NHSN details HRD data processing: deduplication, suppression and Friday publication cadence

Andrea Cool, an analyst with NHSN’s Emergency Preparedness and Response Unit, presented NHSN’s data processing and quality-assurance steps for Hospital Respiratory Data submissions.

Cool reviewed submission pathways: facilities may submit HRD data via the weekly or daily pathway and may use web form entry, CSV upload or API direct upload. She described front-end submission specifications and business rules applied at save time, including required element formats and valid value ranges; for example, a submission will fail if a reported number of adult inpatient beds exceeds the total number of inpatient beds. She said, "A full list of all the business rules applied to each reporting pathway can be found on the NHSN HRD webpage." (Andrea Cool)

Andrea explained the error-notification behavior by method: web-form rejections appear as a pop-up listing errors; CSV uploads present a pop-up and send an email to the submitting user's NHSN account email; API uploads send an email to the system account email in SAMS. After submission succeeds (that is, passes front-end checks), NHSN applies downstream processing steps including deduplication and suppression. Cool described the deduplication rule: the pipeline retains a single record per facility per reporting week chosen based on the completeness of the record for CMS enforcement purposes; if more than one record is equal on completeness criteria, the most recently submitted record is retained. Suppression removes values that fail specified logic checks so implausible values do not feed aggregated metrics and public dissemination.

Cool outlined the daily-to-weekly reconciliation rules: 24 data elements for bed capacity, occupancy and prevalent hospitalizations must be reported as single-day snapshots (the Wednesday of the reporting week) while 21 new-admission data elements are reported as weekly aggregates (Sunday–Saturday sums). She said complete and valid daily data consist of non-null values for all 45 required data elements and passing all logic checks; reconciled weekly records retain the Wednesday snapshot values for the 24 single-day fields and sum daily values to produce weekly totals for the 21 aggregated fields.

Finally, Cool described downstream ingestion and publication: reconciled and deduplicated weekly HRD records with logic checks applied are ingested into a Palantir Foundry CDC data platform for further transformation and aggregation; after applying exclusion criteria to remove inactive hospitals and those not required to report under CMS conditions of participation, aggregate metrics (for example, hospital admission rates and occupancy percentages) are calculated and published to data.cdc.gov and the NHSN HRD website every Friday afternoon. Cool noted where additional HRD information and links are provided in the slide handouts.