Platform
Two independent measures. One priority map.
How the access score and the vulnerability index are built, why they're kept apart, and what still needs work.
The two axes
Axis 1 — Spatial access to healthcare
A two-step floating catchment area (2SFCA) score: not facility counts, but how many facilities a population can reach within 30 or 60 minutes of travel, discounted by how many other people compete for the same facilities.
Inputs
- WorldPop 100m gridded population
- OpenStreetMap facility locations and road network
- MOH facility statistics, used for validation
Axis 2 — Vulnerability
A constructed index combining three components, standardised and combined by principal component analysis (PCA) with published loadings. The reasoning: these compound one another. A community that is poor, heat-exposed, and already managing chronic illness carries far more health need than one that is merely distant, because chronic conditions demand sustained nearby care and distance turns that management into emergency presentation.
Components
- Socioeconomic: Census data on education, housing, and income.
- Environmental exposure: Satellite data on PM2.5 air pollution, heat, and greenery (NDVI, Normalized Difference Vegetation Index).
- Chronic disease burden: Principally diabetes and hypertension, from national health surveys.
The quadrant
Every unit is plotted on both axes, splitting the map into four groups. The two numbers are deliberately kept separate rather than merged into one score, because the quadrant tells a planner not just where the problem is but which problem they are looking at — and those need different interventions.
Priority
High vulnerability, low access
Far from care and least equipped to cope with being far. The target group for siting and outreach.
Served but vulnerable
High vulnerability, good access
Access isn't the barrier here — the underlying vulnerability is. Needs outreach and chronic-care support, not new facilities.
Remote but resourced
Low vulnerability, low access
Distant, but the population can generally absorb it. Lower priority than the priority quadrant.
No identified gap
Low vulnerability, good access
Neither access nor vulnerability stands out. Not a priority for new intervention.
Why two axes, not one score
A single blended score would be easier to rank — and would hide the thing a planner most needs to know. Two regions can produce the same combined score for entirely different reasons: one because it's remote but resilient, the other because it's nearby but overwhelmed by need. Collapsing that distinction into one number erases the difference between 'build a clinic here' and 'fund outreach and chronic-disease management here.' Keeping the axes separate keeps that distinction visible.
Validation approach
Because chronic disease sits inside the vulnerability score, it cannot also be used to test that score. Validation instead relies on three approaches, in order of what's currently feasible:
- 1
Facility validation
The OpenStreetMap facility layer is checked against official MOH figures, region by region, producing a coverage ratio that shows where the map is reliable.
- 2
Leave-one-component-out sensitivity analysis
The vulnerability index is recomputed with each component removed in turn, to see how much any single input drives the result.
- 3
Expert face-validity review
Regional public health practitioners review whether the classification matches what they observe on the ground.
- 4
Clinical outcome validation (later phase)
Testing the index against measured outcomes such as emergency department utilisation and avoidable admissions, once institutional data access is in place.
The two tiers
Tier 1 — Planning tier
In developmentPlanner- and researcher-facing. Built entirely on static, open data. Buildable now, and complete in itself. This is Phase 1.
Tier 2 — Patient tier
Permission-dependentPatient-facing: a map showing nearby emergency departments and how busy each currently is, plus a short triage questionnaire feeding an estimated wait time — so a patient isn't forced to default to the closest ED when a quieter one is reachable.
Fixed safety constraints
- Anyone reporting chest pain, stroke symptoms, or other red-flag presentations is instructed to call 997 and attend the nearest facility, with all routing suppressed.
- Wait times, when shown, are always ranges — never precise figures.
- This tier requires live occupancy feeds from hospitals, SFDA clearance as clinical decision support, and emergency physician oversight — none of which currently exist. The realistic entry point is a single-cluster pilot, not a national launch.
- This tier is not available. Nothing on this site should be used to decide where to seek care.
Data sources
| Source | What it provides | Resolution | Licence |
|---|---|---|---|
| WorldPop | Gridded population counts | 100m | CC BY 4.0 |
| OpenStreetMap | Facility locations, road network | Point / vector | ODbL |
| GHSL (Global Human Settlement Layer) | Built-up area and settlement classification | 100m | CC BY 4.0 |
| VIIRS (Visible Infrared Imaging Radiometer Suite) | Night-time lights, as an economic activity proxy | ~500m | Public domain (NOAA) |
| ACAG PM2.5 (Atmospheric Composition Analysis Group) | Fine particulate air pollution | ~1km | Open, attribution required |
| ERA5 | Reanalysis climate data, including heat exposure | ~9km, hourly | Copernicus open licence |
| MODIS NDVI (Normalized Difference Vegetation Index) | Greenery / vegetation cover | 250m | Public domain (NASA) |
| GASTAT | 2022 census population and socioeconomic indicators | Region / governorate | Open government data |
| MOH Statistical Yearbook | Facility counts, used for validation | Region | Open government data |
| HDX (Humanitarian Data Exchange) | Administrative boundary polygons | ADM1 (region) | Varies by dataset, open |
Limitations
- OpenStreetMap facility coverage is uneven and weakest in rural areas; the validation module quantifies this rather than hiding it.
- Chronic disease data exists only at 13-region resolution with wide confidence intervals; governorate-level small-area estimation is the priority methodological upgrade.
- Thirteen regions is too few units for stable spatial modelling; governorate-level analysis (roughly 139 units) is strongly preferred, contingent on obtaining boundary polygons.
- Phase 1 has no external validation against clinical outcomes; this is stated, not hidden, and is planned for a later phase.
- Income is not a census variable in Saudi Arabia and may need to be proxied by other socioeconomic indicators.