Health Workforce Density By Country Data: Dataset Structure and Field Coverage

Health workforce density expresses how many health workers serve a given population — physicians and nurses converted into per-capita terms so countries of different sizes can be compared. Every serious source treats it as a country-year panel rather than a snapshot. The WHO Global Health Observatory stores rows keyed by country, year and indicator; OECD Health Statistics publishes country-year observations disaggregated by age, sex and socio-economic status on selected measures; and the World Bank's hospital-beds series alone runs 265 entities across roughly 64 years, from 1960 to 2023.

The companion WHO Data Portal (data.who.int) sits on top of the same indicator catalog behind a public OData relay API. It consolidates country profiles for all Member States, Triple Billion dashboards that use a 2018 baseline with 2025 projections, and downloadable CSVs — the under-five mortality extract weighs in at 11.2 MB, which gives a sense of file scale per indicator. For workforce work, the GHO repository remains the programmatic route; the portal is the faster way to eyeball one country's profile.

OECD Health Statistics (OECD.Stat) trades breadth of countries for depth of detail: about 86 health-domain dataflows under the OECD.ELS.HD family cover 38 member countries plus partner economies (51 reference areas observed), some series stretching back to 1960. Individual flows return roughly 2,000 to 34,000 rows as CSV, and the SDMX API requires no registration or key. The World Bank Health Nutrition and Population Indicators hold 658 health-topic series within 1,498 total WDI indicators, covering 217 economies recognized by the Bank plus regional and income aggregates, with CSV, XML, Excel and JSON delivery over a key-free REST API.

A reproducible merge takes five steps:

One caution the walkthrough encodes: vintages drift. OECD health-status flows were observed running 1998 through 2025, while the Bank's hospital-beds series ends at 2023, so a merged table should always record which year each source's latest observation stops.