, benchmarks any-hospital coverage against a re-weighted friction-surface estimate, and uses documented emergency tags as a completeness indicator.Source: §1 · p. 2

2 Data and methods

The study builds nested hospital-capability tiers, routes free-flow drive times on the OSM road network, and overlays coverage with population data for country-level estimates and a friction-surface benchmark.

2.1 Hospital inventory and capability tiers

  • The inventory contains 217,841 OSM hospitals; tiers distinguish any hospital, emergency-tagged hospitals and emergency hospitals with tertiary markers.Clinics, doctors' practices and pharmacies are excluded. The emergency tag indicates a documented department, while tertiary markers are a helipad, heliport or runway on the grounds, or a university/teaching name; a separate helipad tier tests dependence on emergency tagging. Helipads were reviewed against aerial imagery across the emergency-tagged set, but university-name markers were not similarly reviewed.Source: §2.1 · pp. 2–3, Table 1 · p. 3

2.2 Road network and drive-time isochrones

  • Hospitals were snapped to a routable OSM graph, and bounded multi-source Dijkstra searches computed outbound travel times from each hospital to 60 minutes.The graph uses selected motor-road classes at default free-flow speeds, overridden by tagged speed limits; service roads, tracks, paths and ferries are excluded, and one-way restrictions are respected. Times omit congestion, road condition, borders and time of day, and treating outbound times as symmetric is an approximation.Source: §2.2 · p. 3
  • Reached road vertices were converted into near-convex concave-hull polygons for 15-, 30- and 60-minute bands, then dissolved by tier and band.The hull approach was chosen for tractability at planet scale; 1,194 hospitals snapped to vertices disconnected from drivable roads. Plausibility checks bounded polygon area, and a quasi-random 2% source sample was compared with a union of 1.5 km vertex buffers, a conservative reference rather than ground truth.Source: §2.2 · pp. 3–4, Figure 1 · p. 5

2.3 Population, land area and country attribution

  • Population and land-area coverage were aggregated by country using the 2025 GHS-POP grid and Natural Earth boundaries, with hospital locations assigned to countries by point-in-polygon.Country attribution uses a nearest-polygon fallback within 0.5 degrees; cells outside all country polygons remain in global totals but not country tables.Source: §2.3 · p. 4

2.4 Benchmark against friction-surface travel time

  • The any-hospital benchmark resamples the Weiss friction-surface travel-time raster to the same 2025 population grid to isolate access-model differences.Population is cross-tabulated by friction-surface 60-minute class and membership in the any-hospital and emergency 60-minute polygons.Source: §2.4 · p. 4

2.5 Indicators

  • Country indicators include population shares within each tier and time band, the any-hospital-to-emergency capability gap, documented-emergency share and hospitals per 100,000 people.Country statistics cover 162 countries and territories with at least one million inhabitants; benchmark countries require friction-surface data for at least 90% of their population, and group aggregates are population-weighted. Figure 1 illustrates dissolved time bands for any hospitals versus emergency-tagged hospitals in East Africa.Source: §2.5 · p. 4, Figure 1 · p. 5

3 Results

The mapped hospital inventory varies substantially across income groups, and coverage declines across emergency and tertiary capability tiers. Geographic and documented-emergency gaps are large, while network and friction-surface estimates agree globally but differ across countries and shift modestly with isochrone construction.

3.1 The hospital inventory

  • Mapped hospital density is 3.4-fold higher in high- than low-income groups, while variation in emergency-tagging reflects tagging practice.The documented-emergency share is 11.7% globally and 3.7% in low-income countries.Source: §3.1 · p. 5

3.2 Global coverage by tier and band

  • At 60 minutes, coverage is 96.2% for any hospital, 78.5% for emergency-tagged hospitals, and 40.3% for tertiary-marker hospitals.Land coverage is 39.4% for any hospital and 11.4% for tertiary-marker hospitals; defining the tertiary tier by helipad alone leaves its population estimate unchanged.Source: §3.2 · pp. 5–6, Table 3 · p. 6

3.3 Geography of the gaps

  • Africa holds 66.1% of people beyond an hour of any hospital, while tertiary-marker coverage ranges from 86.7% in high-income to 5.6% in low-income countries.The uncovered geography changes by tier; for example, 90% of India's population is beyond 60 minutes of a tertiary-marker hospital.Source: Figure 2 · p. 7, §3.3 · pp. 8–9, Figure 3 · p. 9

3.4 The capability gap and the documented-emergency share

  • 1.45 billion people live within an hour of any hospital but not a documented emergency hospital; below about 10% emergency-tag share, coverage may reflect tagging rather than capability.Above about 20% tag share, emergency coverage is close to any-hospital coverage across countries.Source: §3.4 · p. 9, Table 5 · p. 10, Figure 4 · p. 10

3.5 Benchmark against the friction-surface product

  • 96.2% network coverage versus 96.7% friction-surface coverage is close globally, but country estimates differ by more than five points in 34 countries.The comparison combines differences in facility definitions, access models and data vintages, so it cannot isolate the methods.Source: §3.5 · p. 11, Table 6 · p. 11, Figure 5 · p. 12

3.6 Sensitivity to the isochrone construction

  • Changing the hull ratio from 0.85 to 0.3 lowers 60-minute coverage from 96.2% to 95.5% for any hospital, 78.5% to 77.1% for emergency, and 40.3% to 39.3% for tertiary-marker hospitals.Strict buffer-based estimates are lower bounds; the tertiary correction is indicative only, based on 128 sampled hospitals.Source: Table 7 · p. 12, §3.6 · p. 13, Figure 6 · p. 14

4 Discussion

Tiered estimates show that broad hospital access does not imply access to time-critical care, while network and friction-surface estimates diverge most in poorly mapped settings. Emergency-tag completeness, inventory variation, proxy markers, and travel-time assumptions qualify how the coverage figures should be interpreted.

4.1 What the tiers add

  • 1.45 billion people lose 60-minute coverage when access requires a documented emergency department, while only 40.3% are within an hour of a tertiary-marker hospital.The tertiary estimate rests on a helipad marker that gives the same coverage whether or not the emergency tag is required.Source: §4.1 · p. 14

4.2 Network routing and friction surfaces

  • Friction surfaces are more optimistic than network routing in the poorest countries, where unmapped roads can make raster-reachable locations unreachable on a car network.The comparison is confounded by differences in facility sets, access models, and data vintage; both methods may be optimistic relative to actual journeys.Source: §4.2 · p. 15

4.3 The emergency tag as a data-quality indicator

  • Emergency-tag prevalence varies implausibly across countries, so the emergency tier is a lower bound and comparisons are best confined to countries with similar tag completeness.Above roughly 20% documented-emergency tagging, the tagged subset covers almost everyone covered by the full hospital set.Source: §4.3 · p. 15

4.4 Limitations

  • OSM inventory completeness and the meaning of “hospital” vary by country, while university names and helipads are imperfect proxies for tertiary capability.The study also assumes free-flow driving with an available vehicle, and its potential-access polygons do not measure beds, staff, supplies, or cost.Source: §4.4 · pp. 15–16
  • Hull isochrones overstate 60-minute coverage by about 5% for any hospital and 6% for tertiary-marker hospitals, without changing tier ordering.The paper notes that hulls can enclose unreachable pockets or bridge unreached terrain in sparse networks.Source: §4.4 · p. 16

4.5 Future work

  • Future work could replace hulls with buffered-edge isochrones or raster travel-time surfaces, combine OSM with national facility lists, and track successive planet files.These proposals target hull approximation, inventory and attribute gaps, and changes in access versus changes in mapping.Source: §4.5 · p. 16

5 Conclusion

Routing from 217,841 OSM hospitals places 96.2% of the population within a 60-minute drive, but coverage falls for documented emergency and tertiary care. The results distinguish broad hospital access from access to time-critical care and identify mapping and tagging completeness as important qualifications.

5 Conclusion

  • 96.2% of the world's population is within a 60-minute free-flow drive of an OSM hospital, within one percentage point of the friction-surface estimate using the same 2025 population grid.The analysis routes from 217,841 hospitals.Source: §5 · p. 16
  • 78.5% of people are within an hour of a hospital with a documented emergency department, compared with 40.3% for a hospital with a tertiary marker.Defining the tertiary tier by helipad alone leaves its coverage estimate unchanged.Source: §5 · p. 16
  • The 1.45 billion people between the hospital and emergency tiers are concentrated in South and East Asia and Nigeria, and part of the gap reflects data.The 312 million people beyond an hour of any hospital lie in a belt from the Sahel to Madagascar.Source: §5 · p. 16

Data and code availability

The analysis code and derived dataset are available under the Open Database License, while the routing pipeline is specified for reproduction with open tools. Third-party inputs are available from their publishers.

Data and code availability

  • The analysis code and derived dataset are available from the project repository under the Open Database License as an OpenStreetMap derivative.Larger polygon files are provided as release assets from the same location.Source: Data and code availability · p. 16
  • The routing pipeline itself is not released, but the paper specifies it in sufficient detail for reproduction with open tools.Source: Data and code availability · pp. 16–17
  • GHS-POP, Natural Earth, the World Bank income classification, and the Malaria Atlas Project raster are available from their publishers.Source: Data and code availability · p. 17

Declarations

The author reports no specific funding and states that the work was conducted independently of the employer. The study used public aggregate and geographic data and required no ethics approval.

Declarations

  • The work received no specific funding and was carried out in the author's own time.Source: Declarations · p. 17
  • The author is employed by Safeture AB, but the employer did not fund the work or participate in its design, analysis, or writing.Source: Declarations · p. 17
  • The study used publicly available aggregate and geographic data, involved no human participants, and required no ethics approval.Source: Declarations · p. 17
Loading 2609.12696v1…