MA pbp section b1a inpat hosp

What it is

1. 1A CMS Plan Benefit Package (PBP) Section B-1a mirror table giving the member cost-sharing design for the INPATIENT HOSPITAL (acute) benefit of every Medicare Advantage plan: copays/coinsurance/deductible by day-interval and tier, prior-auth/referral rules, and benefit limits.
2. 2It is the authoritative source for 'what does an MA plan charge a member for an inpatient stay' and underpins MA benefit benchmarking / plan-design comparison for the cost-data catalog.

Grain

How to read it

1. 1To get one current-design row per plan for a plan year, pick the latest quarter per plan: e.g.
2. 2`SELECT bid_id, pbp_a_plan_type, pbp_b1a_copay_mcs_amt_int1_t1, pbp_b1a_copay_mcs_bgnd_int1_t1, pbp_b1a_copay_mcs_endd_int1_t1 FROM medicare_silver.pbp_section_b1a_inpat_hosp WHERE report_year=2026 AND quarter=(SELECT max(quarter) FROM medicare_silver.pbp_section_b1a_inpat_hosp WHERE report_year=2026)`.
3. 3Answers: 'what is the per-day inpatient copay (and for which days) for each MA plan in 2026?' To read a plan's day-tiered copay, walk int1->int2->int3 with their bgnd/endd day boundaries (e.g. $375/day for days 1-7, then $0).

Traps

1. 1NO is_current / latest flag exists. The table keeps EVERY quarterly snapshot, so there are multiple rows per plan-year (179,479 rows vs only 92,166 bid x year combos vs 19,819 distinct plans). A query like `WHERE report_year=2023` returns up to 4 rows per plan and will double/quadruple-count. ALWAYS filter to one quarter (e.g. quarter=max per year, or cms_vintage_sequence=max per year).
2. 2The grain is (bid_id, report_year, quarter) -- NOT bid_id alone, NOT bid_id+report_year, and `version` is NOT part of it. The profiler reported a 25-column superset key only because the inpatient `maxplan` benefit columns contain the substring 'plan'; those are benefit attributes, not identifiers.
3. 3Two parallel copay encodings: a plan populates EITHER the flat per-tier copay (pbp_b1a_copay_mcs_amt_t1) when cost does not vary by day, OR the per-day-interval set (pbp_b1a_copay_mcs_amt_int1_t1 + bgnd/endd day bounds) when it does. Reading only one family misses ~half the plans (in 2026Q1: 5,108 use flat _t1, 4,046 use _int1_t1). 'mcs' = Medicare-Covered Stay; that is the inpatient field -- not _ad (additional days), _up (uncovered), _lrd (lifetime reserve), or _nmcs.
4. 4Coinsurance (pbp_b1a_coins_mcs_pct) and deductible (pbp_b1a_ded_amt) are essentially UNUSED for MA inpatient (0 populated in 2026Q1) -- MA inpatient cost-sharing is copay-based. Empty here means 'not how this benefit is structured', not missing data.
5. 5The `_bonly` column family (e.g. pbp_b1a_coins_pct_bonly, copay_amt_ps_bonly) is populated ONLY for the ~2% of rows that are Part-B-only plans (pbp_a_ben_cov='2'); it is always null for standard Part-A plans (ben_cov='1'). Use the non-bonly fields for normal plans.
6. 6asserted_at (2 values), parse_revision, parser_artifact_hash, sha256_content_hash, source_url, cms_vintage_sequence are silver-layer PROVENANCE/load metadata, not business data. asserted_at is the ingest date (May 2026), not a plan date; cms_vintage_sequence resets to 1 each year (it is the quarter-rank within the year, not a global version). effective_period_start/end and observed_period are fully redundant with report_year (Jan 1 / Dec 31 / Jan 1).
7. 7Early years (2010-2017) have ONLY a Q1 snapshot while 2018+ have several quarters, and some year/quarter combos are absent (e.g. 2020 has Q1,Q2,Q4 but no Q3; 2024 has Q1,Q3,Q4 but no Q2). Do not assume four quarters per year. Within-year benefit drift is tiny though (only 2 of 7,741 plans changed inpatient copay between 2023 Q1 and Q4), so any single quarter is a fair representation of that plan-year's design -- but you must still pick exactly one.

Columns (492)

Query it

describe_data_source(source_fqn="medicare_silver.pbp_section_b1a_inpat_hosp")     # the card, live
query_substrate(sql="SELECT * FROM medicare_silver.pbp_section_b1a_inpat_hosp WHERE bid_id = '<value>' AND report_year = '<value>' AND quarter = '<value>' LIMIT 100")