WHO Global Health Observatory Data Repository

The central health-statistics repository of the World Health Organization, resolved by Datadory into typed, join-ready rows. It is organized around three themes - Universal Health Coverage, Health Emergencies and Health and Well-Being - plus roughly 35 topic areas covering malaria, HIV, tuberculosis, noncommunicable diseases, immunization, water and sanitation, tobacco control and the rest of the global-health canon.

This is the evidence base behind the charts everyone quotes: physician and nursing density, hospital beds per 10,000 population, service coverage indices, cause-specific mortality from the Global Health Estimates, health spending by country. Verified during the August 2026 research pass at 3,093 distinct indicator codes, each resolving to observation rows that carry the geography, the year, the sex and age splits, an attributed origin where the indicator distinguishes sources, and the uncertainty bounds around the estimate. As published it is a browsing experience built for humans; as delivered it is a schema built for models.

Two verified observations from the hospital-bed density indicator, laid out exactly as they arrive:

indicator_code : WHOSIS_000004    indicator_name: Hospital beds per 10,000 population
spatial_type   : COUNTRY          spatial_dim: AZE         parent_location: Europe (EUR)
time_dim       : 2016
dim1           : SEX_MLE          (male)
value          : "143"            numeric_value: 143.4084514

indicator_code : WHOSIS_000004    indicator_name: Hospital beds per 10,000 population
spatial_type   : GLOBAL           spatial_dim: GLOBAL
time_dim       : 2004
dim1           : SEX_BTSX         (both sexes)
value          : "176"            numeric_value: 175.9135015

Read the shape, not just the numbers. One row per indicator x geography x year x disaggregation, with the display-formatted value, the typed numeric estimate and the last-loaded stamp travelling together. The same code yields sibling rows for each sex split, and world aggregates occupy the same geography dimension as countries - so 143 stays attributable to Azerbaijani male-reported bed density in 2016, forever.

Start with a sample of this dataset cut to the indicators and countries you track.

Twenty-four documented fields cover every row, all twenty-four verified during the August 2026 research pass. The locating quartet - IndicatorCode, SpatialDim with its type, TimeDim with its type - pins each row to a concept, a place and a year, and the ParentLocation pair preserves the regional hierarchy so rollups can be rebuilt rather than trusted.

The measurement set carries the published display value, the numeric point estimate and the lower and upper bounds of the uncertainty interval, so a figure never detaches from its error bars. Six disaggregation slots (three dimension-type/value pairs) absorb sex, age, wealth quintile, urban/rural and whatever else a given indicator publishes. Integrity rides along too: the attributed source where an indicator distinguishes origins, free-text comments, the last-loaded stamp, and normalized period keys for clean calendar joins.

Series commonly span 1990 to the present, giving three decades of comparably-defined context on the flagship indicators, and selected measures reach back to the 1970s. Each record carries its own period bounds and a last-loaded timestamp - the hospital-bed family was observed carrying an October 2024 load stamp - so recency is a queryable attribute of every row rather than a claim about the catalog as a whole.

For panel construction the useful property is stability: a country-year row from 1995 joins cleanly to the same country-year row from 2024, because the indicator codes and dimension structure hold still while the values move. Where an indicator's history runs thinner than its neighbors, the missing years show up as absent rows - data-shaped silence you can measure, not gaps you discover after publication.

Health economists build the 194-country panels that papers and preprints run on. Policy analysts at ministries, agencies and NGOs benchmark national performance against regional and income-group peers inside the same schema. Insurers and actuaries lean on decades of mortality, capacity and service-coverage series for pricing and market-entry scenarios, pharma and medtech teams size territories from immunization, disease-program and financing denominators, and journalists resolve the famous league-table charts into rows early enough to check them.

See the persona pages for data scientists and journalists and academics working in health care services.

Because the published form is a browsing experience over 3,093 indicator codes, each with its own dimension configuration - and the analysis lives in the assembly. Building one multi-country, multi-indicator panel by hand means enumerating codes one by one, harmonizing disaggregations that change per indicator, and separating un-disaggregated rows from sex-split siblings before they silently double-count. Datadory does that work once, upstream, and ships the result as one normalized schema.

Field definitions are verified against the live catalog rather than guessed from documentation, indicator-specific breakdowns fold under additional fields on request, and everything arrives delivered daily, weekly, or hourly over API, files, or your warehouse.

Start with a sample cut to the themes, geographies and years you actually need, then let the panel grow from there.