PBP partb rx drugs (cleaned view)

What it is

1. 1A semantic (gold) view of the Medicare Advantage Part B Rx (Part-B-covered drugs, e.g. clinician-administered chemo/biologics/insulin) member cost-share, parsed from CMS PBP (Plan Benefit Package) PUF section B15.
2. 2One row gives the in-network member cost-sharing (coinsurance %, copay $, IRA $35 insulin cap) for one MA/MA-PD plan benefit package in one CMS PBP release.
3. 3It is the benefit-design / cost-share side of Medicare, not provider-paid prices; use it to see what a beneficiary pays out-of-pocket for Part B drugs by plan and year.

Grain

How to read it

1. 1Get the current (2026) in-network Part B drug member cost-share for every MA plan, deduped to one row per plan: SELECT contract_id, plan_id, segment_id, coinsurance_pct, copay_amount, cost_share_string, ira_insulin_copay_month_amount FROM medicare_semantic.vw_pbp_partb_rx_drugs WHERE report_year = 2026 AND quarter = (SELECT MAX(quarter) FROM medicare_semantic.vw_pbp_partb_rx_drugs WHERE report_year = 2026).
2. 2Answers 'what does a beneficiary pay (e.g.
3. 320% coinsurance) for Part-B-administered drugs under each MA plan, and which plans apply the IRA $35 insulin cap?'

Traps

1. 1NO is_current column and NO single 'live' row per plan-year. In republish years (2018-2020, 2022-2024) a plan appears up to 4 times -- once per quarter (quarter 1-4) -- with IDENTICAL benefit values but different source_url/sha256_content_hash. To get one row per plan per benefit-year you MUST dedup, e.g. keep MAX(quarter) per (contract_id,plan_id,segment_id,report_year). Summing/AVG-ing across rows without this 4x-counts those years.
2. 2report_year (not quarter) is the real time axis and is REQUIRED in the grain key. The profiler reported quarter as the time column and offered a candidate key (contract_id,plan_id,segment_id,limit_period_code,benefit_amount_period_code,confidence_tier[,quarter]) that is NOT unique (55,564 of 189,385) and is built from all-NULL/constant placeholder columns. The true minimal key is (contract_id, plan_id, segment_id, report_year, quarter).
3. 3Roughly a dozen columns are pure CONSTANTS in this view (benefit_category, network_type, offered_status, numeric_amount_observability, null_amount_reason, annualization_policy, combined_limit_policy, consumer_document_reconciliation_status, confidence_tier, partb_rx_drugs_sub_benefit, source_pbp_section, source_silver_table, lmi_authorities) and another ~10 are 100% NULL (benefit_amount*, limit_*, referral_required, max_oop_applies, normalization_*). These are shared semantic-view template fields; do not filter or join on them expecting signal.
4. 4Nulls carry temporal meaning. The IRA insulin columns (has_ira_insulin_coinsurance, ira_insulin_*) are 100% NULL before report_year 2024, then populated -- a null means 'pre-Inflation-Reduction-Act / not applicable', not false; the $35 cap only appears 2024+. Chemo coinsurance columns are null before report_year 2010 (chemo cost-share was copay-only earlier). Filtering on these silently drops all older years.
5. 5coinsurance_pct (71.3% filled) is the headline member cost-share, but coinsurance_pct_max is filled MORE (85.4%): ~26,695 rows have the _max set while the headline coinsurance_pct is null, because the _min/_max columns aggregate across the standard + chemo + insulin sub-amounts. Use coinsurance_pct/copay_amount for the plain plan-level cost-share; use the _min/_max for the full range. version is exactly one value per grain key (the CMS file version), so it is redundant with the key, not a grain dimension.

Columns (66)

Query it

describe_data_source(source_fqn="medicare_semantic.vw_pbp_partb_rx_drugs")     # the card, live
query_substrate(sql="SELECT * FROM medicare_semantic.vw_pbp_partb_rx_drugs WHERE contract_id = '<value>' AND plan_id = '<value>' AND segment_id = '<value>' AND report_year = '<value>' AND quarter = '<value>' LIMIT 100")