cumulus-library-data-metrics 10.0.0

Data quality and characterization metrics for Cumulus

Data Metrics

A Cumulus-based implementation of the qualifier metrics.

Implemented Metrics

The following qualifier metrics are implemented (per December 2025 qualifier definitions).

• c_attachment_count
• c_content_type_use
• c_pt_count
• c_pt_deceased_count
• c_resource_count
• c_resources_per_pt
• c_system_use
• c_us_core_v6_count *
• q_date_recent
• q_ref_target_pop
• q_ref_target_valid
• q_system_use
• q_valid_us_core_v6 *

* These are US Core profile-based metrics, and the following profiles are not yet implemented:

• Implantable Device (due to the difficulty in identify implantable records)
• Some Observation profiles and also its various Vital Signs sub-profiles like Blood Pressure (just haven't gotten around to them yet)

Installing

pip install cumulus-library-data-metrics

Running the Metrics

These metrics are designed as a Cumulus Library study and are run using the cumulus-library command.

Local Ndjson

Let's say you have a collection of FHIR-formatted NDJSON files. They can all be in one folder or in organized subfolders.

Here's a sample command to run against that pile of NDJSON data:

cumulus-library build \
  --db-type duckdb \
  --database output-tables.db \
  --load-ndjson-dir path/to/ndjson/root \
  --target data_metrics

And then you can load output-tables.db in a DuckDB session and see the results. Or read below to export the counts tables.

Visualization

The metrics can also be reviewed in an interactive web interface by installing and running the open source Cumulus Data Metrics Reporting Tool. When generating a metrics file for this view, the output-mode:aggregate flag should be used. E.g.,

cumulus-library build \
--option output-mode:aggregate \
--option min-bucket-size:0 \
--db-type duckdb \
--database src/data/metrics.duckdb \
--target data_metrics \
--load-ndjson-dir {path/to/ndjson/root}

Athena

Here's a sample command to run against your Cumulus data in Athena:

cumulus-library build \
  --database your-glue-database \
  --workgroup your-athena-workgroup \
  --profile your-aws-credentials-profile \
  --target data_metrics

And then you can see the resulting tables in Athena. Or read below to export the counts tables.

Exporting Counts

For the metrics that have exportable counts (the characterization metrics mostly), you can easily export those using Cumulus Library, by replacing build in the above commands with export ./output-folder. Like so:

cumulus-library export \
  ./output-folder \
  --db-type duckdb \
  --database output-tables.db \
  --target data_metrics

Aggregate counts

This study generates CUBE output by default. If it's easier to work with simple aggregate counts of every value combination (that is, without the partial value combinations that CUBE() generates), run the build step with --option output-mode:aggregate.

That is, run it like:

cumulus-library build --option output-mode:aggregate ...

Bucket sizes

To help preserve privacy, this study ignores any count results of less than ten.

For example, if there are only two male patients that died at age 55, that combination of male & 55 will be dropped from the c_pt_deceased_count table.

This makes it easier to share the count results with other institutions. But if that's not a concern and you want the fine-grained details, you can run the build step with --option min-bucket-size:0 to turn this feature off. Or use another value to change the bucket threshold (the default value is 10).