WHO Global Health Observatory (GHO) indicators: sample notebook
A worked notebook for WHO Global Health Observatory (GHO) indicators, with 5 code cells and the output each one printed. On Scigantic it opens in a hosted JupyterLab with the dataset already mounted, so nothing is downloaded first.
- Provider
- World Health Organization
- License
- WHO data terms: free use, reproduction and distribution for public health purposes with attribution; no use in promotion of a commercial enterprise (https://www.who.int/about/policies/publishing/data-policy/terms-and-conditions)
- Format
- Parquet (in sampled files)
- Stored at
- Amazon S3, s3://scigantic-who (us-east-1)
- Access
- Free and public. No account with the data provider is needed.
The notebook
WHO Global Health Observatory
Every Global Health Observatory indicator, mounted read-only as parquet under gho/. WHO's own API serves one indicator per request and has no bulk download; here data/ .parquet holds each indicator, observations.parquet holds all of them, and indicators.parquet / dimension_values.parquet turn codes into names.
This notebook finds an indicator, plots it, decodes its dimension codes, then ranks the causes of under-five deaths for one country. Everything runs on DuckDB over the mount.
import os
import duckdb
import pandas as pd
import matplotlib.pyplot as plt
ARCHIVE = os.environ.get("SCIGANTIC_MOUNT_PATH", "/mnt/archive")
con = duckdb.connect()
def q(sql, **params):
return con.execute(sql.format(a=ARCHIVE, **params)).df()
print(open(f"{ARCHIVE}/_MIRROR.txt").read().split("\n\n")[0])
q("select count(*) as indicators, sum(row_count) as observations, "
"count(*) filter (where row_count = 0) as empty_indicators from '{a}/gho/indicators.parquet'")
Output:
Scigantic mirror of WHO Global Health Observatory indicators observations empty_indicators 0 3099 9944921.0 612
1. Find an indicator
Search indicators.parquet by name. row_count is how many observations the indicator actually has; a few hundred codes exist in WHO's catalog with no data behind them, and names starting with Archived point at a replacement code.
q("""
select IndicatorCode, IndicatorName, row_count, first_year, last_year, n_spatial_units
from '{a}/gho/indicators.parquet'
where IndicatorName ilike '%under-five mortality%' and row_count > 0
order by row_count desc
""")
Output:
IndicatorCode IndicatorName \ 0 MDG_0000000007 Under-five mortality rate (probability of dyin... row_count first_year last_year n_spatial_units 0 64510 1931 2024 249
2. One indicator, three countries
MDG_0000000001 is the under-five mortality rate (deaths per 1,000 live births). Read it from its own file under gho/data/, keep the both-sexes series (Dim1 = 'SEX_BTSX'), and plot. Value is the published string with its uncertainty interval; NumericValue, Low and High are the numbers.
u5 = q("""
select SpatialDim as iso3, TimeDim as year, NumericValue as rate, Low, High, Value
from '{a}/gho/data/MDG_0000000001.parquet'
where SpatialDimType = 'COUNTRY' and SpatialDim in ('BRA', 'IND', 'NGA')
and Dim1 = 'SEX_BTSX' and TimeDim >= 2000
order by iso3, year
""")
fig, ax = plt.subplots(figsize=(8, 4))
for iso3, g in u5.groupby("iso3"):
ax.plot(g["year"], g["rate"], marker="o", ms=3, label=iso3)
ax.fill_between(g["year"], g["Low"], g["High"], alpha=0.15)
ax.set_ylabel("deaths per 1,000 live births")
ax.set_title("Under-five mortality rate, both sexes (WHO GHO MDG_0000000001)")
ax.legend()
plt.show()
u5.tail(3)
Output:
iso3 year rate Low High Value 72 NGA 2022 70.163303 62.129077 79.329716 70.2 [62.1-79.3] 73 NGA 2023 70.130311 61.113004 80.759944 70.1 [61.1-80.8] 74 NGA 2024 69.771281 59.634972 82.162548 69.8 [59.6-82.2]
3. Decode the codes
SpatialDim, Dim1..Dim3 and friends are codes. dimension_values.parquet maps every (Dimension, Code) to a label and its parent, so one join names the country and the sex.
q("""
select o.SpatialDim as code, c.Title as country, c.ParentTitle as who_region,
o.Dim1 as sex_code, s.Title as sex, o.TimeDim as year, o.Value
from '{a}/gho/data/MDG_0000000001.parquet' o
join '{a}/gho/dimension_values.parquet' c on c.Dimension = 'COUNTRY' and c.Code = o.SpatialDim
join '{a}/gho/dimension_values.parquet' s on s.Dimension = o.Dim1Type and s.Code = o.Dim1
where o.SpatialDim = 'NGA' and o.TimeDim = (select max(TimeDim) from '{a}/gho/data/MDG_0000000001.parquet')
order by sex_code
""")
Output:
code country who_region sex_code sex year Value 0 NGA Nigeria Africa SEX_BTSX Both sexes 2024 69.8 [59.6-82.2] 1 NGA Nigeria Africa SEX_FMLE Female 2024 63.2 [53.7-74.8] 2 NGA Nigeria Africa SEX_MLE Male 2024 76.0 [64.8-89.9]
4. What are those children dying of?
MORT_100 is the number of under-five deaths by cause (WHO and partner estimates), by country and year, with the cause in Dim3 under the CHILDCAUSE dimension and the age band in Dim2. Here every other column is pinned (AGEGROUP_YEARS0-4 is the whole under-five band; the narrower DAYS0-27 and MONTHS1-59 rows would double count it) and any all-causes total row is dropped so the ranking is not led by its own sum.
causes = q("""
select c.Title as cause, o.TimeDim as year, o.NumericValue as deaths
from '{a}/gho/data/MORT_100.parquet' o
join '{a}/gho/dimension_values.parquet' c on c.Dimension = o.Dim3Type and c.Code = o.Dim3
where o.SpatialDim = 'NGA' and o.Dim3Type = 'CHILDCAUSE' and o.Dim2 = 'AGEGROUP_YEARS0-4'
and o.Dim1 = 'SEX_BTSX' and c.Title not ilike '%all causes%'
and o.TimeDim = (select max(TimeDim) from '{a}/gho/data/MORT_100.parquet')
order by deaths desc
""")
print(f"Nigeria, {causes['year'].iloc[0]}: under-five deaths by cause (WHO MORT_100, both sexes)")
causes.drop(columns='year').round(0)
Output:
Nigeria, 2021: under-five deaths by cause (WHO MORT_100, both sexes)
cause deaths
0 Malaria 180294.0
1 Acute lower respiratory infections 169149.0
2 Diarrhoeal diseases 135857.0
3 Birth asphyxia and birth trauma 79818.0
4 Other Group 1 and Other noncommunicable (neona... 76626.0
5 Prematurity 56905.0
6 Sepsis and other infectious conditions of the ... 41386.0
7 Injuries 35231.0
8 Meningitis/encephalitis 28600.0
9 Congenital anomalies 18905.0
10 Tuberculosis 14805.0
11 HIV/AIDS 10640.0
12 Measles 3489.0
...
Where to go next
- gho/observations.parquet is every indicator in one file, ordered by IndicatorCode, so where IndicatorCode = ... prunes to a few row groups.
- The GHE_* burden series (DALYs, YLLs, YLDs by cause) exist in the GHO only as global and GHE-region aggregates for 2000 and 2012; WHO's country-level Global Health Estimates are a separate download, not in the API and not here.
- "All ages" is spelled three ways across indicators (AGEGROUP_AGEAll, AGEGROUP_YEARSALL, AGEGROUP_ALLAges); check dimension_values for the one your indicator uses.
- dimension_values.parquet is keyed by (Dimension, Code); a code string alone is ambiguous across dimensions.
- _MANIFEST.json records when this copy was pulled; WHO revises indicators continuously and the Date column says when a row last changed.
Run it
Open this notebook to run it, see the dataset page for the full description, or browse the full catalog.