VaidhyaMegha/health-determinants-kg

Health Determinants Knowledge Graph

239,802 nodes. 239,795 edges. 211 countries × 35 years of World Bank development indicators — the socioeconomic, demographic, environmental, water and nutrition conditions that shape population health, as one graph.

Built with Samyama Graph. Loader and ETL: samyama-ai/health-determinants-kg.

Licence

Data	CC-BY-4.0 — attribution required, commercial use permitted
Source	World Bank World Development Indicators
Loader code	Apache-2.0, in the GitHub repo. This does not govern the data.

Verified at the source: the World Bank's licensing page states CC-BY-4.0 is the default for datasets the World Bank produces, and that it "allows users to copy, modify and distribute data in any format for any purpose, including commercial use", with attribution and an indication of changes.

Attribution to carry downstream:

Contains data from the World Bank World Development Indicators, licensed CC-BY-4.0. Restructured as a graph; values unmodified.

⚠️ What was removed, and why

The upstream repository builds this graph from more than one source. The WHO-derived portion has been excluded from this dataset, because WHO's data policy permits only "non-commercial, not-for-profit use of the Data for public health purposes" — terms a commercial publisher cannot meet.

45,833 nodes were removed from the 285,635 in the source snapshot:

Removed	indicator_code	Nodes
WaterResource	basic_water, basic_sanitation, safely_managed_water, safely_managed_sanitation	44,013
EnvironmentalFactor	AIR_QUALITY	1,820

Their incident edges went with them. What remains is World Bank data only, and this dataset carries a single licence rather than a mixture.

If you need the WHO indicators, get them from WHO GHO directly under their own terms.

What this is

For each of 211 countries, indicator time series from 1990 to 2024 across five domains, each attached to the country as its own node so that values, years and indicator codes stay first-class rather than collapsing into columns.

The graph shape is deliberately simple — a star around Country — which makes cross-domain questions one hop apart: pair environmental burden against nutrition outcomes, or water access against income level, without a join.

Income level	Countries
High income	81
Upper middle income	54
Lower middle income	49
Low income	25
Not classified	2

Files

Nodes carry an id; edges reference those ids as src and tgt. Join on id.

nodes/

File	Rows	Columns
demographicprofile.csv	107,088	id, demographicprofile_id, indicator_code, indicator_name, value, year
socioeconomicindicator.csv	52,367	id, socioeconomicindicator_id, indicator_code, indicator_name, value, year
environmentalfactor.csv	33,847	id, environmentalfactor_id, indicator_code, indicator_name, value, year
waterresource.csv	32,209	id, waterresource_id, indicator_code, indicator_name, value, year
nutritionindicator.csv	14,073	id, nutritionindicator_id, indicator_code, indicator_name, value, year
country.csv	211	id, income_level, iso_code, name, region_wb
region.csv	7	id, code, name

edges/

File	Rows	Connects
demographic_of.csv	107,088	Country → DemographicProfile
has_indicator.csv	52,367	Country → SocioeconomicIndicator
environment_of.csv	33,847	Country → EnvironmentalFactor
water_resource_of.csv	32,209	Country → WaterResource
nutrition_status.csv	14,073	Country → NutritionIndicator
in_region.csv	211	Country → Region

Also included

health-determinants.sgsnap — the same filtered graph as a Samyama snapshot, importable in one step. It was produced by removing the WHO rows from the source snapshot and rewriting the header, so the snapshot and the CSVs contain exactly the same data.

Usage

from datasets import load_dataset

countries = load_dataset("VaidhyaMegha/health-determinants-kg", "country", revision="v1.0")
print(countries["train"].num_rows)          # 211

docker run --rm -p 8080:8080 -p 6379:6379 public.ecr.aws/f9f6l5u4/samyama-graph:1.1.0
curl -X POST localhost:8080/api/tenants -H 'Content-Type: application/json' \
     -d '{"id":"health-determinants","name":"Health Determinants KG"}'
curl -X POST localhost:8080/api/tenants/health-determinants/snapshot/import \
     -F "file=@health-determinants.sgsnap"

// Highest GNI per capita, 2023
MATCH (c:Country)-[:HAS_INDICATOR]->(s:SocioeconomicIndicator)
WHERE s.indicator_code = 'NY.GNP.PCAP.CD' AND s.year = 2023
RETURN c.name, s.value ORDER BY s.value DESC LIMIT 5
// Bermuda 134,500 · Norway 101,640 · Switzerland 95,740 ·
// Luxembourg 88,480 · Iceland 82,630

// Cross-domain in one hop: low-income countries carrying both an
// environmental and a nutrition burden
MATCH (c:Country {income_level: 'Low income'})-[:ENVIRONMENT_OF]->(e:EnvironmentalFactor),
      (c)-[:NUTRITION_STATUS]->(n:NutritionIndicator)
WHERE e.year = 2020 AND n.year = 2020
RETURN c.name, e.indicator_name, e.value, n.indicator_name, n.value
LIMIT 20

Provenance

1. Original source — World Bank World Development Indicators, CC-BY-4.0, commercial use permitted with attribution.

2. ETL — samyama-ai/health-determinants-kg, which maps the WDI series onto a country-centred property graph.

3. Snapshot — taken 2026-04-03 (sgsnap v2, engine v0.7.0), 285,635 nodes.

4. This dataset — the WHO-derived rows removed (45,833 nodes and their edges), the header rewritten, then flattened to CSV. Totals asserted against the filtered snapshot header — 239,802 nodes and 239,795 edges — and re-verified after upload.

Versioning

Version	Date	Contents
v1.0	2026-08-17	Initial release — World Bank WDI only, WHO rows excluded

load_dataset("VaidhyaMegha/health-determinants-kg", "country", revision="v1.0")

Data changes get a new tag; card corrections land on main.

Limitations

• World Bank only. The upstream repo describes four sources. WHO was removed for the licence reason above, and UNDP HDI is not present in this snapshot at all — the HDI phase found no data and skipped, so there are zero indicator_code = 'HDI' rows. Treat the repo's source list as describing the ETL's capability, not this export.
• Sparse by construction. Not every country has every indicator in every year; the World Bank's own coverage varies widely by indicator and by decade.
• Values are as published, including revisions and estimates. Nothing was recomputed, imputed or gap-filled.
• Country is the only unit. No subnational data, and the 211 Country nodes include aggregates and territories as the World Bank classifies them, not only sovereign states.
• The graph is a star. Every edge is Country → indicator; there are no indicator-to-indicator relationships to traverse.
• region.csv holds 7 World Bank regions; Country.region_wb carries the same information as a property.

Verification

Totals match the filtered snapshot header exactly, and zero WHO rows survive the filter (checked explicitly on both affected labels after conversion). Spot-checked against published World Bank figures: 2023 GNI per capita ranks Bermuda, Norway, Switzerland, Luxembourg and Iceland at the top, and the income-level distribution (81 high / 54 upper-middle / 49 lower-middle / 25 low) matches the World Bank's classification.

Citation

CC-BY-4.0 requires attribution.

Original source:

World Bank. World Development Indicators. https://datatopics.worldbank.org/world-development-indicators/
Licensed CC-BY-4.0. https://creativecommons.org/licenses/by/4.0/

This graph build:

Health Determinants Knowledge Graph, v1.0 (239,802 nodes, 239,795 edges).
Built with Samyama Graph. https://huggingface.co/datasets/VaidhyaMegha/health-determinants-kg
ETL: https://github.com/samyama-ai/health-determinants-kg
Source data: World Bank World Development Indicators, CC-BY-4.0.
Restructured as a graph; WHO-derived rows excluded; values unmodified.

@misc{health_determinants_kg_samyama,
  title        = {Health Determinants Knowledge Graph},
  howpublished = {\url{https://huggingface.co/datasets/VaidhyaMegha/health-determinants-kg}},
  version      = {v1.0},
  note         = {Derived from World Bank World Development Indicators, CC-BY-4.0.
                  WHO-derived rows excluded.
                  ETL: \url{https://github.com/samyama-ai/health-determinants-kg}},
  year         = {2026}
}

The three pieces

Piece	Where
Code — ETL, schema, loaders	github.com/samyama-ai/health-determinants-kg
Data — this dataset, and the raw source it derives from	here, and World Bank WDI
Graph — importable snapshot	health-determinants.sgsnap in this repository