Start with the release that actually exists
CMS first published its 2026 State Drug Utilization dataset on July 10. The rows currently displayed are for quarter 1. Each row can identify a state, an 11-digit National Drug Code, fee-for-service or managed-care utilization, units reimbursed, prescription count, and reimbursement amounts. That is genuinely current named-product context, but it is a partial-year release. The 2025 file remains the latest dataset containing all four calendar quarters; SDUD records can still be revised.
That first boundary belongs in any chart, memo, or software import. Calling a Q1 extract “2026 utilization” without its quarter invites a reader to compare three months with a full year. A usable evidence record should preserve the dataset year, quarter, retrieval date, utilization type, and whether later revisions were checked.
Read the row before reading a story into it
A public SDUD row says that a state reported utilization for a covered outpatient drug at the NDC-11 level. It does not say that the drug was ordered for, dispensed to, or administered to a nursing-home resident. The file has no resident age or diagnosis, prescriber, pharmacy, facility, place of service, or long-term-care flag. Reimbursement amounts are before manufacturer rebates, so they are not net program spend either.
CMS also suppresses counts below 11 and applies secondary suppression where another cell could expose the protected value. A suppressed value is unknown, not zero. Preserve the dataset's Suppression Used flag instead of silently converting an empty reported field to zero; that would produce a tidy picture and a false result.
Do not manufacture the missing facility-by-drug row
CMS's open Facility-Level MDS Frequency data gets closer to the setting: it reports assessment responses by nursing-home facility and includes medication classes and review-process items. It does not name an ingredient or NDC. SDUD does the reverse: it names the product at state level without identifying nursing-home use.
Those two files do not become a facility ingredient dataset when they are joined. Multiplying a facility's MDS class percentage by a state NDC share mixes different populations, denominators, reporting events, time periods, and suppression rules. It would assign precision that neither source supplies. The honest output is two separately labelled signals and an explicit gap.
Give each question its own evidence lane
A small consultant pharmacy practice can keep the boundary visible without building a research warehouse. Route each question to the narrowest source that can answer it:
- Facility screening: use Facility-Level MDS Frequency or a defined Care Compare measure for class-level and process signals, retaining the measure version, period, denominator limits, and suppression.
- Named-drug context: use SDUD or Medicare Part D public-use files for clearly labelled state, geography, or prescriber context—not as evidence of nursing-home administration or appropriateness.
- Resident review: use the current MAR or eMAR, orders, diagnoses, indication, dose, duration, labs, response, and other resident-specific facts available through the responsible care team.
- Research-grade linkage: for a current Medicare Part D study, one possible route is the restricted monthly resident-level MDS and Part D Event files in the CCW VRDC, with valid stay-and-date attribution, a data-use agreement, and privacy review. It remains Part D fill evidence—not a complete record of facility medication administration or other payers.
Make an analytics tool show its provenance
If software presents a facility drug benchmark, ask for one row's complete route back to source. Which dataset and release supplied the value? Is the observation an assessment, claim, dispense, order, or administration? What population and denominator does it use? How were NDCs normalized, revisions handled, and suppressed values represented? Which field identifies the facility rather than a prescriber address or state?
A polished chart is not an answer to those questions. Ask the supplier to export the source ID, period, grain, transformation, and suppression status alongside the result. If the facility attribution comes from a restricted linkage, request the method, permitted output rules, and validation evidence. If it comes from combining unrelated public percentages, leave the claim out of the recommendation workflow.
Use the new file for what it can prove
The 2026 SDUD release is useful for current state-and-NDC context and for testing a data pipeline against a new quarter. It does not close the gap among the national sources reviewed between a nursing-home facility and a named drug. For an individual recommendation, the proportionate next step remains local: verify the resident record, preserve the source and date, and keep broader public utilization in a separately labelled context field.