MA pbp section b1b inpat hosp psych

What it is

1. 1CMS Medicare Advantage Plan Benefit Package (PBP) Section B-1b: the member cost-sharing design (copays, coinsurance, deductible, prior-auth, day-based intervals, OOP caps) that each MA plan files for INPATIENT HOSPITAL - PSYCHIATRIC services.
2. 2It is the benefit-design / cost-share-rules source for MA inpatient psych, used to compare what plans charge enrollees for a psychiatric hospital stay year over year.

Grain

How to read it

1. 1Get one row per plan-year (de-duplicating the quarterly snapshots) and read the primary Medicare-covered-stay inpatient-psych copay: SELECT contract_id, plan_id, segment_id, pbp_b1b_copay_mcs_amt_int1_t1 AS copay_days1to5, pbp_b1b_copay_mcs_bgnd_int1_t1 AS begin_day, pbp_b1b_copay_mcs_endd_int1_t1 AS end_day, pbp_b1b_auth_yn FROM pbp_section_b1b_inpat_hosp_psych WHERE report_year=2026 AND quarter=1 AND pbp_a_ben_cov='1'.
2. 2Answers: 'for plan H#### in 2026, what does a member pay per day for the first days of a psychiatric inpatient stay, and is prior authorization required?' (median first-interval copay ~$333/day for days 1-5, then $0 days 6-90; auth required for ~87% of plans).

Traps

1. 1MINIMAL GRAIN is 5 columns: contract_id + plan_id + segment_id + report_year + quarter = 200,746 distinct = exactly the row count. The profiler's 14-column key is a SUPERSET — it needlessly pads with bid_id and six pbp_b1b_maxplan_* columns. Do NOT add bid_id (NULL in 7,574 old rows) or version to the key.
2. 2`quarter` is the killer trap. It is the CMS WITHIN-YEAR release snapshot, not a benefit attribute. 2002-2017 have only Q1, but 2018+ carry Q1-Q4 (e.g. 2023 = 4 near-identical copies, ~7,750 rows each). Counting plans across years WITHOUT pinning quarter inflates recent years up to 4x. There are 39 distinct year x quarter snapshots vs only 25 report_years. To get one row per plan-year, filter to one quarter (e.g. MAX(quarter) per year) or GROUP BY year only.
3. 3No `is_current` column and no snapshot fan-out: the table is already deduped to ONE physical row per logical key (max rows/key=1; exactly 1 cms_vintage_sequence / asserted_at / parse_revision / sha256 per key). It is a 'best-vintage' silver snapshot, NOT SCD-2 history — there is no current-flag to filter and no de-dup step needed beyond picking a quarter.
4. 4Cost-share columns are SPARSE and split into day-phase FAMILIES, so naive 'pick the copay column' fails. The populated family for inpatient psych is `mcs` (Medicare-Covered Stay): use pbp_b1b_copay_mcs_amt_int{1,2,3}_t1 (~4,034 plans in 2026) with paired bgnd/endd day boundaries. The `ad` (additional-days) and `lrd` (lifetime-reserve-days) families are tiny (73 / 210 plans). Verified day structure: int1 = days 1-5 (~$333/day copay), int2 = days 6-90 ($0). pct_* columns hold % coinsurance instead of $.
5. 5Y/N flags use CMS encoding 1=Yes, 2=No (NOT 1/0, NOT true/false), and NULL means 'not applicable / benefit not covered'. E.g. ded_yn=2 for ~all plans (no separate psych deductible); auth_yn=1 for the vast majority (prior auth required). Also observed_period == effective_period_start == Jan 1 of report_year for every row (redundant), and `version` (PBP software version) varies plan-to-plan within a single year and is NOT a time/dedup key.
6. 6Sibling table: pbp_section_b1a_inpat_hosp is ACUTE inpatient hospital; this b1b table is the PSYCHIATRIC inpatient sub-section (Medicare's psych benefit has the unique 190-day lifetime limit and its own day-interval cost-sharing). Don't confuse the two; join both for total inpatient cost-share. All `*_bonly` (Part-B-only-enrollee) columns are effectively empty in modern data.

Columns (392)

Query it

describe_data_source(source_fqn="medicare_silver.pbp_section_b1b_inpat_hosp_psych")     # the card, live
query_substrate(sql="SELECT * FROM medicare_silver.pbp_section_b1b_inpat_hosp_psych WHERE contract_id = '<value>' AND plan_id = '<value>' AND segment_id = '<value>' AND report_year = '<value>' AND quarter = '<value>' LIMIT 100")