Drug spending

Medicare drug spending from two programs: Part D (pharmacy-dispensed prescriptions, reported per drug and manufacturer) and Part B (physician-administered drugs, reported per HCPCS billing code). Part D figures below use CMS's 'Overall' manufacturer roll-up rows only, so no drug is double-counted across its manufacturers.

Top drugs by total spending ( 2024 )

All single-year figures below use the latest year available in both programs, so Part B and Part D are always compared like-for-like even if one file gains a new year first.

Spending trend by program

Drugs billed in both programs

Joined through dim_drug, which conforms Part B and Part D on exact (upper-cased) brand + generic name matches — treat the overlap as a lower bound, since name formatting differs between the two files.

Caveats

• Part D total shown here ≠ Part D total on /prescribers. This page's Part D figures come from the CMS spending-by-drug file (2023 program-wide roll-up: $275.8B, 'Overall' manufacturer rows in fct_part_d_drug_spending). The /prescribers page reports Part D from the CMS prescriber-profile file ($288.4B total drug cost; $284.8B MAPD+PDP in fct_prescriber_profile). The two CMS sources roll up different universes — drug-level spending vs per-NPI prescriber totals — and neither reconciles to the other; the site convention is to stay within one source per page.
• 'Overall' manufacturer rows are kept for drug-level analysis. Part D's spending file publishes both per-manufacturer rows and an 'Overall' roll-up for each drug. fct_part_d_drug_spending keeps both and flags the roll-up via is_manufacturer_rollup. Every Part D chart on this page filters on that flag; summing total_spending without it double-counts.
• Part B has no manufacturer breakdown. CMS publishes Part B spending per HCPCS billing code only, so a single code can cover several branded drugs — the brand_name column occasionally lists multiple values. Summing Part B total_spending needs no roll-up filter.
• Unmarketed drug-years are dropped, suppressed cells are kept. Both facts drop rows where every metric for a drug-year is null (the drug wasn't marketed that year — 11,038 rows dropped in Part D). Individual cells that CMS suppressed for privacy stay NULL and are never coalesced to 0, so per-beneficiary and per-claim-derived metrics disappear on small cells rather than reading as $0.
• dim_drug join is name-based. The Part B ↔ Part D overlap table joins on drug_key, a hash of upper-cased brand + generic names. Name formatting differs between the two files, so treat the overlap as a lower bound on drugs billed in both programs. Part B rows whose HCPCS code carries no drug name (e.g. not-otherwise-classified codes) have drug_key = NULL and are excluded from the join.